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: head/sys/netinet/ipfw/ip_fw_dynamic.c 215317 2010-11-14 20:38:11Z dim $");
28
29#define DEB(x)
30#define DDB(x) x
31
32/*
33 * Dynamic rule support for ipfw
34 */
35
36#if !defined(KLD_MODULE)
37#include "opt_ipfw.h"
38#include "opt_ipdivert.h"
39#include "opt_ipdn.h"
40#include "opt_inet.h"
41#ifndef INET
42#error IPFIREWALL requires INET.
43#endif /* INET */
44#endif
45#include "opt_inet6.h"
46#include "opt_ipsec.h"
47
48#include <sys/param.h>
49#include <sys/systm.h>
50#include <sys/malloc.h>
51#include <sys/mbuf.h>
52#include <sys/kernel.h>
53#include <sys/lock.h>
54#include <sys/socket.h>
55#include <sys/sysctl.h>
56#include <sys/syslog.h>
57#include <net/ethernet.h> /* for ETHERTYPE_IP */
58#include <net/if.h>
59#include <net/vnet.h>
60
61#include <netinet/in.h>
62#include <netinet/ip.h>
63#include <netinet/ip_var.h> /* ip_defttl */
64#include <netinet/ip_fw.h>
65#include <netinet/ipfw/ip_fw_private.h>
66#include <netinet/tcp_var.h>
67#include <netinet/udp.h>
68
69#include <netinet/ip6.h> /* IN6_ARE_ADDR_EQUAL */
70#ifdef INET6
71#include <netinet6/in6_var.h>
72#include <netinet6/ip6_var.h>
73#endif
74
75#include <machine/in_cksum.h> /* XXX for in_cksum */
76
77#ifdef MAC
78#include <security/mac/mac_framework.h>
79#endif
80
81/*
82 * Description of dynamic rules.
83 *
84 * Dynamic rules are stored in lists accessed through a hash table
85 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
86 * be modified through the sysctl variable dyn_buckets which is
87 * updated when the table becomes empty.
88 *
89 * XXX currently there is only one list, ipfw_dyn.
90 *
91 * When a packet is received, its address fields are first masked
92 * with the mask defined for the rule, then hashed, then matched
93 * against the entries in the corresponding list.
94 * Dynamic rules can be used for different purposes:
95 * + stateful rules;
96 * + enforcing limits on the number of sessions;
97 * + in-kernel NAT (not implemented yet)
98 *
99 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
100 * measured in seconds and depending on the flags.
101 *
102 * The total number of dynamic rules is stored in dyn_count.
103 * The max number of dynamic rules is dyn_max. When we reach
104 * the maximum number of rules we do not create anymore. This is
105 * done to avoid consuming too much memory, but also too much
106 * time when searching on each packet (ideally, we should try instead
107 * to put a limit on the length of the list on each bucket...).
108 *
109 * Each dynamic rule holds a pointer to the parent ipfw rule so
110 * we know what action to perform. Dynamic rules are removed when
111 * the parent rule is deleted. XXX we should make them survive.
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/*
119 * Static variables followed by global ones
120 */
121STATIC_VNET_DEFINE(ipfw_dyn_rule **, ipfw_dyn_v);
122STATIC_VNET_DEFINE(u_int32_t, dyn_buckets);
123STATIC_VNET_DEFINE(u_int32_t, curr_dyn_buckets);
124STATIC_VNET_DEFINE(struct callout, ipfw_timeout);
125#define V_ipfw_dyn_v VNET(ipfw_dyn_v)
126#define V_dyn_buckets VNET(dyn_buckets)
127#define V_curr_dyn_buckets VNET(curr_dyn_buckets)
128#define V_ipfw_timeout VNET(ipfw_timeout)
129
130static uma_zone_t ipfw_dyn_rule_zone;
131#ifndef __FreeBSD__
132DEFINE_SPINLOCK(ipfw_dyn_mtx);
133#else
134static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */
135#endif
136
137#define IPFW_DYN_LOCK_INIT() \
138 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF)
139#define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx)
140#define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx)
141#define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx)
142#define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED)
143
144void
145ipfw_dyn_unlock(void)
146{
147 IPFW_DYN_UNLOCK();
148}
149
150/*
151 * Timeouts for various events in handing dynamic rules.
152 */
153STATIC_VNET_DEFINE(u_int32_t, dyn_ack_lifetime);
154STATIC_VNET_DEFINE(u_int32_t, dyn_syn_lifetime);
155STATIC_VNET_DEFINE(u_int32_t, dyn_fin_lifetime);
156STATIC_VNET_DEFINE(u_int32_t, dyn_rst_lifetime);
157STATIC_VNET_DEFINE(u_int32_t, dyn_udp_lifetime);
158STATIC_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
175STATIC_VNET_DEFINE(u_int32_t, dyn_keepalive_interval);
176STATIC_VNET_DEFINE(u_int32_t, dyn_keepalive_period);
177STATIC_VNET_DEFINE(u_int32_t, dyn_keepalive);
178
179#define V_dyn_keepalive_interval VNET(dyn_keepalive_interval)
180#define V_dyn_keepalive_period VNET(dyn_keepalive_period)
181#define V_dyn_keepalive VNET(dyn_keepalive)
182
183STATIC_VNET_DEFINE(u_int32_t, dyn_count); /* # of dynamic rules */
184STATIC_VNET_DEFINE(u_int32_t, dyn_max); /* max # of dynamic rules */
185
186#define V_dyn_count VNET(dyn_count)
187#define V_dyn_max VNET(dyn_max)
188
189#ifdef SYSCTL_NODE
190
191SYSBEGIN(f2)
192
193SYSCTL_DECL(_net_inet_ip_fw);
194SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets,
195 CTLFLAG_RW, &VNET_NAME(dyn_buckets), 0,
196 "Number of dyn. buckets");
197SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets,
198 CTLFLAG_RD, &VNET_NAME(curr_dyn_buckets), 0,
199 "Current Number of dyn. buckets");
200SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_count,
201 CTLFLAG_RD, &VNET_NAME(dyn_count), 0,
202 "Number of dyn. rules");
203SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_max,
204 CTLFLAG_RW, &VNET_NAME(dyn_max), 0,
205 "Max number of dyn. rules");
206SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime,
207 CTLFLAG_RW, &VNET_NAME(dyn_ack_lifetime), 0,
208 "Lifetime of dyn. rules for acks");
209SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime,
210 CTLFLAG_RW, &VNET_NAME(dyn_syn_lifetime), 0,
211 "Lifetime of dyn. rules for syn");
212SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime,
213 CTLFLAG_RW, &VNET_NAME(dyn_fin_lifetime), 0,
214 "Lifetime of dyn. rules for fin");
215SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime,
216 CTLFLAG_RW, &VNET_NAME(dyn_rst_lifetime), 0,
217 "Lifetime of dyn. rules for rst");
218SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime,
219 CTLFLAG_RW, &VNET_NAME(dyn_udp_lifetime), 0,
220 "Lifetime of dyn. rules for UDP");
221SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime,
222 CTLFLAG_RW, &VNET_NAME(dyn_short_lifetime), 0,
223 "Lifetime of dyn. rules for other situations");
224SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive,
225 CTLFLAG_RW, &VNET_NAME(dyn_keepalive), 0,
226 "Enable keepalives for dyn. rules");
227
228SYSEND
229
230#endif /* SYSCTL_NODE */
231
232
233static __inline int
234hash_packet6(struct ipfw_flow_id *id)
235{
236 u_int32_t i;
237 i = (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^
238 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^
239 (id->src_ip6.__u6_addr.__u6_addr32[2]) ^
240 (id->src_ip6.__u6_addr.__u6_addr32[3]) ^
241 (id->dst_port) ^ (id->src_port);
242 return i;
243}
244
245/*
246 * IMPORTANT: the hash function for dynamic rules must be commutative
247 * in source and destination (ip,port), because rules are bidirectional
248 * and we want to find both in the same bucket.
249 */
250static __inline int
251hash_packet(struct ipfw_flow_id *id)
252{
253 u_int32_t i;
254
255#ifdef INET6
256 if (IS_IP6_FLOW_ID(id))
257 i = hash_packet6(id);
258 else
259#endif /* INET6 */
260 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
261 i &= (V_curr_dyn_buckets - 1);
262 return i;
263}
264
265static __inline void
266unlink_dyn_rule_print(struct ipfw_flow_id *id)
267{
268 struct in_addr da;
269#ifdef INET6
270 char src[INET6_ADDRSTRLEN], dst[INET6_ADDRSTRLEN];
271#else
272 char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN];
273#endif
274
275#ifdef INET6
276 if (IS_IP6_FLOW_ID(id)) {
277 ip6_sprintf(src, &id->src_ip6);
278 ip6_sprintf(dst, &id->dst_ip6);
279 } else
280#endif
281 {
282 da.s_addr = htonl(id->src_ip);
283 inet_ntoa_r(da, src);
284 da.s_addr = htonl(id->dst_ip);
285 inet_ntoa_r(da, dst);
286 }
287 printf("ipfw: unlink entry %s %d -> %s %d, %d left\n",
288 src, id->src_port, dst, id->dst_port, V_dyn_count - 1);
289}
290
291/**
292 * unlink a dynamic rule from a chain. prev is a pointer to
293 * the previous one, q is a pointer to the rule to delete,
294 * head is a pointer to the head of the queue.
295 * Modifies q and potentially also head.
296 */
297#define UNLINK_DYN_RULE(prev, head, q) { \
298 ipfw_dyn_rule *old_q = q; \
299 \
300 /* remove a refcount to the parent */ \
301 if (q->dyn_type == O_LIMIT) \
302 q->parent->count--; \
303 DEB(unlink_dyn_rule_print(&q->id);) \
304 if (prev != NULL) \
305 prev->next = q = q->next; \
306 else \
307 head = q = q->next; \
308 V_dyn_count--; \
309 uma_zfree(ipfw_dyn_rule_zone, old_q); }
310
311#define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
312
313/**
314 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
315 *
316 * If keep_me == NULL, rules are deleted even if not expired,
317 * otherwise only expired rules are removed.
318 *
319 * The value of the second parameter is also used to point to identify
320 * a rule we absolutely do not want to remove (e.g. because we are
321 * holding a reference to it -- this is the case with O_LIMIT_PARENT
322 * rules). The pointer is only used for comparison, so any non-null
323 * value will do.
