?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
kqueue: Use wakeup_one based on # of threads sleep on kqueue
[dragonfly.git] / sys / net / ipfw / ip_fw2.c
blob12455eac74166b35d5f56f429170787cb918eac5
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 * $FreeBSD: src/sys/netinet/ip_fw2.c,v 1.6.2.12 2003/04/08 10:42:32 maxim Exp $
26 */
27
28 /*
29 * Implement IP packet firewall (new version)
30 */
31
32 #include "opt_ipfw.h"
33 #include "opt_inet.h"
34 #ifndef INET
35 #error IPFIREWALL requires INET.
36 #endif /* INET */
37
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/malloc.h>
41 #include <sys/mbuf.h>
42 #include <sys/kernel.h>
43 #include <sys/proc.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/sysctl.h>
47 #include <sys/syslog.h>
48 #include <sys/ucred.h>
49 #include <sys/in_cksum.h>
50 #include <sys/lock.h>
51
52 #include <net/if.h>
53 #include <net/route.h>
54 #include <net/pfil.h>
55 #include <net/dummynet/ip_dummynet.h>
56
57 #include <sys/thread2.h>
58 #include <sys/mplock2.h>
59 #include <net/netmsg2.h>
60
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/in_var.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip_var.h>
67 #include <netinet/ip_icmp.h>
68 #include <netinet/tcp.h>
69 #include <netinet/tcp_timer.h>
70 #include <netinet/tcp_var.h>
71 #include <netinet/tcpip.h>
72 #include <netinet/udp.h>
73 #include <netinet/udp_var.h>
74 #include <netinet/ip_divert.h>
75 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
76
77 #include <net/ipfw/ip_fw2.h>
78
79 #ifdef IPFIREWALL_DEBUG
80 #define DPRINTF(fmt, ...) \
81 do { \
82 if (fw_debug > 0) \
83 kprintf(fmt, __VA_ARGS__); \
84 } while (0)
85 #else
86 #define DPRINTF(fmt, ...) ((void)0)
87 #endif
88
89 /*
90 * Description about per-CPU rule duplication:
91 *
92 * Module loading/unloading and all ioctl operations are serialized
93 * by netisr0, so we don't have any ordering or locking problems.
94 *
95 * Following graph shows how operation on per-CPU rule list is
96 * performed [2 CPU case]:
97 *
98 * CPU0 CPU1
99 *
100 * netisr0 <------------------------------------+
101 * domsg |
102 * | |
103 * | netmsg |
104 * | |
105 * V |
106 * ifnet0 |
107 * : | netmsg
108 * :(delete/add...) |
109 * : |
110 * : netmsg |
111 * forwardmsg---------->ifnet1 |
112 * : |
113 * :(delete/add...) |
114 * : |
115 * : |
116 * replymsg--------------+
117 *
118 *
119 *
120 *
121 * Rules which will not create states (dyn rules) [2 CPU case]
122 *
123 * CPU0 CPU1
124 * layer3_chain layer3_chain
125 * | |
126 * V V
127 * +-------+ sibling +-------+ sibling
128 * | rule1 |--------->| rule1 |--------->NULL
129 * +-------+ +-------+
130 * | |
131 * |next |next
132 * V V
133 * +-------+ sibling +-------+ sibling
134 * | rule2 |--------->| rule2 |--------->NULL
135 * +-------+ +-------+
136 *
137 * ip_fw.sibling:
138 * 1) Ease statistics calculation during IP_FW_GET. We only need to
139 * iterate layer3_chain on CPU0; the current rule's duplication on
140 * the other CPUs could safely be read-only accessed by using
141 * ip_fw.sibling
142 * 2) Accelerate rule insertion and deletion, e.g. rule insertion:
143 * a) In netisr0 (on CPU0) rule3 is determined to be inserted between
144 * rule1 and rule2. To make this decision we need to iterate the
145 * layer3_chain on CPU0. The netmsg, which is used to insert the
146 * rule, will contain rule1 on CPU0 as prev_rule and rule2 on CPU0
147 * as next_rule
148 * b) After the insertion on CPU0 is done, we will move on to CPU1.
149 * But instead of relocating the rule3's position on CPU1 by
150 * iterating the layer3_chain on CPU1, we set the netmsg's prev_rule
151 * to rule1->sibling and next_rule to rule2->sibling before the
152 * netmsg is forwarded to CPU1 from CPU0
153 *
154 *
155 *
156 * Rules which will create states (dyn rules) [2 CPU case]
157 * (unnecessary parts are omitted; they are same as in the previous figure)
158 *
159 * CPU0 CPU1
160 *
161 * +-------+ +-------+
162 * | rule1 | | rule1 |
163 * +-------+ +-------+
164 * ^ | | ^
165 * | |stub stub| |
166 * | | | |
167 * | +----+ +----+ |
168 * | | | |
169 * | V V |
170 * | +--------------------+ |
171 * | | rule_stub | |
172 * | | (read-only shared) | |
173 * | | | |
174 * | | back pointer array | |
175 * | | (indexed by cpuid) | |
176 * | | | |
177 * +----|---------[0] | |
178 * | [1]--------|----+
179 * | |
180 * +--------------------+
181 * ^ ^
182 * | |
183 * ........|............|............
184 * : | | :
185 * : |stub |stub :
186 * : | | :
187 * : +---------+ +---------+ :
188 * : | state1a | | state1b | .... :
189 * : +---------+ +---------+ :
190 * : :
191 * : states table :
192 * : (shared) :
193 * : (protected by dyn_lock) :
194 * ..................................
195 *
196 * [state1a and state1b are states created by rule1]
197 *
198 * ip_fw_stub:
199 * This structure is introduced so that shared (locked) state table could
200 * work with per-CPU (duplicated) static rules. It mainly bridges states
201 * and static rules and serves as static rule's place holder (a read-only
202 * shared part of duplicated rules) from states point of view.
203 *
204 * IPFW_RULE_F_STATE (only for rules which create states):
205 * o During rule installation, this flag is turned on after rule's
206 * duplications reach all CPUs, to avoid at least following race:
207 * 1) rule1 is duplicated on CPU0 and is not duplicated on CPU1 yet
208 * 2) rule1 creates state1
209 * 3) state1 is located on CPU1 by check-state
210 * But rule1 is not duplicated on CPU1 yet
211 * o During rule deletion, this flag is turned off before deleting states
212 * created by the rule and before deleting the rule itself, so no
213 * more states will be created by the to-be-deleted rule even when its
214 * duplication on certain CPUs are not eliminated yet.
215 */
216
217 #define IPFW_AUTOINC_STEP_MIN 1
218 #define IPFW_AUTOINC_STEP_MAX 1000
219 #define IPFW_AUTOINC_STEP_DEF 100
220
221 #define IPFW_DEFAULT_RULE 65535 /* rulenum for the default rule */
222 #define IPFW_DEFAULT_SET 31 /* set number for the default rule */
223
224 struct netmsg_ipfw {
225 struct netmsg_base base;
226 const struct ipfw_ioc_rule *ioc_rule;
227 struct ip_fw *next_rule;
228 struct ip_fw *prev_rule;
229 struct ip_fw *sibling;
230 struct ip_fw_stub *stub;
231 };
232
233 struct netmsg_del {
234 struct netmsg_base base;
235 struct ip_fw *start_rule;
236 struct ip_fw *prev_rule;
237 uint16_t rulenum;
238 uint8_t from_set;
239 uint8_t to_set;
240 };
241
242 struct netmsg_zent {
243 struct netmsg_base base;
244 struct ip_fw *start_rule;
245 uint16_t rulenum;
246 uint16_t log_only;
247 };
248
249 struct ipfw_context {
250 struct ip_fw *ipfw_layer3_chain; /* list of rules for layer3 */
251 struct ip_fw *ipfw_default_rule; /* default rule */
252 uint64_t ipfw_norule_counter; /* counter for ipfw_log(NULL) */
253
254 /*
255 * ipfw_set_disable contains one bit per set value (0..31).
256 * If the bit is set, all rules with the corresponding set
257 * are disabled. Set IPDW_DEFAULT_SET is reserved for the
258 * default rule and CANNOT be disabled.
259 */
260 uint32_t ipfw_set_disable;
261 uint32_t ipfw_gen; /* generation of rule list */
262 };
263
264 static struct ipfw_context *ipfw_ctx[MAXCPU];
265
266 #ifdef KLD_MODULE
267 /*
268 * Module can not be unloaded, if there are references to
269 * certains rules of ipfw(4), e.g. dummynet(4)
270 */
271 static int ipfw_refcnt;
272 #endif
273
274 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
275
276 /*
277 * Following two global variables are accessed and
278 * updated only on CPU0
279 */
280 static uint32_t static_count; /* # of static rules */
281 static uint32_t static_ioc_len; /* bytes of static rules */
282
283 /*
284 * If 1, then ipfw static rules are being flushed,
285 * ipfw_chk() will skip to the default rule.