324 */
325static void
326remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
327{
328 static u_int32_t last_remove = 0;
329
330#define FORCE (keep_me == NULL)
331
332 ipfw_dyn_rule *prev, *q;
333 int i, pass = 0, max_pass = 0;
334
335 IPFW_DYN_LOCK_ASSERT();
336
337 if (V_ipfw_dyn_v == NULL || V_dyn_count == 0)
338 return;
339 /* do not expire more than once per second, it is useless */
340 if (!FORCE && last_remove == time_uptime)
341 return;
342 last_remove = time_uptime;
343
344 /*
345 * because O_LIMIT refer to parent rules, during the first pass only
346 * remove child and mark any pending LIMIT_PARENT, and remove
347 * them in a second pass.
348 */
349next_pass:
350 for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
351 for (prev=NULL, q = V_ipfw_dyn_v[i] ; q ; ) {
352 /*
353 * Logic can become complex here, so we split tests.
354 */
355 if (q == keep_me)
356 goto next;
357 if (rule != NULL && rule != q->rule)
358 goto next; /* not the one we are looking for */
359 if (q->dyn_type == O_LIMIT_PARENT) {
360 /*
361 * handle parent in the second pass,
362 * record we need one.
363 */
364 max_pass = 1;
365 if (pass == 0)
366 goto next;
367 if (FORCE && q->count != 0 ) {
368 /* XXX should not happen! */
369 printf("ipfw: OUCH! cannot remove rule,"
370 " count %d\n", q->count);
371 }
372 } else {
373 if (!FORCE &&
374 !TIME_LEQ( q->expire, time_uptime ))
375 goto next;
376 }
377 if (q->dyn_type != O_LIMIT_PARENT || !q->count) {
378 UNLINK_DYN_RULE(prev, V_ipfw_dyn_v[i], q);
379 continue;
380 }
381next:
382 prev=q;
383 q=q->next;
384 }
385 }
386 if (pass++ < max_pass)
387 goto next_pass;
388}
389
390void
391ipfw_remove_dyn_children(struct ip_fw *rule)
392{
393 IPFW_DYN_LOCK();
394 remove_dyn_rule(rule, NULL /* force removal */);
395 IPFW_DYN_UNLOCK();
396}
397
398/**
399 * lookup a dynamic rule, locked version
400 */
401static ipfw_dyn_rule *
402lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction,
403 struct tcphdr *tcp)
404{
405 /*
406 * stateful ipfw extensions.
407 * Lookup into dynamic session queue
408 */
409#define MATCH_REVERSE 0
410#define MATCH_FORWARD 1
411#define MATCH_NONE 2
412#define MATCH_UNKNOWN 3
413 int i, dir = MATCH_NONE;
414 ipfw_dyn_rule *prev, *q=NULL;
415
416 IPFW_DYN_LOCK_ASSERT();
417
418 if (V_ipfw_dyn_v == NULL)
419 goto done; /* not found */
420 i = hash_packet( pkt );
421 for (prev=NULL, q = V_ipfw_dyn_v[i] ; q != NULL ; ) {
422 if (q->dyn_type == O_LIMIT_PARENT && q->count)
423 goto next;
424 if (TIME_LEQ( q->expire, time_uptime)) { /* expire entry */
425 UNLINK_DYN_RULE(prev, V_ipfw_dyn_v[i], q);
426 continue;
427 }
428 if (pkt->proto == q->id.proto &&
429 q->dyn_type != O_LIMIT_PARENT) {
430 if (IS_IP6_FLOW_ID(pkt)) {
431 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
432 &(q->id.src_ip6)) &&
433 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
434 &(q->id.dst_ip6)) &&
435 pkt->src_port == q->id.src_port &&
436 pkt->dst_port == q->id.dst_port ) {
437 dir = MATCH_FORWARD;
438 break;
439 }
440 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
441 &(q->id.dst_ip6)) &&
442 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
443 &(q->id.src_ip6)) &&
444 pkt->src_port == q->id.dst_port &&
445 pkt->dst_port == q->id.src_port ) {
446 dir = MATCH_REVERSE;
447 break;
448 }
449 } else {
450 if (pkt->src_ip == q->id.src_ip &&
451 pkt->dst_ip == q->id.dst_ip &&
452 pkt->src_port == q->id.src_port &&
453 pkt->dst_port == q->id.dst_port ) {
454 dir = MATCH_FORWARD;
455 break;
456 }
457 if (pkt->src_ip == q->id.dst_ip &&
458 pkt->dst_ip == q->id.src_ip &&
459 pkt->src_port == q->id.dst_port &&
460 pkt->dst_port == q->id.src_port ) {
461 dir = MATCH_REVERSE;
462 break;
463 }
464 }
465 }
466next:
467 prev = q;
468 q = q->next;
469 }
470 if (q == NULL)
471 goto done; /* q = NULL, not found */
472
473 if ( prev != NULL) { /* found and not in front */
474 prev->next = q->next;
475 q->next = V_ipfw_dyn_v[i];
476 V_ipfw_dyn_v[i] = q;
477 }
478 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
479 u_char flags = pkt->_flags & (TH_FIN|TH_SYN|TH_RST);
480
481#define BOTH_SYN (TH_SYN | (TH_SYN << 8))
482#define BOTH_FIN (TH_FIN | (TH_FIN << 8))
483 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
484 switch (q->state) {
485 case TH_SYN: /* opening */
486 q->expire = time_uptime + V_dyn_syn_lifetime;
487 break;
488
489 case BOTH_SYN: /* move to established */
490 case BOTH_SYN | TH_FIN : /* one side tries to close */
491 case BOTH_SYN | (TH_FIN << 8) :
492 if (tcp) {
493#define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
494 u_int32_t ack = ntohl(tcp->th_ack);
495 if (dir == MATCH_FORWARD) {
496 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
497 q->ack_fwd = ack;
498 else { /* ignore out-of-sequence */
499 break;
500 }
501 } else {
502 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
503 q->ack_rev = ack;
504 else { /* ignore out-of-sequence */
505 break;
506 }
507 }
508 }
509 q->expire = time_uptime + V_dyn_ack_lifetime;
510 break;
511
512 case BOTH_SYN | BOTH_FIN: /* both sides closed */
513 if (V_dyn_fin_lifetime >= V_dyn_keepalive_period)
514 V_dyn_fin_lifetime = V_dyn_keepalive_period - 1;
515 q->expire = time_uptime + V_dyn_fin_lifetime;
516 break;
517
518 default:
519#if 0
520 /*
521 * reset or some invalid combination, but can also
522 * occur if we use keep-state the wrong way.
523 */
524 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
525 printf("invalid state: 0x%x\n", q->state);
526#endif
527 if (V_dyn_rst_lifetime >= V_dyn_keepalive_period)
528 V_dyn_rst_lifetime = V_dyn_keepalive_period - 1;
529 q->expire = time_uptime + V_dyn_rst_lifetime;
530 break;
531 }
532 } else if (pkt->proto == IPPROTO_UDP) {
533 q->expire = time_uptime + V_dyn_udp_lifetime;
534 } else {
535 /* other protocols */
536 q->expire = time_uptime + V_dyn_short_lifetime;
537 }
538done:
539 if (match_direction)
540 *match_direction = dir;
541 return q;
542}
543
544ipfw_dyn_rule *
545ipfw_lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
546 struct tcphdr *tcp)
547{
548 ipfw_dyn_rule *q;
549
550 IPFW_DYN_LOCK();
551 q = lookup_dyn_rule_locked(pkt, match_direction, tcp);
552 if (q == NULL)
553 IPFW_DYN_UNLOCK();
554 /* NB: return table locked when q is not NULL */
555 return q;
556}
557
558static void
559realloc_dynamic_table(void)
560{
561 IPFW_DYN_LOCK_ASSERT();
562
563 /*
564 * Try reallocation, make sure we have a power of 2 and do
565 * not allow more than 64k entries. In case of overflow,
566 * default to 1024.
567 */
568
569 if (V_dyn_buckets > 65536)
570 V_dyn_buckets = 1024;
571 if ((V_dyn_buckets & (V_dyn_buckets-1)) != 0) { /* not a power of 2 */
572 V_dyn_buckets = V_curr_dyn_buckets; /* reset */
573 return;
574 }
575 V_curr_dyn_buckets = V_dyn_buckets;
576 if (V_ipfw_dyn_v != NULL)
577 free(V_ipfw_dyn_v, M_IPFW);
578 for (;;) {
579 V_ipfw_dyn_v = malloc(V_curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
580 M_IPFW, M_NOWAIT | M_ZERO);
581 if (V_ipfw_dyn_v != NULL || V_curr_dyn_buckets <= 2)
582 break;
583 V_curr_dyn_buckets /= 2;
584 }
585}
586
587/**
588 * Install state of type 'type' for a dynamic session.
589 * The hash table contains two type of rules:
590 * - regular rules (O_KEEP_STATE)
591 * - rules for sessions with limited number of sess per user
592 * (O_LIMIT). When they are created, the parent is
593 * increased by 1, and decreased on delete. In this case,
594 * the third parameter is the parent rule and not the chain.
595 * - "parent" rules for the above (O_LIMIT_PARENT).
596 */
597static ipfw_dyn_rule *
598add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
599{
600 ipfw_dyn_rule *r;
601 int i;
602
603 IPFW_DYN_LOCK_ASSERT();
604
605 if (V_ipfw_dyn_v == NULL ||
606 (V_dyn_count == 0 && V_dyn_buckets != V_curr_dyn_buckets)) {
607 realloc_dynamic_table();
608 if (V_ipfw_dyn_v == NULL)
609 return NULL; /* failed ! */
610 }
611 i = hash_packet(id);
612
613 r = uma_zalloc(ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO);
614 if (r == NULL) {
615 printf ("ipfw: sorry cannot allocate state\n");
616 return NULL;
617 }
618
619 /* increase refcount on parent, and set pointer */
620 if (dyn_type == O_LIMIT) {
621 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
622 if ( parent->dyn_type != O_LIMIT_PARENT)
623 panic("invalid parent");
624 parent->count++;
625 r->parent = parent;
626 rule = parent->rule;
627 }
628
629 r->id = *id;
630 r->expire = time_uptime + V_dyn_syn_lifetime;
631 r->rule = rule;
632 r->dyn_type = dyn_type;
633 r->pcnt = r->bcnt = 0;
634 r->count = 0;
635
636 r->bucket = i;
637 r->next = V_ipfw_dyn_v[i];
638 V_ipfw_dyn_v[i] = r;
639 V_dyn_count++;
640 DEB({
641 struct in_addr da;
642#ifdef INET6
643 char src[INET6_ADDRSTRLEN];
644 char dst[INET6_ADDRSTRLEN];
645#else
646 char src[INET_ADDRSTRLEN];
647 char dst[INET_ADDRSTRLEN];
648#endif
649
650#ifdef INET6
651 if (IS_IP6_FLOW_ID(&(r->id))) {
652 ip6_sprintf(src, &r->id.src_ip6);
653 ip6_sprintf(dst, &r->id.dst_ip6);
654 } else
655#endif
656 {
657 da.s_addr = htonl(r->id.src_ip);
658 inet_ntoa_r(da, src);
659 da.s_addr = htonl(r->id.dst_ip);
660 inet_ntoa_r(da, dst);
661 }
662 printf("ipfw: add dyn entry ty %d %s %d -> %s %d, total %d\n",
663 dyn_type, src, r->id.src_port, dst, r->id.dst_port,
664 V_dyn_count);
665 })
666 return r;
667}
668
669/**
670 * lookup dynamic parent rule using pkt and rule as search keys.