286 */
287 static int ipfw_flushing;
288
289 static int fw_verbose;
290 static int verbose_limit;
291
292 static int fw_debug;
293 static int autoinc_step = IPFW_AUTOINC_STEP_DEF;
294
295 static int ipfw_sysctl_enable(SYSCTL_HANDLER_ARGS);
296 static int ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS);
297 static int ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS);
298 static int ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS);
299 static int ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS);
300
301 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
302 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, enable, CTLTYPE_INT | CTLFLAG_RW,
303 &fw_enable, 0, ipfw_sysctl_enable, "I", "Enable ipfw");
304 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLTYPE_INT | CTLFLAG_RW,
305 &autoinc_step, 0, ipfw_sysctl_autoinc_step, "I",
306 "Rule number autincrement step");
307 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW,
308 &fw_one_pass, 0,
309 "Only do a single pass through ipfw when using dummynet(4)");
310 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
311 &fw_debug, 0, "Enable printing of debug ip_fw statements");
312 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW,
313 &fw_verbose, 0, "Log matches to ipfw rules");
314 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
315 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
316
317 /*
318 * Description of dynamic rules.
319 *
320 * Dynamic rules are stored in lists accessed through a hash table
321 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
322 * be modified through the sysctl variable dyn_buckets which is
323 * updated when the table becomes empty.
324 *
325 * XXX currently there is only one list, ipfw_dyn.
326 *
327 * When a packet is received, its address fields are first masked
328 * with the mask defined for the rule, then hashed, then matched
329 * against the entries in the corresponding list.
330 * Dynamic rules can be used for different purposes:
331 * + stateful rules;
332 * + enforcing limits on the number of sessions;
333 * + in-kernel NAT (not implemented yet)
334 *
335 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
336 * measured in seconds and depending on the flags.
337 *
338 * The total number of dynamic rules is stored in dyn_count.
339 * The max number of dynamic rules is dyn_max. When we reach
340 * the maximum number of rules we do not create anymore. This is
341 * done to avoid consuming too much memory, but also too much
342 * time when searching on each packet (ideally, we should try instead
343 * to put a limit on the length of the list on each bucket...).
344 *
345 * Each dynamic rule holds a pointer to the parent ipfw rule so
346 * we know what action to perform. Dynamic rules are removed when
347 * the parent rule is deleted. XXX we should make them survive.
348 *
349 * There are some limitations with dynamic rules -- we do not
350 * obey the 'randomized match', and we do not do multiple
351 * passes through the firewall. XXX check the latter!!!
352 *
353 * NOTE about the SHARED LOCKMGR LOCK during dynamic rule looking up:
354 * Only TCP state transition will change dynamic rule's state and ack
355 * sequences, while all packets of one TCP connection only goes through
356 * one TCP thread, so it is safe to use shared lockmgr lock during dynamic
357 * rule looking up. The keep alive callout uses exclusive lockmgr lock
358 * when it tries to find suitable dynamic rules to send keep alive, so
359 * it will not see half updated state and ack sequences. Though the expire
360 * field updating looks racy for other protocols, the resolution (second)
361 * of expire field makes this kind of race harmless.
362 * XXX statistics' updating is _not_ MPsafe!!!
363 * XXX once UDP output path is fixed, we could use lockless dynamic rule
364 * hash table
365 */
366 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
367 static uint32_t dyn_buckets = 256; /* must be power of 2 */
368 static uint32_t curr_dyn_buckets = 256; /* must be power of 2 */
369 static uint32_t dyn_buckets_gen; /* generation of dyn buckets array */
370 static struct lock dyn_lock; /* dynamic rules' hash table lock */
371
372 static struct netmsg_base ipfw_timeout_netmsg; /* schedule ipfw timeout */
373 static struct callout ipfw_timeout_h;
374
375 /*
376 * Timeouts for various events in handing dynamic rules.
377 */
378 static uint32_t dyn_ack_lifetime = 300;
379 static uint32_t dyn_syn_lifetime = 20;
380 static uint32_t dyn_fin_lifetime = 1;
381 static uint32_t dyn_rst_lifetime = 1;
382 static uint32_t dyn_udp_lifetime = 10;
383 static uint32_t dyn_short_lifetime = 5;
384
385 /*
386 * Keepalives are sent if dyn_keepalive is set. They are sent every
387 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
388 * seconds of lifetime of a rule.
389 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
390 * than dyn_keepalive_period.
391 */
392
393 static uint32_t dyn_keepalive_interval = 20;
394 static uint32_t dyn_keepalive_period = 5;
395 static uint32_t dyn_keepalive = 1; /* do send keepalives */
396
397 static uint32_t dyn_count; /* # of dynamic rules */
398 static uint32_t dyn_max = 4096; /* max # of dynamic rules */
399
400 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLTYPE_INT | CTLFLAG_RW,
401 &dyn_buckets, 0, ipfw_sysctl_dyn_buckets, "I", "Number of dyn. buckets");
402 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
403 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
404 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
405 &dyn_count, 0, "Number of dyn. rules");
406 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
407 &dyn_max, 0, "Max number of dyn. rules");
408 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
409 &static_count, 0, "Number of static rules");
410 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
411 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
412 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
413 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
414 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime,
415 CTLTYPE_INT | CTLFLAG_RW, &dyn_fin_lifetime, 0, ipfw_sysctl_dyn_fin, "I",
416 "Lifetime of dyn. rules for fin");
417 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime,
418 CTLTYPE_INT | CTLFLAG_RW, &dyn_rst_lifetime, 0, ipfw_sysctl_dyn_rst, "I",
419 "Lifetime of dyn. rules for rst");
420 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
421 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
422 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
423 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
424 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
425 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
426
427 static ip_fw_chk_t ipfw_chk;
428 static void ipfw_tick(void *);
429
430 static __inline int
431 ipfw_free_rule(struct ip_fw *rule)
432 {
433 KASSERT(rule->cpuid == mycpuid, ("rule freed on cpu%d", mycpuid));
434 KASSERT(rule->refcnt > 0, ("invalid refcnt %u", rule->refcnt));
435 rule->refcnt--;
436 if (rule->refcnt == 0) {
437 kfree(rule, M_IPFW);
438 return 1;
439 }
440 return 0;
441 }
442
443 static void
444 ipfw_unref_rule(void *priv)
445 {
446 ipfw_free_rule(priv);
447 #ifdef KLD_MODULE
448 atomic_subtract_int(&ipfw_refcnt, 1);
449 #endif
450 }
451
452 static __inline void
453 ipfw_ref_rule(struct ip_fw *rule)
454 {
455 KASSERT(rule->cpuid == mycpuid, ("rule used on cpu%d", mycpuid));
456 #ifdef KLD_MODULE
457 atomic_add_int(&ipfw_refcnt, 1);
458 #endif
459 rule->refcnt++;
460 }
461
462 /*
463 * This macro maps an ip pointer into a layer3 header pointer of type T
464 */
465 #define L3HDR(T, ip) ((T *)((uint32_t *)(ip) + (ip)->ip_hl))
466
467 static __inline int
468 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
469 {
470 int type = L3HDR(struct icmp,ip)->icmp_type;
471
472 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1 << type)));
473 }
474
475 #define TT ((1 << ICMP_ECHO) | \
476 (1 << ICMP_ROUTERSOLICIT) | \
477 (1 << ICMP_TSTAMP) | \
478 (1 << ICMP_IREQ) | \
479 (1 << ICMP_MASKREQ))
480
481 static int
482 is_icmp_query(struct ip *ip)
483 {
484 int type = L3HDR(struct icmp, ip)->icmp_type;
485
486 return (type <= ICMP_MAXTYPE && (TT & (1 << type)));
487 }
488
489 #undef TT
490
491 /*
492 * The following checks use two arrays of 8 or 16 bits to store the
493 * bits that we want set or clear, respectively. They are in the
494 * low and high half of cmd->arg1 or cmd->d[0].
495 *
496 * We scan options and store the bits we find set. We succeed if
497 *
498 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
499 *
500 * The code is sometimes optimized not to store additional variables.