671 * If the lookup fails, then install one.
672 */
673static ipfw_dyn_rule *
674lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
675{
676 ipfw_dyn_rule *q;
677 int i;
678
679 IPFW_DYN_LOCK_ASSERT();
680
681 if (V_ipfw_dyn_v) {
682 int is_v6 = IS_IP6_FLOW_ID(pkt);
683 i = hash_packet( pkt );
684 for (q = V_ipfw_dyn_v[i] ; q != NULL ; q=q->next)
685 if (q->dyn_type == O_LIMIT_PARENT &&
686 rule== q->rule &&
687 pkt->proto == q->id.proto &&
688 pkt->src_port == q->id.src_port &&
689 pkt->dst_port == q->id.dst_port &&
690 (
691 (is_v6 &&
692 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
693 &(q->id.src_ip6)) &&
694 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
695 &(q->id.dst_ip6))) ||
696 (!is_v6 &&
697 pkt->src_ip == q->id.src_ip &&
698 pkt->dst_ip == q->id.dst_ip)
699 )
700 ) {
701 q->expire = time_uptime + V_dyn_short_lifetime;
702 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);)
703 return q;
704 }
705 }
706 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
707}
708
709/**
710 * Install dynamic state for rule type cmd->o.opcode
711 *
712 * Returns 1 (failure) if state is not installed because of errors or because
713 * session limitations are enforced.
714 */
715int
716ipfw_install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
717 struct ip_fw_args *args, uint32_t tablearg)
718{
719 static int last_log;
720 ipfw_dyn_rule *q;
721 struct in_addr da;
722#ifdef INET6
723 char src[INET6_ADDRSTRLEN + 2], dst[INET6_ADDRSTRLEN + 2];
724#else
725 char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN];
726#endif
727
728 src[0] = '\0';
729 dst[0] = '\0';
730
731 IPFW_DYN_LOCK();
732
733 DEB(
734#ifdef INET6
735 if (IS_IP6_FLOW_ID(&(args->f_id))) {
736 ip6_sprintf(src, &args->f_id.src_ip6);
737 ip6_sprintf(dst, &args->f_id.dst_ip6);
738 } else
739#endif
740 {
741 da.s_addr = htonl(args->f_id.src_ip);
742 inet_ntoa_r(da, src);
743 da.s_addr = htonl(args->f_id.dst_ip);
744 inet_ntoa_r(da, dst);
745 }
746 printf("ipfw: %s: type %d %s %u -> %s %u\n",
747 __func__, cmd->o.opcode, src, args->f_id.src_port,
748 dst, args->f_id.dst_port);
749 src[0] = '\0';
750 dst[0] = '\0';
751 )
752
753 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
754
755 if (q != NULL) { /* should never occur */
756 if (last_log != time_uptime) {
757 last_log = time_uptime;
758 printf("ipfw: %s: entry already present, done\n",
759 __func__);
760 }
761 IPFW_DYN_UNLOCK();
762 return (0);
763 }
764
765 if (V_dyn_count >= V_dyn_max)
766 /* Run out of slots, try to remove any expired rule. */
767 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
768
769 if (V_dyn_count >= V_dyn_max) {
770 if (last_log != time_uptime) {
771 last_log = time_uptime;
772 printf("ipfw: %s: Too many dynamic rules\n", __func__);
773 }
774 IPFW_DYN_UNLOCK();
775 return (1); /* cannot install, notify caller */
776 }
777
778 switch (cmd->o.opcode) {
779 case O_KEEP_STATE: /* bidir rule */
780 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
781 break;
782
783 case O_LIMIT: { /* limit number of sessions */
784 struct ipfw_flow_id id;
785 ipfw_dyn_rule *parent;
786 uint32_t conn_limit;
787 uint16_t limit_mask = cmd->limit_mask;
788
789 conn_limit = (cmd->conn_limit == IP_FW_TABLEARG) ?
790 tablearg : cmd->conn_limit;
791
792 DEB(
793 if (cmd->conn_limit == IP_FW_TABLEARG)
794 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u "
795 "(tablearg)\n", __func__, conn_limit);
796 else
797 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u\n",
798 __func__, conn_limit);
799 )
800
801 id.dst_ip = id.src_ip = id.dst_port = id.src_port = 0;
802 id.proto = args->f_id.proto;
803 id.addr_type = args->f_id.addr_type;
804 id.fib = M_GETFIB(args->m);
805
806 if (IS_IP6_FLOW_ID (&(args->f_id))) {
807 if (limit_mask & DYN_SRC_ADDR)
808 id.src_ip6 = args->f_id.src_ip6;
809 if (limit_mask & DYN_DST_ADDR)
810 id.dst_ip6 = args->f_id.dst_ip6;
811 } else {
812 if (limit_mask & DYN_SRC_ADDR)
813 id.src_ip = args->f_id.src_ip;
814 if (limit_mask & DYN_DST_ADDR)
815 id.dst_ip = args->f_id.dst_ip;
816 }
817 if (limit_mask & DYN_SRC_PORT)
818 id.src_port = args->f_id.src_port;
819 if (limit_mask & DYN_DST_PORT)
820 id.dst_port = args->f_id.dst_port;
821 if ((parent = lookup_dyn_parent(&id, rule)) == NULL) {
822 printf("ipfw: %s: add parent failed\n", __func__);
823 IPFW_DYN_UNLOCK();
824 return (1);
825 }
826
827 if (parent->count >= conn_limit) {
828 /* See if we can remove some expired rule. */
829 remove_dyn_rule(rule, parent);
830 if (parent->count >= conn_limit) {
831 if (V_fw_verbose && last_log != time_uptime) {
832 last_log = time_uptime;
833#ifdef INET6
834 /*
835 * XXX IPv6 flows are not
836 * supported yet.
837 */
838 if (IS_IP6_FLOW_ID(&(args->f_id))) {
839 char ip6buf[INET6_ADDRSTRLEN];
840 snprintf(src, sizeof(src),
841 "[%s]", ip6_sprintf(ip6buf,
842 &args->f_id.src_ip6));
843 snprintf(dst, sizeof(dst),
844 "[%s]", ip6_sprintf(ip6buf,
845 &args->f_id.dst_ip6));
846 } else
847#endif
848 {
849 da.s_addr =
850 htonl(args->f_id.src_ip);
851 inet_ntoa_r(da, src);
852 da.s_addr =
853 htonl(args->f_id.dst_ip);
854 inet_ntoa_r(da, dst);
855 }
856 log(LOG_SECURITY | LOG_DEBUG,
857 "ipfw: %d %s %s:%u -> %s:%u, %s\n",
858 parent->rule->rulenum,
859 "drop session",
860 src, (args->f_id.src_port),
861 dst, (args->f_id.dst_port),
862 "too many entries");
863 }
864 IPFW_DYN_UNLOCK();
865 return (1);
866 }
867 }
868 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
869 break;
870 }
871 default:
872 printf("ipfw: %s: unknown dynamic rule type %u\n",
873 __func__, cmd->o.opcode);
874 IPFW_DYN_UNLOCK();
875 return (1);
876 }
877
878 /* XXX just set lifetime */
879 lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
880
881 IPFW_DYN_UNLOCK();
882 return (0);
883}
884
885/*
886 * Generate a TCP packet, containing either a RST or a keepalive.
887 * When flags & TH_RST, we are sending a RST packet, because of a
888 * "reset" action matched the packet.
889 * Otherwise we are sending a keepalive, and flags & TH_
890 * The 'replyto' mbuf is the mbuf being replied to, if any, and is required
891 * so that MAC can label the reply appropriately.