501 */
502
503 static int
504 flags_match(ipfw_insn *cmd, uint8_t bits)
505 {
506 u_char want_clear;
507 bits = ~bits;
508
509 if (((cmd->arg1 & 0xff) & bits) != 0)
510 return 0; /* some bits we want set were clear */
511
512 want_clear = (cmd->arg1 >> 8) & 0xff;
513 if ((want_clear & bits) != want_clear)
514 return 0; /* some bits we want clear were set */
515 return 1;
516 }
517
518 static int
519 ipopts_match(struct ip *ip, ipfw_insn *cmd)
520 {
521 int optlen, bits = 0;
522 u_char *cp = (u_char *)(ip + 1);
523 int x = (ip->ip_hl << 2) - sizeof(struct ip);
524
525 for (; x > 0; x -= optlen, cp += optlen) {
526 int opt = cp[IPOPT_OPTVAL];
527
528 if (opt == IPOPT_EOL)
529 break;
530
531 if (opt == IPOPT_NOP) {
532 optlen = 1;
533 } else {
534 optlen = cp[IPOPT_OLEN];
535 if (optlen <= 0 || optlen > x)
536 return 0; /* invalid or truncated */
537 }
538
539 switch (opt) {
540 case IPOPT_LSRR:
541 bits |= IP_FW_IPOPT_LSRR;
542 break;
543
544 case IPOPT_SSRR:
545 bits |= IP_FW_IPOPT_SSRR;
546 break;
547
548 case IPOPT_RR:
549 bits |= IP_FW_IPOPT_RR;
550 break;
551
552 case IPOPT_TS:
553 bits |= IP_FW_IPOPT_TS;
554 break;
555
556 default:
557 break;
558 }
559 }
560 return (flags_match(cmd, bits));
561 }
562
563 static int
564 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
565 {
566 int optlen, bits = 0;
567 struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
568 u_char *cp = (u_char *)(tcp + 1);
569 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
570
571 for (; x > 0; x -= optlen, cp += optlen) {
572 int opt = cp[0];
573
574 if (opt == TCPOPT_EOL)
575 break;
576
577 if (opt == TCPOPT_NOP) {
578 optlen = 1;
579 } else {
580 optlen = cp[1];
581 if (optlen <= 0)
582 break;
583 }
584
585 switch (opt) {
586 case TCPOPT_MAXSEG:
587 bits |= IP_FW_TCPOPT_MSS;
588 break;
589
590 case TCPOPT_WINDOW:
591 bits |= IP_FW_TCPOPT_WINDOW;
592 break;
593
594 case TCPOPT_SACK_PERMITTED:
595 case TCPOPT_SACK:
596 bits |= IP_FW_TCPOPT_SACK;
597 break;
598
599 case TCPOPT_TIMESTAMP:
600 bits |= IP_FW_TCPOPT_TS;
601 break;
602
603 case TCPOPT_CC:
604 case TCPOPT_CCNEW:
605 case TCPOPT_CCECHO:
606 bits |= IP_FW_TCPOPT_CC;
607 break;
608
609 default:
610 break;
611 }
612 }
613 return (flags_match(cmd, bits));
614 }
615
616 static int
617 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
618 {
619 if (ifp == NULL) /* no iface with this packet, match fails */
620 return 0;
621
622 /* Check by name or by IP address */
623 if (cmd->name[0] != '\0') { /* match by name */
624 /* Check name */
625 if (cmd->p.glob) {
626 if (kfnmatch(cmd->name, ifp->if_xname, 0) == 0)
627 return(1);
628 } else {
629 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
630 return(1);
631 }
632 } else {
633 struct ifaddr_container *ifac;
634
635 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
636 struct ifaddr *ia = ifac->ifa;
637
638 if (ia->ifa_addr == NULL)
639 continue;
640 if (ia->ifa_addr->sa_family != AF_INET)
641 continue;
642 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
643 (ia->ifa_addr))->sin_addr.s_addr)
644 return(1); /* match */
645 }
646 }
647 return(0); /* no match, fail ... */
648 }
649
650 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
651
652 /*
653 * We enter here when we have a rule with O_LOG.
654 * XXX this function alone takes about 2Kbytes of code!
655 */
656 static void
657 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
658 struct mbuf *m, struct ifnet *oif)
659 {
660 char *action;
661 int limit_reached = 0;
662 char action2[40], proto[48], fragment[28];
663
664 fragment[0] = '\0';
665 proto[0] = '\0';
666
667 if (f == NULL) { /* bogus pkt */
668 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
669
670 if (verbose_limit != 0 &&
671 ctx->ipfw_norule_counter >= verbose_limit)
672 return;
673 ctx->ipfw_norule_counter++;
674 if (ctx->ipfw_norule_counter == verbose_limit)
675 limit_reached = verbose_limit;
676 action = "Refuse";
677 } else { /* O_LOG is the first action, find the real one */
678 ipfw_insn *cmd = ACTION_PTR(f);
679 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
680
681 if (l->max_log != 0 && l->log_left == 0)
682 return;
683 l->log_left--;
684 if (l->log_left == 0)
685 limit_reached = l->max_log;
686 cmd += F_LEN(cmd); /* point to first action */
687 if (cmd->opcode == O_PROB)
688 cmd += F_LEN(cmd);
689
690 action = action2;
691 switch (cmd->opcode) {
692 case O_DENY:
693 action = "Deny";
694 break;
695
696 case O_REJECT:
697 if (cmd->arg1==ICMP_REJECT_RST) {
698 action = "Reset";
699 } else if (cmd->arg1==ICMP_UNREACH_HOST) {
700 action = "Reject";
701 } else {
702 ksnprintf(SNPARGS(action2, 0), "Unreach %d",
703 cmd->arg1);
704 }
705 break;
706
707 case O_ACCEPT:
708 action = "Accept";
709 break;
710
711 case O_COUNT:
712 action = "Count";
713 break;
714
715 case O_DIVERT:
716 ksnprintf(SNPARGS(action2, 0), "Divert %d", cmd->arg1);
717 break;
718
719 case O_TEE:
720 ksnprintf(SNPARGS(action2, 0), "Tee %d", cmd->arg1);
721 break;
722
723 case O_SKIPTO:
724 ksnprintf(SNPARGS(action2, 0), "SkipTo %d", cmd->arg1);
725 break;
726
727 case O_PIPE:
728 ksnprintf(SNPARGS(action2, 0), "Pipe %d", cmd->arg1);
729 break;
730
731 case O_QUEUE:
732 ksnprintf(SNPARGS(action2, 0), "Queue %d", cmd->arg1);
733 break;
734
735 case O_FORWARD_IP:
736 {
737 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
738 int len;
739
740 len = ksnprintf(SNPARGS(action2, 0),
741 "Forward to %s",
742 inet_ntoa(sa->sa.sin_addr));
743 if (sa->sa.sin_port) {
744 ksnprintf(SNPARGS(action2, len), ":%d",
745 sa->sa.sin_port);
746 }
747 }
748 break;
749
750 default:
751 action = "UNKNOWN";
752 break;
753 }
754 }
755
756 if (hlen == 0) { /* non-ip */
757 ksnprintf(SNPARGS(proto, 0), "MAC");
758 } else {
759 struct ip *ip = mtod(m, struct ip *);
760 /* these three are all aliases to the same thing */
761 struct icmp *const icmp = L3HDR(struct icmp, ip);
762 struct tcphdr *const tcp = (struct tcphdr *)icmp;
763 struct udphdr *const udp = (struct udphdr *)icmp;
764
765 int ip_off, offset, ip_len;
766 int len;
767
768 if (eh != NULL) { /* layer 2 packets are as on the wire */
769 ip_off = ntohs(ip->ip_off);
770 ip_len = ntohs(ip->ip_len);
771 } else {
772 ip_off = ip->ip_off;
773 ip_len = ip->ip_len;
774 }
775 offset = ip_off & IP_OFFMASK;
776 switch (ip->ip_p) {
777 case IPPROTO_TCP:
778 len = ksnprintf(SNPARGS(proto, 0), "TCP %s",
779 inet_ntoa(ip->ip_src));
780 if (offset == 0) {
781 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
782 ntohs(tcp->th_sport),
783 inet_ntoa(ip->ip_dst),
784 ntohs(tcp->th_dport));
785 } else {
786 ksnprintf(SNPARGS(proto, len), " %s",
787 inet_ntoa(ip->ip_dst));
788 }
789 break;
790
791 case IPPROTO_UDP:
792 len = ksnprintf(SNPARGS(proto, 0), "UDP %s",
793 inet_ntoa(ip->ip_src));
794 if (offset == 0) {
795 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
796 ntohs(udp->uh_sport),
797 inet_ntoa(ip->ip_dst),
798 ntohs(udp->uh_dport));
799 } else {
800 ksnprintf(SNPARGS(proto, len), " %s",
801 inet_ntoa(ip->ip_dst));
802 }
803 break;
804
805 case IPPROTO_ICMP:
806 if (offset == 0) {
807 len = ksnprintf(SNPARGS(proto, 0),
808 "ICMP:%u.%u ",
809 icmp->icmp_type,
810 icmp->icmp_code);
811 } else {
812 len = ksnprintf(SNPARGS(proto, 0), "ICMP ");
813 }
814 len += ksnprintf(SNPARGS(proto, len), "%s",
815 inet_ntoa(ip->ip_src));
816 ksnprintf(SNPARGS(proto, len), " %s",
817 inet_ntoa(ip->ip_dst));
818 break;
819
820 default:
821 len = ksnprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
822 inet_ntoa(ip->ip_src));
823 ksnprintf(SNPARGS(proto, len), " %s",
824 inet_ntoa(ip->ip_dst));
825 break;
826 }
827
828 if (ip_off & (IP_MF | IP_OFFMASK)) {
829 ksnprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
830 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
831 offset << 3, (ip_off & IP_MF) ? "+" : "");
832 }
833 }
834
835 if (oif || m->m_pkthdr.rcvif) {
836 log(LOG_SECURITY | LOG_INFO,
837 "ipfw: %d %s %s %s via %s%s\n",
838 f ? f->rulenum : -1,
839 action, proto, oif ? "out" : "in",
840 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
841 fragment);
842 } else {
843 log(LOG_SECURITY | LOG_INFO,
844 "ipfw: %d %s %s [no if info]%s\n",
845 f ? f->rulenum : -1,
846 action, proto, fragment);
847 }
848
849 if (limit_reached) {
850 log(LOG_SECURITY | LOG_NOTICE,
851 "ipfw: limit %d reached on entry %d\n",
852 limit_reached, f ? f->rulenum : -1);
853 }
854 }
855
856 #undef SNPARGS
857
858 /*
859 * IMPORTANT: the hash function for dynamic rules must be commutative
860 * in source and destination (ip,port), because rules are bidirectional
861 * and we want to find both in the same bucket.