892 */
893struct mbuf *
894ipfw_send_pkt(struct mbuf *replyto, struct ipfw_flow_id *id, u_int32_t seq,
895 u_int32_t ack, int flags)
896{
897 struct mbuf *m = NULL; /* stupid compiler */
898 int len, dir;
899 struct ip *h = NULL; /* stupid compiler */
900#ifdef INET6
901 struct ip6_hdr *h6 = NULL;
902#endif
903 struct tcphdr *th = NULL;
904
905 MGETHDR(m, M_DONTWAIT, MT_DATA);
906 if (m == NULL)
907 return (NULL);
908
909 M_SETFIB(m, id->fib);
910#ifdef MAC
911 if (replyto != NULL)
912 mac_netinet_firewall_reply(replyto, m);
913 else
914 mac_netinet_firewall_send(m);
915#else
916 (void)replyto; /* don't warn about unused arg */
917#endif
918
919 switch (id->addr_type) {
920 case 4:
921 len = sizeof(struct ip) + sizeof(struct tcphdr);
922 break;
923#ifdef INET6
924 case 6:
925 len = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
926 break;
927#endif
928 default:
929 /* XXX: log me?!? */
930 FREE_PKT(m);
931 return (NULL);
932 }
933 dir = ((flags & (TH_SYN | TH_RST)) == TH_SYN);
934
935 m->m_data += max_linkhdr;
936 m->m_flags |= M_SKIP_FIREWALL;
937 m->m_pkthdr.len = m->m_len = len;
938 m->m_pkthdr.rcvif = NULL;
939 bzero(m->m_data, len);
940
941 switch (id->addr_type) {
942 case 4:
943 h = mtod(m, struct ip *);
944
945 /* prepare for checksum */
946 h->ip_p = IPPROTO_TCP;
947 h->ip_len = htons(sizeof(struct tcphdr));
948 if (dir) {
949 h->ip_src.s_addr = htonl(id->src_ip);
950 h->ip_dst.s_addr = htonl(id->dst_ip);
951 } else {
952 h->ip_src.s_addr = htonl(id->dst_ip);
953 h->ip_dst.s_addr = htonl(id->src_ip);
954 }
955
956 th = (struct tcphdr *)(h + 1);
957 break;
958#ifdef INET6
959 case 6:
960 h6 = mtod(m, struct ip6_hdr *);
961
962 /* prepare for checksum */
963 h6->ip6_nxt = IPPROTO_TCP;
964 h6->ip6_plen = htons(sizeof(struct tcphdr));
965 if (dir) {
966 h6->ip6_src = id->src_ip6;
967 h6->ip6_dst = id->dst_ip6;
968 } else {
969 h6->ip6_src = id->dst_ip6;
970 h6->ip6_dst = id->src_ip6;
971 }
972
973 th = (struct tcphdr *)(h6 + 1);
974 break;
975#endif
976 }
977
978 if (dir) {
979 th->th_sport = htons(id->src_port);
980 th->th_dport = htons(id->dst_port);
981 } else {
982 th->th_sport = htons(id->dst_port);
983 th->th_dport = htons(id->src_port);
984 }
985 th->th_off = sizeof(struct tcphdr) >> 2;
986
987 if (flags & TH_RST) {
988 if (flags & TH_ACK) {
989 th->th_seq = htonl(ack);
990 th->th_flags = TH_RST;
991 } else {
992 if (flags & TH_SYN)
993 seq++;
994 th->th_ack = htonl(seq);
995 th->th_flags = TH_RST | TH_ACK;
996 }
997 } else {
998 /*
999 * Keepalive - use caller provided sequence numbers
1000 */
1001 th->th_seq = htonl(seq);
1002 th->th_ack = htonl(ack);
1003 th->th_flags = TH_ACK;
1004 }
1005
1006 switch (id->addr_type) {
1007 case 4:
1008 th->th_sum = in_cksum(m, len);
1009
1010 /* finish the ip header */
1011 h->ip_v = 4;
1012 h->ip_hl = sizeof(*h) >> 2;
1013 h->ip_tos = IPTOS_LOWDELAY;
1014 h->ip_off = 0;
1015 /* ip_len must be in host format for ip_output */
1016 h->ip_len = len;
1017 h->ip_ttl = V_ip_defttl;
1018 h->ip_sum = 0;
1019 break;
1020#ifdef INET6
1021 case 6:
1022 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(*h6),
1023 sizeof(struct tcphdr));
1024
1025 /* finish the ip6 header */
1026 h6->ip6_vfc |= IPV6_VERSION;
1027 h6->ip6_hlim = IPV6_DEFHLIM;
1028 break;
1029#endif
1030 }
1031
1032 return (m);
1033}
1034
1035/*
1036 * This procedure is only used to handle keepalives. It is invoked
1037 * every dyn_keepalive_period
1038 */
1039static void
1040ipfw_tick(void * vnetx)
1041{
1042 struct mbuf *m0, *m, *mnext, **mtailp;
1043#ifdef INET6
1044 struct mbuf *m6, **m6_tailp;
1045#endif
1046 int i;
1047 ipfw_dyn_rule *q;
1048#ifdef VIMAGE
1049 struct vnet *vp = vnetx;
1050#endif
1051
1052 CURVNET_SET(vp);
1053 if (V_dyn_keepalive == 0 || V_ipfw_dyn_v == NULL || V_dyn_count == 0)
1054 goto done;
1055
1056 /*
1057 * We make a chain of packets to go out here -- not deferring
1058 * until after we drop the IPFW dynamic rule lock would result
1059 * in a lock order reversal with the normal packet input -> ipfw
1060 * call stack.
1061 */
1062 m0 = NULL;
1063 mtailp = &m0;
1064#ifdef INET6
1065 m6 = NULL;
1066 m6_tailp = &m6;
1067#endif
1068 IPFW_DYN_LOCK();
1069 for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
1070 for (q = V_ipfw_dyn_v[i] ; q ; q = q->next ) {
1071 if (q->dyn_type == O_LIMIT_PARENT)
1072 continue;
1073 if (q->id.proto != IPPROTO_TCP)
1074 continue;
1075 if ( (q->state & BOTH_SYN) != BOTH_SYN)
1076 continue;
1077 if (TIME_LEQ(time_uptime + V_dyn_keepalive_interval,
1078 q->expire))
1079 continue; /* too early */
1080 if (TIME_LEQ(q->expire, time_uptime))
1081 continue; /* too late, rule expired */
1082
1083 m = ipfw_send_pkt(NULL, &(q->id), q->ack_rev - 1,
1084 q->ack_fwd, TH_SYN);
1085 mnext = ipfw_send_pkt(NULL, &(q->id), q->ack_fwd - 1,
1086 q->ack_rev, 0);
1087
1088 switch (q->id.addr_type) {
1089 case 4:
1090 if (m != NULL) {
1091 *mtailp = m;
1092 mtailp = &(*mtailp)->m_nextpkt;
1093 }
1094 if (mnext != NULL) {
1095 *mtailp = mnext;
1096 mtailp = &(*mtailp)->m_nextpkt;
1097 }
1098 break;
1099#ifdef INET6
1100 case 6:
1101 if (m != NULL) {
1102 *m6_tailp = m;
1103 m6_tailp = &(*m6_tailp)->m_nextpkt;
1104 }
1105 if (mnext != NULL) {
1106 *m6_tailp = mnext;
1107 m6_tailp = &(*m6_tailp)->m_nextpkt;
1108 }
1109 break;
1110#endif
1111 }
1112
1113 m = mnext = NULL;
1114 }
1115 }
1116 IPFW_DYN_UNLOCK();
1117 for (m = mnext = m0; m != NULL; m = mnext) {
1118 mnext = m->m_nextpkt;
1119 m->m_nextpkt = NULL;
1120 ip_output(m, NULL, NULL, 0, NULL, NULL);
1121 }
1122#ifdef INET6
1123 for (m = mnext = m6; m != NULL; m = mnext) {
1124 mnext = m->m_nextpkt;
1125 m->m_nextpkt = NULL;
1126 ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1127 }
1128#endif
1129done:
1130 callout_reset_on(&V_ipfw_timeout, V_dyn_keepalive_period * hz,
1131 ipfw_tick, vnetx, 0);
1132 CURVNET_RESTORE();
1133}
1134
1135void
1136ipfw_dyn_attach(void)
1137{
1138 ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule",
1139 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL,
1140 UMA_ALIGN_PTR, 0);
1141
1142 IPFW_DYN_LOCK_INIT();
1143}
1144
1145void
1146ipfw_dyn_detach(void)
1147{
1148 uma_zdestroy(ipfw_dyn_rule_zone);
1149 IPFW_DYN_LOCK_DESTROY();
1150}
1151
1152void
1153ipfw_dyn_init(void)
1154{
1155 V_ipfw_dyn_v = NULL;
1156 V_dyn_buckets = 256; /* must be power of 2 */
1157 V_curr_dyn_buckets = 256; /* must be power of 2 */
1158
1159 V_dyn_ack_lifetime = 300;
1160 V_dyn_syn_lifetime = 20;
1161 V_dyn_fin_lifetime = 1;
1162 V_dyn_rst_lifetime = 1;
1163 V_dyn_udp_lifetime = 10;
1164 V_dyn_short_lifetime = 5;
1165
1166 V_dyn_keepalive_interval = 20;
1167 V_dyn_keepalive_period = 5;
1168 V_dyn_keepalive = 1; /* do send keepalives */
1169
1170 V_dyn_max = 4096; /* max # of dynamic rules */
1171 callout_init(&V_ipfw_timeout, CALLOUT_MPSAFE);
1172 callout_reset_on(&V_ipfw_timeout, hz, ipfw_tick, curvnet, 0);
1173}
1174
1175void
1176ipfw_dyn_uninit(int pass)
1177{
1178 if (pass == 0)
1179 callout_drain(&V_ipfw_timeout);
1180 else {
1181 if (V_ipfw_dyn_v != NULL)
1182 free(V_ipfw_dyn_v, M_IPFW);
1183 }
1184}
1185
1186int
1187ipfw_dyn_len(void)
1188{
1189 return (V_ipfw_dyn_v == NULL) ? 0 :
1190 (V_dyn_count * sizeof(ipfw_dyn_rule));
1191}
1192
1193void
1194ipfw_get_dynamic(char **pbp, const char *ep)
1195{
1196 ipfw_dyn_rule *p, *last = NULL;
1197 char *bp;
1198 int i;
1199
1200 if (V_ipfw_dyn_v == NULL)
1201 return;
1202 bp = *pbp;
1203
1204 IPFW_DYN_LOCK();
1205 for (i = 0 ; i < V_curr_dyn_buckets; i++)
1206 for (p = V_ipfw_dyn_v[i] ; p != NULL; p = p->next) {
1207 if (bp + sizeof *p <= ep) {
1208 ipfw_dyn_rule *dst =
1209 (ipfw_dyn_rule *)bp;
1210 bcopy(p, dst, sizeof *p);
1211 bcopy(&(p->rule->rulenum), &(dst->rule),
1212 sizeof(p->rule->rulenum));
1213 /*
1214 * store set number into high word of
1215 * dst->rule pointer.
1216 */
1217 bcopy(&(p->rule->set),
1218 (char *)&dst->rule +
1219 sizeof(p->rule->rulenum),
1220 sizeof(p->rule->set));
1221 /*
1222 * store a non-null value in "next".
1223 * The userland code will interpret a
1224 * NULL here as a marker
1225 * for the last dynamic rule.
1226 */
1227 bcopy(&dst, &dst->next, sizeof(dst));
1228 last = dst;
1229 dst->expire =
1230 TIME_LEQ(dst->expire, time_uptime) ?
1231 0 : dst->expire - time_uptime ;
1232 bp += sizeof(ipfw_dyn_rule);
1233 }
1234 }
1235 IPFW_DYN_UNLOCK();
1236 if (last != NULL) /* mark last dynamic rule */
1237 bzero(&last->next, sizeof(last));
1238 *pbp = bp;
1239}
1240/* end of file */
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: head/sys/netinet/ipfw/ip_fw_dynamic.c 215317 2010-11-14 20:38:11Z dim $");
28
29#define DEB(x)
30#define DDB(x) x
31
32/*
33 * Dynamic rule support for ipfw
34 */
35
36#if !defined(KLD_MODULE)
37#include "opt_ipfw.h"
38#include "opt_ipdivert.h"
39#include "opt_ipdn.h"
40#include "opt_inet.h"
41#ifndef INET
42#error IPFIREWALL requires INET.