862 */
863 static __inline int
864 hash_packet(struct ipfw_flow_id *id)
865 {
866 uint32_t i;
867
868 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
869 i &= (curr_dyn_buckets - 1);
870 return i;
871 }
872
873 /**
874 * unlink a dynamic rule from a chain. prev is a pointer to
875 * the previous one, q is a pointer to the rule to delete,
876 * head is a pointer to the head of the queue.
877 * Modifies q and potentially also head.
878 */
879 #define UNLINK_DYN_RULE(prev, head, q) \
880 do { \
881 ipfw_dyn_rule *old_q = q; \
882 \
883 /* remove a refcount to the parent */ \
884 if (q->dyn_type == O_LIMIT) \
885 q->parent->count--; \
886 DPRINTF("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n", \
887 q->id.src_ip, q->id.src_port, \
888 q->id.dst_ip, q->id.dst_port, dyn_count - 1); \
889 if (prev != NULL) \
890 prev->next = q = q->next; \
891 else \
892 head = q = q->next; \
893 KASSERT(dyn_count > 0, ("invalid dyn count %u", dyn_count)); \
894 dyn_count--; \
895 kfree(old_q, M_IPFW); \
896 } while (0)
897
898 #define TIME_LEQ(a, b) ((int)((a) - (b)) <= 0)
899
900 /**
901 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
902 *
903 * If keep_me == NULL, rules are deleted even if not expired,
904 * otherwise only expired rules are removed.
905 *
906 * The value of the second parameter is also used to point to identify
907 * a rule we absolutely do not want to remove (e.g. because we are
908 * holding a reference to it -- this is the case with O_LIMIT_PARENT
909 * rules). The pointer is only used for comparison, so any non-null
910 * value will do.
911 */
912 static void
913 remove_dyn_rule_locked(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
914 {
915 static time_t last_remove = 0; /* XXX */
916
917 #define FORCE (keep_me == NULL)
918
919 ipfw_dyn_rule *prev, *q;
920 int i, pass = 0, max_pass = 0, unlinked = 0;
921
922 if (ipfw_dyn_v == NULL || dyn_count == 0)
923 return;
924 /* do not expire more than once per second, it is useless */
925 if (!FORCE && last_remove == time_uptime)
926 return;
927 last_remove = time_uptime;
928
929 /*
930 * because O_LIMIT refer to parent rules, during the first pass only
931 * remove child and mark any pending LIMIT_PARENT, and remove
932 * them in a second pass.
933 */
934 next_pass:
935 for (i = 0; i < curr_dyn_buckets; i++) {
936 for (prev = NULL, q = ipfw_dyn_v[i]; q;) {
937 /*
938 * Logic can become complex here, so we split tests.
939 */
940 if (q == keep_me)
941 goto next;
942 if (rule != NULL && rule->stub != q->stub)
943 goto next; /* not the one we are looking for */
944 if (q->dyn_type == O_LIMIT_PARENT) {
945 /*
946 * handle parent in the second pass,
947 * record we need one.
948 */
949 max_pass = 1;
950 if (pass == 0)
951 goto next;
952 if (FORCE && q->count != 0) {
953 /* XXX should not happen! */
954 kprintf("OUCH! cannot remove rule, "
955 "count %d\n", q->count);
956 }
957 } else {
958 if (!FORCE && !TIME_LEQ(q->expire, time_second))
959 goto next;
960 }
961 unlinked = 1;
962 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
963 continue;
964 next:
965 prev = q;
966 q = q->next;
967 }
968 }
969 if (pass++ < max_pass)
970 goto next_pass;
971
972 if (unlinked)
973 ++dyn_buckets_gen;
974
975 #undef FORCE
976 }
977
978 /**
979 * lookup a dynamic rule.