43#endif /* INET */
44#endif
45#include "opt_inet6.h"
46#include "opt_ipsec.h"
47
48#include <sys/param.h>
49#include <sys/systm.h>
50#include <sys/malloc.h>
51#include <sys/mbuf.h>
52#include <sys/kernel.h>
53#include <sys/lock.h>
54#include <sys/socket.h>
55#include <sys/sysctl.h>
56#include <sys/syslog.h>
57#include <net/ethernet.h> /* for ETHERTYPE_IP */
58#include <net/if.h>
59#include <net/vnet.h>
60
61#include <netinet/in.h>
62#include <netinet/ip.h>
63#include <netinet/ip_var.h> /* ip_defttl */
64#include <netinet/ip_fw.h>
65#include <netinet/ipfw/ip_fw_private.h>
66#include <netinet/tcp_var.h>
67#include <netinet/udp.h>
68
69#include <netinet/ip6.h> /* IN6_ARE_ADDR_EQUAL */
70#ifdef INET6
71#include <netinet6/in6_var.h>
72#include <netinet6/ip6_var.h>
73#endif
74
75#include <machine/in_cksum.h> /* XXX for in_cksum */
76
77#ifdef MAC
78#include <security/mac/mac_framework.h>
79#endif
80
81/*
82 * Description of dynamic rules.
83 *
84 * Dynamic rules are stored in lists accessed through a hash table
85 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
86 * be modified through the sysctl variable dyn_buckets which is
87 * updated when the table becomes empty.
88 *
89 * XXX currently there is only one list, ipfw_dyn.
90 *
91 * When a packet is received, its address fields are first masked
92 * with the mask defined for the rule, then hashed, then matched
93 * against the entries in the corresponding list.
94 * Dynamic rules can be used for different purposes:
95 * + stateful rules;
96 * + enforcing limits on the number of sessions;
97 * + in-kernel NAT (not implemented yet)
98 *
99 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
100 * measured in seconds and depending on the flags.
101 *
102 * The total number of dynamic rules is stored in dyn_count.
103 * The max number of dynamic rules is dyn_max. When we reach
104 * the maximum number of rules we do not create anymore. This is
105 * done to avoid consuming too much memory, but also too much
106 * time when searching on each packet (ideally, we should try instead
107 * to put a limit on the length of the list on each bucket...).
108 *
109 * Each dynamic rule holds a pointer to the parent ipfw rule so
110 * we know what action to perform. Dynamic rules are removed when
111 * the parent rule is deleted. XXX we should make them survive.
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/*
119 * Static variables followed by global ones
120 */
121STATIC_VNET_DEFINE(ipfw_dyn_rule **, ipfw_dyn_v);
122STATIC_VNET_DEFINE(u_int32_t, dyn_buckets);
123STATIC_VNET_DEFINE(u_int32_t, curr_dyn_buckets);
124STATIC_VNET_DEFINE(struct callout, ipfw_timeout);
125#define V_ipfw_dyn_v VNET(ipfw_dyn_v)
126#define V_dyn_buckets VNET(dyn_buckets)
127#define V_curr_dyn_buckets VNET(curr_dyn_buckets)
128#define V_ipfw_timeout VNET(ipfw_timeout)
129
130static uma_zone_t ipfw_dyn_rule_zone;
131#ifndef __FreeBSD__
132DEFINE_SPINLOCK(ipfw_dyn_mtx);
133#else
134static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */
135#endif
136
137#define IPFW_DYN_LOCK_INIT() \
138 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF)
139#define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx)
140#define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx)
141#define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx)
142#define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED)
143
144void
145ipfw_dyn_unlock(void)
146{
147 IPFW_DYN_UNLOCK();
148}
149
150/*
151 * Timeouts for various events in handing dynamic rules.
152 */
153STATIC_VNET_DEFINE(u_int32_t, dyn_ack_lifetime);
154STATIC_VNET_DEFINE(u_int32_t, dyn_syn_lifetime);
155STATIC_VNET_DEFINE(u_int32_t, dyn_fin_lifetime);
156STATIC_VNET_DEFINE(u_int32_t, dyn_rst_lifetime);
157STATIC_VNET_DEFINE(u_int32_t, dyn_udp_lifetime);
158STATIC_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
175STATIC_VNET_DEFINE(u_int32_t, dyn_keepalive_interval);
176STATIC_VNET_DEFINE(u_int32_t, dyn_keepalive_period);
177STATIC_VNET_DEFINE(u_int32_t, dyn_keepalive);
178
179#define V_dyn_keepalive_interval VNET(dyn_keepalive_interval)
180#define V_dyn_keepalive_period VNET(dyn_keepalive_period)
181#define V_dyn_keepalive VNET(dyn_keepalive)
182
183STATIC_VNET_DEFINE(u_int32_t, dyn_count); /* # of dynamic rules */
184STATIC_VNET_DEFINE(u_int32_t, dyn_max); /* max # of dynamic rules */
185
186#define V_dyn_count VNET(dyn_count)
187#define V_dyn_max VNET(dyn_max)
188
189#ifdef SYSCTL_NODE
190
191SYSBEGIN(f2)
192
193SYSCTL_DECL(_net_inet_ip_fw);
194SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets,
195 CTLFLAG_RW, &VNET_NAME(dyn_buckets), 0,
196 "Number of dyn. buckets");
197SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets,
198 CTLFLAG_RD, &VNET_NAME(curr_dyn_buckets), 0,
199 "Current Number of dyn. buckets");
200SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_count,
201 CTLFLAG_RD, &VNET_NAME(dyn_count), 0,
202 "Number of dyn. rules");
203SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_max,
204 CTLFLAG_RW, &VNET_NAME(dyn_max), 0,
205 "Max number of dyn. rules");
206SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime,
207 CTLFLAG_RW, &VNET_NAME(dyn_ack_lifetime), 0,
208 "Lifetime of dyn. rules for acks");
209SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime,
210 CTLFLAG_RW, &VNET_NAME(dyn_syn_lifetime), 0,
211 "Lifetime of dyn. rules for syn");
212SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime,
213 CTLFLAG_RW, &VNET_NAME(dyn_fin_lifetime), 0,
214 "Lifetime of dyn. rules for fin");
215SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime,
216 CTLFLAG_RW, &VNET_NAME(dyn_rst_lifetime), 0,
217 "Lifetime of dyn. rules for rst");
218SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime,
219 CTLFLAG_RW, &VNET_NAME(dyn_udp_lifetime), 0,
220 "Lifetime of dyn. rules for UDP");
221SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime,
222 CTLFLAG_RW, &VNET_NAME(dyn_short_lifetime), 0,
223 "Lifetime of dyn. rules for other situations");
224SYSCTL_VNET_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive,
225 CTLFLAG_RW, &VNET_NAME(dyn_keepalive), 0,
226 "Enable keepalives for dyn. rules");
227
228SYSEND
229
230#endif /* SYSCTL_NODE */
231
232
233static __inline int
234hash_packet6(struct ipfw_flow_id *id)
235{
236 u_int32_t i;
237 i = (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^
238 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^
239 (id->src_ip6.__u6_addr.__u6_addr32[2]) ^
240 (id->src_ip6.__u6_addr.__u6_addr32[3]) ^
241 (id->dst_port) ^ (id->src_port);
242 return i;
243}
244
245/*
246 * IMPORTANT: the hash function for dynamic rules must be commutative
247 * in source and destination (ip,port), because rules are bidirectional
248 * and we want to find both in the same bucket.
249 */
250static __inline int
251hash_packet(struct ipfw_flow_id *id)
252{
253 u_int32_t i;
254
255#ifdef INET6
256 if (IS_IP6_FLOW_ID(id))
257 i = hash_packet6(id);
258 else
259#endif /* INET6 */
260 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
261 i &= (V_curr_dyn_buckets - 1);
262 return i;
263}
264
265static __inline void
266unlink_dyn_rule_print(struct ipfw_flow_id *id)
267{
268 struct in_addr da;
269#ifdef INET6
270 char src[INET6_ADDRSTRLEN], dst[INET6_ADDRSTRLEN];
271#else
272 char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN];
273#endif
274
275#ifdef INET6
276 if (IS_IP6_FLOW_ID(id)) {
277 ip6_sprintf(src, &id->src_ip6);
278 ip6_sprintf(dst, &id->dst_ip6);
279 } else
280#endif
281 {
282 da.s_addr = htonl(id->src_ip);
283 inet_ntoa_r(da, src);
284 da.s_addr = htonl(id->dst_ip);
285 inet_ntoa_r(da, dst);
286 }
287 printf("ipfw: unlink entry %s %d -> %s %d, %d left\n",
288 src, id->src_port, dst, id->dst_port, V_dyn_count - 1);
289}
290
291/**
292 * unlink a dynamic rule from a chain. prev is a pointer to
293 * the previous one, q is a pointer to the rule to delete,
294 * head is a pointer to the head of the queue.
295 * Modifies q and potentially also head.
296 */
297#define UNLINK_DYN_RULE(prev, head, q) { \
298 ipfw_dyn_rule *old_q = q; \
299 \
300 /* remove a refcount to the parent */ \
301 if (q->dyn_type == O_LIMIT) \
302 q->parent->count--; \
303 DEB(unlink_dyn_rule_print(&q->id);) \
304 if (prev != NULL) \
305 prev->next = q = q->next; \
306 else \
307 head = q = q->next; \
308 V_dyn_count--; \
309 uma_zfree(ipfw_dyn_rule_zone, old_q); }
310
311#define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
312
313/**
314 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
315 *
316 * If keep_me == NULL, rules are deleted even if not expired,
317 * otherwise only expired rules are removed.
318 *
319 * The value of the second parameter is also used to point to identify
320 * a rule we absolutely do not want to remove (e.g. because we are
321 * holding a reference to it -- this is the case with O_LIMIT_PARENT
322 * rules). The pointer is only used for comparison, so any non-null
323 * value will do.
324 */
325static void
326remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
327{
328 static u_int32_t last_remove = 0;
329
330#define FORCE (keep_me == NULL)
331
332 ipfw_dyn_rule *prev, *q;
333 int i, pass = 0, max_pass = 0;
334
335 IPFW_DYN_LOCK_ASSERT();
336
337 if (V_ipfw_dyn_v == NULL || V_dyn_count == 0)
338 return;
339 /* do not expire more than once per second, it is useless */
340 if (!FORCE && last_remove == time_uptime)
341 return;
342 last_remove = time_uptime;
343
344 /*
345 * because O_LIMIT refer to parent rules, during the first pass only
346 * remove child and mark any pending LIMIT_PARENT, and remove
347 * them in a second pass.
348 */
349next_pass:
350 for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
351 for (prev=NULL, q = V_ipfw_dyn_v[i] ; q ; ) {
352 /*
353 * Logic can become complex here, so we split tests.
354 */
355 if (q == keep_me)
356 goto next;
357 if (rule != NULL && rule != q->rule)
358 goto next; /* not the one we are looking for */
359 if (q->dyn_type == O_LIMIT_PARENT) {
360 /*
361 * handle parent in the second pass,
362 * record we need one.
363 */
364 max_pass = 1;
365 if (pass == 0)
366 goto next;
367 if (FORCE && q->count != 0 ) {
368 /* XXX should not happen! */
369 printf("ipfw: OUCH! cannot remove rule,"
370 " count %d\n", q->count);
371 }
372 } else {
373 if (!FORCE &&
374 !TIME_LEQ( q->expire, time_uptime ))
375 goto next;
376 }
377 if (q->dyn_type != O_LIMIT_PARENT || !q->count) {
378 UNLINK_DYN_RULE(prev, V_ipfw_dyn_v[i], q);
379 continue;
380 }
381next:
382 prev=q;
383 q=q->next;
384 }
385 }
386 if (pass++ < max_pass)
387 goto next_pass;
388}
389
390void
391ipfw_remove_dyn_children(struct ip_fw *rule)
392{
393 IPFW_DYN_LOCK();
394 remove_dyn_rule(rule, NULL /* force removal */);
395 IPFW_DYN_UNLOCK();
396}
397
398/**
399 * lookup a dynamic rule, locked version
400 */
401static ipfw_dyn_rule *
402lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction,
403 struct tcphdr *tcp)
404{
405 /*
406 * stateful ipfw extensions.
407 * Lookup into dynamic session queue
408 */
409#define MATCH_REVERSE 0
410#define MATCH_FORWARD 1
411#define MATCH_NONE 2
412#define MATCH_UNKNOWN 3
413 int i, dir = MATCH_NONE;
414 ipfw_dyn_rule *prev, *q=NULL;
415
416 IPFW_DYN_LOCK_ASSERT();
417
418 if (V_ipfw_dyn_v == NULL)
419 goto done; /* not found */
420 i = hash_packet( pkt );
421 for (prev=NULL, q = V_ipfw_dyn_v[i] ; q != NULL ; ) {
422 if (q->dyn_type == O_LIMIT_PARENT && q->count)
423 goto next;
424 if (TIME_LEQ( q->expire, time_uptime)) { /* expire entry */
425 UNLINK_DYN_RULE(prev, V_ipfw_dyn_v[i], q);
426 continue;
427 }
428 if (pkt->proto == q->id.proto &&
429 q->dyn_type != O_LIMIT_PARENT) {
430 if (IS_IP6_FLOW_ID(pkt)) {
431 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
432 &(q->id.src_ip6)) &&
433 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
434 &(q->id.dst_ip6)) &&
435 pkt->src_port == q->id.src_port &&
436 pkt->dst_port == q->id.dst_port ) {
437 dir = MATCH_FORWARD;
438 break;
439 }
440 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
441 &(q->id.dst_ip6)) &&
442 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
443 &(q->id.src_ip6)) &&
444 pkt->src_port == q->id.dst_port &&
445 pkt->dst_port == q->id.src_port ) {
446 dir = MATCH_REVERSE;
447 break;
448 }
449 } else {
450 if (pkt->src_ip == q->id.src_ip &&
451 pkt->dst_ip == q->id.dst_ip &&
452 pkt->src_port == q->id.src_port &&
453 pkt->dst_port == q->id.dst_port ) {
454 dir = MATCH_FORWARD;
455 break;
456 }
457 if (pkt->src_ip == q->id.dst_ip &&
458 pkt->dst_ip == q->id.src_ip &&
459 pkt->src_port == q->id.dst_port &&
460 pkt->dst_port == q->id.src_port ) {
461 dir = MATCH_REVERSE;
462 break;
463 }
464 }
465 }
466next:
467 prev = q;
468 q = q->next;
469 }
470 if (q == NULL)
471 goto done; /* q = NULL, not found */
472
473 if ( prev != NULL) { /* found and not in front */
474 prev->next = q->next;
475 q->next = V_ipfw_dyn_v[i];
476 V_ipfw_dyn_v[i] = q;
477 }
478 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
479 u_char flags = pkt->_flags & (TH_FIN|TH_SYN|TH_RST);
480
481#define BOTH_SYN (TH_SYN | (TH_SYN << 8))
482#define BOTH_FIN (TH_FIN | (TH_FIN << 8))
483 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
484 switch (q->state) {
485 case TH_SYN: /* opening */
486 q->expire = time_uptime + V_dyn_syn_lifetime;
487 break;
488
489 case BOTH_SYN: /* move to established */
490 case BOTH_SYN | TH_FIN : /* one side tries to close */
491 case BOTH_SYN | (TH_FIN << 8) :
492 if (tcp) {
493#define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
494 u_int32_t ack = ntohl(tcp->th_ack);
495 if (dir == MATCH_FORWARD) {
496 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
497 q->ack_fwd = ack;
498 else { /* ignore out-of-sequence */
499 break;
500 }
501 } else {
502 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
503 q->ack_rev = ack;
504 else { /* ignore out-of-sequence */
505 break;
506 }
507 }
508 }
509 q->expire = time_uptime + V_dyn_ack_lifetime;
510 break;
511
512 case BOTH_SYN | BOTH_FIN: /* both sides closed */
513 if (V_dyn_fin_lifetime >= V_dyn_keepalive_period)
514 V_dyn_fin_lifetime = V_dyn_keepalive_period - 1;
515 q->expire = time_uptime + V_dyn_fin_lifetime;
516 break;
517
518 default:
519#if 0
520 /*
521 * reset or some invalid combination, but can also
522 * occur if we use keep-state the wrong way.
523 */
524 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
525 printf("invalid state: 0x%x\n", q->state);
526#endif
527 if (V_dyn_rst_lifetime >= V_dyn_keepalive_period)
528 V_dyn_rst_lifetime = V_dyn_keepalive_period - 1;
529 q->expire = time_uptime + V_dyn_rst_lifetime;
530 break;
531 }
532 } else if (pkt->proto == IPPROTO_UDP) {
533 q->expire = time_uptime + V_dyn_udp_lifetime;
534 } else {
535 /* other protocols */
536 q->expire = time_uptime + V_dyn_short_lifetime;
537 }
538done:
539 if (match_direction)
540 *match_direction = dir;
541 return q;
542}
543
544ipfw_dyn_rule *
545ipfw_lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
546 struct tcphdr *tcp)
547{
548 ipfw_dyn_rule *q;
549
550 IPFW_DYN_LOCK();
551 q = lookup_dyn_rule_locked(pkt, match_direction, tcp);
552 if (q == NULL)
553 IPFW_DYN_UNLOCK();
554 /* NB: return table locked when q is not NULL */
555 return q;
556}
557
558static void
559realloc_dynamic_table(void)
560{
561 IPFW_DYN_LOCK_ASSERT();
562
563 /*
564 * Try reallocation, make sure we have a power of 2 and do
565 * not allow more than 64k entries. In case of overflow,
566 * default to 1024.
567 */
568
569 if (V_dyn_buckets > 65536)
570 V_dyn_buckets = 1024;
571 if ((V_dyn_buckets & (V_dyn_buckets-1)) != 0) { /* not a power of 2 */
572 V_dyn_buckets = V_curr_dyn_buckets; /* reset */
573 return;
574 }
575 V_curr_dyn_buckets = V_dyn_buckets;
576 if (V_ipfw_dyn_v != NULL)
577 free(V_ipfw_dyn_v, M_IPFW);
578 for (;;) {
579 V_ipfw_dyn_v = malloc(V_curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
580 M_IPFW, M_NOWAIT | M_ZERO);
581 if (V_ipfw_dyn_v != NULL || V_curr_dyn_buckets <= 2)
582 break;
583 V_curr_dyn_buckets /= 2;
584 }
585}
586
587/**
588 * Install state of type 'type' for a dynamic session.
589 * The hash table contains two type of rules:
590 * - regular rules (O_KEEP_STATE)
591 * - rules for sessions with limited number of sess per user
592 * (O_LIMIT). When they are created, the parent is
593 * increased by 1, and decreased on delete. In this case,
594 * the third parameter is the parent rule and not the chain.
595 * - "parent" rules for the above (O_LIMIT_PARENT).
596 */
597static ipfw_dyn_rule *
598add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
599{
600 ipfw_dyn_rule *r;
601 int i;
602
603 IPFW_DYN_LOCK_ASSERT();
604
605 if (V_ipfw_dyn_v == NULL ||
606 (V_dyn_count == 0 && V_dyn_buckets != V_curr_dyn_buckets)) {
607 realloc_dynamic_table();
608 if (V_ipfw_dyn_v == NULL)
609 return NULL; /* failed ! */
610 }
611 i = hash_packet(id);
612
613 r = uma_zalloc(ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO);
614 if (r == NULL) {
615 printf ("ipfw: sorry cannot allocate state\n");
616 return NULL;
617 }
618
619 /* increase refcount on parent, and set pointer */
620 if (dyn_type == O_LIMIT) {
621 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
622 if ( parent->dyn_type != O_LIMIT_PARENT)
623 panic("invalid parent");
624 parent->count++;
625 r->parent = parent;
626 rule = parent->rule;
627 }
628
629 r->id = *id;
630 r->expire = time_uptime + V_dyn_syn_lifetime;
631 r->rule = rule;
632 r->dyn_type = dyn_type;
633 r->pcnt = r->bcnt = 0;
634 r->count = 0;
635
636 r->bucket = i;
637 r->next = V_ipfw_dyn_v[i];
638 V_ipfw_dyn_v[i] = r;
639 V_dyn_count++;
640 DEB({
641 struct in_addr da;
642#ifdef INET6
643 char src[INET6_ADDRSTRLEN];
644 char dst[INET6_ADDRSTRLEN];
645#else
646 char src[INET_ADDRSTRLEN];
647 char dst[INET_ADDRSTRLEN];
648#endif
649
650#ifdef INET6
651 if (IS_IP6_FLOW_ID(&(r->id))) {
652 ip6_sprintf(src, &r->id.src_ip6);
653 ip6_sprintf(dst, &r->id.dst_ip6);
654 } else
655#endif
656 {
657 da.s_addr = htonl(r->id.src_ip);
658 inet_ntoa_r(da, src);
659 da.s_addr = htonl(r->id.dst_ip);
660 inet_ntoa_r(da, dst);
661 }
662 printf("ipfw: add dyn entry ty %d %s %d -> %s %d, total %d\n",
663 dyn_type, src, r->id.src_port, dst, r->id.dst_port,
664 V_dyn_count);
665 })
666 return r;
667}
668
669/**
670 * lookup dynamic parent rule using pkt and rule as search keys.
671 * If the lookup fails, then install one.
672 */
673static ipfw_dyn_rule *
674lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
675{
676 ipfw_dyn_rule *q;
677 int i;
678
679 IPFW_DYN_LOCK_ASSERT();
680
681 if (V_ipfw_dyn_v) {
682 int is_v6 = IS_IP6_FLOW_ID(pkt);
683 i = hash_packet( pkt );
684 for (q = V_ipfw_dyn_v[i] ; q != NULL ; q=q->next)
685 if (q->dyn_type == O_LIMIT_PARENT &&
686 rule== q->rule &&
687 pkt->proto == q->id.proto &&
688 pkt->src_port == q->id.src_port &&
689 pkt->dst_port == q->id.dst_port &&
690 (
691 (is_v6 &&
692 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
693 &(q->id.src_ip6)) &&
694 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
695 &(q->id.dst_ip6))) ||
696 (!is_v6 &&
697 pkt->src_ip == q->id.src_ip &&
698 pkt->dst_ip == q->id.dst_ip)
699 )
700 ) {
701 q->expire = time_uptime + V_dyn_short_lifetime;
702 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);)
703 return q;
704 }
705 }
706 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
707}
708
709/**
710 * Install dynamic state for rule type cmd->o.opcode
711 *
712 * Returns 1 (failure) if state is not installed because of errors or because
713 * session limitations are enforced.
714 */
715int
716ipfw_install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
717 struct ip_fw_args *args, uint32_t tablearg)
718{
719 static int last_log;
720 ipfw_dyn_rule *q;
721 struct in_addr da;
722#ifdef INET6
723 char src[INET6_ADDRSTRLEN + 2], dst[INET6_ADDRSTRLEN + 2];
724#else
725 char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN];
726#endif
727
728 src[0] = '\0';
729 dst[0] = '\0';
730
731 IPFW_DYN_LOCK();
732
733 DEB(
734#ifdef INET6
735 if (IS_IP6_FLOW_ID(&(args->f_id))) {
736 ip6_sprintf(src, &args->f_id.src_ip6);
737 ip6_sprintf(dst, &args->f_id.dst_ip6);
738 } else
739#endif
740 {
741 da.s_addr = htonl(args->f_id.src_ip);
742 inet_ntoa_r(da, src);
743 da.s_addr = htonl(args->f_id.dst_ip);
744 inet_ntoa_r(da, dst);
745 }
746 printf("ipfw: %s: type %d %s %u -> %s %u\n",
747 __func__, cmd->o.opcode, src, args->f_id.src_port,
748 dst, args->f_id.dst_port);
749 src[0] = '\0';
750 dst[0] = '\0';
751 )
752
753 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
754
755 if (q != NULL) { /* should never occur */
756 if (last_log != time_uptime) {
757 last_log = time_uptime;
758 printf("ipfw: %s: entry already present, done\n",
759 __func__);
760 }
761 IPFW_DYN_UNLOCK();
762 return (0);
763 }
764
765 if (V_dyn_count >= V_dyn_max)
766 /* Run out of slots, try to remove any expired rule. */
767 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
768
769 if (V_dyn_count >= V_dyn_max) {
770 if (last_log != time_uptime) {
771 last_log = time_uptime;
772 printf("ipfw: %s: Too many dynamic rules\n", __func__);
773 }
774 IPFW_DYN_UNLOCK();
775 return (1); /* cannot install, notify caller */
776 }
777
778 switch (cmd->o.opcode) {
779 case O_KEEP_STATE: /* bidir rule */
780 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
781 break;
782
783 case O_LIMIT: { /* limit number of sessions */
784 struct ipfw_flow_id id;
785 ipfw_dyn_rule *parent;
786 uint32_t conn_limit;
787 uint16_t limit_mask = cmd->limit_mask;
788
789 conn_limit = (cmd->conn_limit == IP_FW_TABLEARG) ?
790 tablearg : cmd->conn_limit;
791
792 DEB(
793 if (cmd->conn_limit == IP_FW_TABLEARG)
794 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u "
795 "(tablearg)\n", __func__, conn_limit);
796 else
797 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u\n",
798 __func__, conn_limit);
799 )
800
801 id.dst_ip = id.src_ip = id.dst_port = id.src_port = 0;
802 id.proto = args->f_id.proto;
803 id.addr_type = args->f_id.addr_type;
804 id.fib = M_GETFIB(args->m);
805
806 if (IS_IP6_FLOW_ID (&(args->f_id))) {
807 if (limit_mask & DYN_SRC_ADDR)
808 id.src_ip6 = args->f_id.src_ip6;
809 if (limit_mask & DYN_DST_ADDR)
810 id.dst_ip6 = args->f_id.dst_ip6;
811 } else {
812 if (limit_mask & DYN_SRC_ADDR)
813 id.src_ip = args->f_id.src_ip;
814 if (limit_mask & DYN_DST_ADDR)
815 id.dst_ip = args->f_id.dst_ip;
816 }
817 if (limit_mask & DYN_SRC_PORT)
818 id.src_port = args->f_id.src_port;
819 if (limit_mask & DYN_DST_PORT)
820 id.dst_port = args->f_id.dst_port;
821 if ((parent = lookup_dyn_parent(&id, rule)) == NULL) {
822 printf("ipfw: %s: add parent failed\n", __func__);
823 IPFW_DYN_UNLOCK();
824 return (1);
825 }
826
827 if (parent->count >= conn_limit) {
828 /* See if we can remove some expired rule. */
829 remove_dyn_rule(rule, parent);
830 if (parent->count >= conn_limit) {
831 if (V_fw_verbose && last_log != time_uptime) {
832 last_log = time_uptime;
833#ifdef INET6
834 /*
835 * XXX IPv6 flows are not
836 * supported yet.
837 */
838 if (IS_IP6_FLOW_ID(&(args->f_id))) {
839 char ip6buf[INET6_ADDRSTRLEN];
840 snprintf(src, sizeof(src),
841 "[%s]", ip6_sprintf(ip6buf,
842 &args->f_id.src_ip6));
843 snprintf(dst, sizeof(dst),
844 "[%s]", ip6_sprintf(ip6buf,
845 &args->f_id.dst_ip6));
846 } else
847#endif
848 {
849 da.s_addr =
850 htonl(args->f_id.src_ip);
851 inet_ntoa_r(da, src);
852 da.s_addr =
853 htonl(args->f_id.dst_ip);
854 inet_ntoa_r(da, dst);
855 }
856 log(LOG_SECURITY | LOG_DEBUG,
857 "ipfw: %d %s %s:%u -> %s:%u, %s\n",
858 parent->rule->rulenum,
859 "drop session",
860 src, (args->f_id.src_port),
861 dst, (args->f_id.dst_port),
862 "too many entries");
863 }
864 IPFW_DYN_UNLOCK();
865 return (1);
866 }
867 }
868 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
869 break;
870 }
871 default:
872 printf("ipfw: %s: unknown dynamic rule type %u\n",
873 __func__, cmd->o.opcode);
874 IPFW_DYN_UNLOCK();
875 return (1);
876 }
877
878 /* XXX just set lifetime */
879 lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
880
881 IPFW_DYN_UNLOCK();
882 return (0);
883}
884
885/*
886 * Generate a TCP packet, containing either a RST or a keepalive.
887 * When flags & TH_RST, we are sending a RST packet, because of a
888 * "reset" action matched the packet.
889 * Otherwise we are sending a keepalive, and flags & TH_
890 * The 'replyto' mbuf is the mbuf being replied to, if any, and is required
891 * so that MAC can label the reply appropriately.
892 */
893struct mbuf *
894ipfw_send_pkt(struct mbuf *replyto, struct ipfw_flow_id *id, u_int32_t seq,
895 u_int32_t ack, int flags)
896{
897 struct mbuf *m = NULL; /* stupid compiler */
898 int len, dir;
899 struct ip *h = NULL; /* stupid compiler */
900#ifdef INET6
901 struct ip6_hdr *h6 = NULL;
902#endif
903 struct tcphdr *th = NULL;
904
905 MGETHDR(m, M_DONTWAIT, MT_DATA);
906 if (m == NULL)
907 return (NULL);
908
909 M_SETFIB(m, id->fib);
910#ifdef MAC
911 if (replyto != NULL)
912 mac_netinet_firewall_reply(replyto, m);
913 else
914 mac_netinet_firewall_send(m);
915#else
916 (void)replyto; /* don't warn about unused arg */
917#endif
918
919 switch (id->addr_type) {
920 case 4:
921 len = sizeof(struct ip) + sizeof(struct tcphdr);
922 break;
923#ifdef INET6
924 case 6:
925 len = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
926 break;
927#endif
928 default:
929 /* XXX: log me?!? */
930 FREE_PKT(m);
931 return (NULL);
932 }
933 dir = ((flags & (TH_SYN | TH_RST)) == TH_SYN);
934
935 m->m_data += max_linkhdr;
936 m->m_flags |= M_SKIP_FIREWALL;
937 m->m_pkthdr.len = m->m_len = len;
938 m->m_pkthdr.rcvif = NULL;
939 bzero(m->m_data, len);
940
941 switch (id->addr_type) {
942 case 4:
943 h = mtod(m, struct ip *);
944
945 /* prepare for checksum */
946 h->ip_p = IPPROTO_TCP;
947 h->ip_len = htons(sizeof(struct tcphdr));
948 if (dir) {
949 h->ip_src.s_addr = htonl(id->src_ip);
950 h->ip_dst.s_addr = htonl(id->dst_ip);
951 } else {
952 h->ip_src.s_addr = htonl(id->dst_ip);
953 h->ip_dst.s_addr = htonl(id->src_ip);
954 }
955
956 th = (struct tcphdr *)(h + 1);
957 break;
958#ifdef INET6
959 case 6:
960 h6 = mtod(m, struct ip6_hdr *);
961
962 /* prepare for checksum */
963 h6->ip6_nxt = IPPROTO_TCP;
964 h6->ip6_plen = htons(sizeof(struct tcphdr));
965 if (dir) {
966 h6->ip6_src = id->src_ip6;
967 h6->ip6_dst = id->dst_ip6;
968 } else {
969 h6->ip6_src = id->dst_ip6;
970 h6->ip6_dst = id->src_ip6;
971 }
972
973 th = (struct tcphdr *)(h6 + 1);
974 break;
975#endif
976 }
977
978 if (dir) {
979 th->th_sport = htons(id->src_port);
980 th->th_dport = htons(id->dst_port);
981 } else {
982 th->th_sport = htons(id->dst_port);
983 th->th_dport = htons(id->src_port);
984 }
985 th->th_off = sizeof(struct tcphdr) >> 2;
986
987 if (flags & TH_RST) {
988 if (flags & TH_ACK) {
989 th->th_seq = htonl(ack);
990 th->th_flags = TH_RST;
991 } else {
992 if (flags & TH_SYN)
993 seq++;
994 th->th_ack = htonl(seq);
995 th->th_flags = TH_RST | TH_ACK;
996 }
997 } else {
998 /*
999 * Keepalive - use caller provided sequence numbers
1000 */
1001 th->th_seq = htonl(seq);
1002 th->th_ack = htonl(ack);
1003 th->th_flags = TH_ACK;
1004 }
1005
1006 switch (id->addr_type) {
1007 case 4:
1008 th->th_sum = in_cksum(m, len);
1009
1010 /* finish the ip header */
1011 h->ip_v = 4;
1012 h->ip_hl = sizeof(*h) >> 2;
1013 h->ip_tos = IPTOS_LOWDELAY;
1014 h->ip_off = 0;
1015 /* ip_len must be in host format for ip_output */
1016 h->ip_len = len;
1017 h->ip_ttl = V_ip_defttl;
1018 h->ip_sum = 0;
1019 break;
1020#ifdef INET6
1021 case 6:
1022 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(*h6),
1023 sizeof(struct tcphdr));
1024
1025 /* finish the ip6 header */
1026 h6->ip6_vfc |= IPV6_VERSION;
1027 h6->ip6_hlim = IPV6_DEFHLIM;
1028 break;
1029#endif
1030 }
1031
1032 return (m);
1033}
1034
1035/*
1036 * This procedure is only used to handle keepalives. It is invoked
1037 * every dyn_keepalive_period
1038 */
1039static void
1040ipfw_tick(void * vnetx)
1041{
1042 struct mbuf *m0, *m, *mnext, **mtailp;
1043#ifdef INET6
1044 struct mbuf *m6, **m6_tailp;
1045#endif
1046 int i;
1047 ipfw_dyn_rule *q;
1048#ifdef VIMAGE
1049 struct vnet *vp = vnetx;
1050#endif
1051
1052 CURVNET_SET(vp);
1053 if (V_dyn_keepalive == 0 || V_ipfw_dyn_v == NULL || V_dyn_count == 0)
1054 goto done;
1055
1056 /*
1057 * We make a chain of packets to go out here -- not deferring
1058 * until after we drop the IPFW dynamic rule lock would result
1059 * in a lock order reversal with the normal packet input -> ipfw
1060 * call stack.
1061 */
1062 m0 = NULL;
1063 mtailp = &m0;
1064#ifdef INET6
1065 m6 = NULL;
1066 m6_tailp = &m6;
1067#endif
1068 IPFW_DYN_LOCK();
1069 for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
1070 for (q = V_ipfw_dyn_v[i] ; q ; q = q->next ) {
1071 if (q->dyn_type == O_LIMIT_PARENT)
1072 continue;
1073 if (q->id.proto != IPPROTO_TCP)
1074 continue;
1075 if ( (q->state & BOTH_SYN) != BOTH_SYN)
1076 continue;
1077 if (TIME_LEQ(time_uptime + V_dyn_keepalive_interval,
1078 q->expire))
1079 continue; /* too early */
1080 if (TIME_LEQ(q->expire, time_uptime))
1081 continue; /* too late, rule expired */
1082
1083 m = ipfw_send_pkt(NULL, &(q->id), q->ack_rev - 1,
1084 q->ack_fwd, TH_SYN);
1085 mnext = ipfw_send_pkt(NULL, &(q->id), q->ack_fwd - 1,
1086 q->ack_rev, 0);
1087
1088 switch (q->id.addr_type) {
1089 case 4:
1090 if (m != NULL) {
1091 *mtailp = m;
1092 mtailp = &(*mtailp)->m_nextpkt;
1093 }
1094 if (mnext != NULL) {
1095 *mtailp = mnext;
1096 mtailp = &(*mtailp)->m_nextpkt;
1097 }
1098 break;
1099#ifdef INET6
1100 case 6:
1101 if (m != NULL) {
1102 *m6_tailp = m;
1103 m6_tailp = &(*m6_tailp)->m_nextpkt;
1104 }
1105 if (mnext != NULL) {
1106 *m6_tailp = mnext;
1107 m6_tailp = &(*m6_tailp)->m_nextpkt;
1108 }
1109 break;
1110#endif
1111 }
1112
1113 m = mnext = NULL;
1114 }
1115 }
1116 IPFW_DYN_UNLOCK();
1117 for (m = mnext = m0; m != NULL; m = mnext) {
1118 mnext = m->m_nextpkt;
1119 m->m_nextpkt = NULL;
1120 ip_output(m, NULL, NULL, 0, NULL, NULL);
1121 }
1122#ifdef INET6
1123 for (m = mnext = m6; m != NULL; m = mnext) {
1124 mnext = m->m_nextpkt;
1125 m->m_nextpkt = NULL;
1126 ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1127 }
1128#endif
1129done:
1130 callout_reset_on(&V_ipfw_timeout, V_dyn_keepalive_period * hz,
1131 ipfw_tick, vnetx, 0);
1132 CURVNET_RESTORE();
1133}
1134
1135void
1136ipfw_dyn_attach(void)
1137{
1138 ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule",
1139 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL,
1140 UMA_ALIGN_PTR, 0);
1141
1142 IPFW_DYN_LOCK_INIT();
1143}
1144
1145void
1146ipfw_dyn_detach(void)
1147{
1148 uma_zdestroy(ipfw_dyn_rule_zone);
1149 IPFW_DYN_LOCK_DESTROY();
1150}
1151
1152void
1153ipfw_dyn_init(void)
1154{
1155 V_ipfw_dyn_v = NULL;
1156 V_dyn_buckets = 256; /* must be power of 2 */
1157 V_curr_dyn_buckets = 256; /* must be power of 2 */
1158
1159 V_dyn_ack_lifetime = 300;
1160 V_dyn_syn_lifetime = 20;
1161 V_dyn_fin_lifetime = 1;
1162 V_dyn_rst_lifetime = 1;
1163 V_dyn_udp_lifetime = 10;
1164 V_dyn_short_lifetime = 5;
1165
1166 V_dyn_keepalive_interval = 20;
1167 V_dyn_keepalive_period = 5;
1168 V_dyn_keepalive = 1; /* do send keepalives */
1169
1170 V_dyn_max = 4096; /* max # of dynamic rules */
1171 callout_init(&V_ipfw_timeout, CALLOUT_MPSAFE);
1172 callout_reset_on(&V_ipfw_timeout, hz, ipfw_tick, curvnet, 0);
1173}
1174
1175void
1176ipfw_dyn_uninit(int pass)
1177{
1178 if (pass == 0)
1179 callout_drain(&V_ipfw_timeout);
1180 else {
1181 if (V_ipfw_dyn_v != NULL)
1182 free(V_ipfw_dyn_v, M_IPFW);
1183 }
1184}
1185
1186int
1187ipfw_dyn_len(void)
1188{
1189 return (V_ipfw_dyn_v == NULL) ? 0 :
1190 (V_dyn_count * sizeof(ipfw_dyn_rule));
1191}
1192
1193void
1194ipfw_get_dynamic(char **pbp, const char *ep)
1195{
1196 ipfw_dyn_rule *p, *last = NULL;
1197 char *bp;
1198 int i;
1199
1200 if (V_ipfw_dyn_v == NULL)
1201 return;
1202 bp = *pbp;
1203
1204 IPFW_DYN_LOCK();
1205 for (i = 0 ; i < V_curr_dyn_buckets; i++)
1206 for (p = V_ipfw_dyn_v[i] ; p != NULL; p = p->next) {
1207 if (bp + sizeof *p <= ep) {
1208 ipfw_dyn_rule *dst =
1209 (ipfw_dyn_rule *)bp;
1210 bcopy(p, dst, sizeof *p);
1211 bcopy(&(p->rule->rulenum), &(dst->rule),
1212 sizeof(p->rule->rulenum));
1213 /*
1214 * store set number into high word of
1215 * dst->rule pointer.
1216 */
1217 bcopy(&(p->rule->set),
1218 (char *)&dst->rule +
1219 sizeof(p->rule->rulenum),
1220 sizeof(p->rule->set));
1221 /*
1222 * store a non-null value in "next".
1223 * The userland code will interpret a
1224 * NULL here as a marker
1225 * for the last dynamic rule.
1226 */
1227 bcopy(&dst, &dst->next, sizeof(dst));
1228 last = dst;
1229 dst->expire =
1230 TIME_LEQ(dst->expire, time_uptime) ?
1231 0 : dst->expire - time_uptime ;
1232 bp += sizeof(ipfw_dyn_rule);
1233 }
1234 }
1235 IPFW_DYN_UNLOCK();
1236 if (last != NULL) /* mark last dynamic rule */
1237 bzero(&last->next, sizeof(last));
1238 *pbp = bp;
1239}
1240/* end of file */