980 */
981 static ipfw_dyn_rule *
982 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
983 struct tcphdr *tcp)
984 {
985 /*
986 * stateful ipfw extensions.
987 * Lookup into dynamic session queue
988 */
989 #define MATCH_REVERSE 0
990 #define MATCH_FORWARD 1
991 #define MATCH_NONE 2
992 #define MATCH_UNKNOWN 3
993 int i, dir = MATCH_NONE;
994 ipfw_dyn_rule *q=NULL;
995
996 if (ipfw_dyn_v == NULL)
997 goto done; /* not found */
998
999 i = hash_packet(pkt);
1000 for (q = ipfw_dyn_v[i]; q != NULL;) {
1001 if (q->dyn_type == O_LIMIT_PARENT)
1002 goto next;
1003
1004 if (TIME_LEQ(q->expire, time_second)) {
1005 /*
1006 * Entry expired; skip.
1007 * Let ipfw_tick() take care of it
1008 */
1009 goto next;
1010 }
1011
1012 if (pkt->proto == q->id.proto) {
1013 if (pkt->src_ip == q->id.src_ip &&
1014 pkt->dst_ip == q->id.dst_ip &&
1015 pkt->src_port == q->id.src_port &&
1016 pkt->dst_port == q->id.dst_port) {
1017 dir = MATCH_FORWARD;
1018 break;
1019 }
1020 if (pkt->src_ip == q->id.dst_ip &&
1021 pkt->dst_ip == q->id.src_ip &&
1022 pkt->src_port == q->id.dst_port &&
1023 pkt->dst_port == q->id.src_port) {
1024 dir = MATCH_REVERSE;
1025 break;
1026 }
1027 }
1028 next:
1029 q = q->next;
1030 }
1031 if (q == NULL)
1032 goto done; /* q = NULL, not found */
1033
1034 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
1035 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
1036
1037 #define BOTH_SYN (TH_SYN | (TH_SYN << 8))
1038 #define BOTH_FIN (TH_FIN | (TH_FIN << 8))
1039
1040 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
1041 switch (q->state) {
1042 case TH_SYN: /* opening */
1043 q->expire = time_second + dyn_syn_lifetime;
1044 break;
1045
1046 case BOTH_SYN: /* move to established */
1047 case BOTH_SYN | TH_FIN : /* one side tries to close */
1048 case BOTH_SYN | (TH_FIN << 8) :
1049 if (tcp) {
1050 uint32_t ack = ntohl(tcp->th_ack);
1051
1052 #define _SEQ_GE(a, b) ((int)(a) - (int)(b) >= 0)
1053
1054 if (dir == MATCH_FORWARD) {
1055 if (q->ack_fwd == 0 ||
1056 _SEQ_GE(ack, q->ack_fwd))
1057 q->ack_fwd = ack;
1058 else /* ignore out-of-sequence */
1059 break;
1060 } else {
1061 if (q->ack_rev == 0 ||
1062 _SEQ_GE(ack, q->ack_rev))
1063 q->ack_rev = ack;
1064 else /* ignore out-of-sequence */
1065 break;
1066 }
1067 #undef _SEQ_GE
1068 }
1069 q->expire = time_second + dyn_ack_lifetime;
1070 break;
1071
1072 case BOTH_SYN | BOTH_FIN: /* both sides closed */
1073 KKASSERT(dyn_fin_lifetime < dyn_keepalive_period);
1074 q->expire = time_second + dyn_fin_lifetime;
1075 break;
1076
1077 default:
1078 #if 0
1079 /*
1080 * reset or some invalid combination, but can also
1081 * occur if we use keep-state the wrong way.
1082 */
1083 if ((q->state & ((TH_RST << 8) | TH_RST)) == 0)
1084 kprintf("invalid state: 0x%x\n", q->state);
1085 #endif
1086 KKASSERT(dyn_rst_lifetime < dyn_keepalive_period);
1087 q->expire = time_second + dyn_rst_lifetime;
1088 break;
1089 }
1090 } else if (pkt->proto == IPPROTO_UDP) {
1091 q->expire = time_second + dyn_udp_lifetime;
1092 } else {
1093 /* other protocols */
1094 q->expire = time_second + dyn_short_lifetime;
1095 }
1096 done:
1097 if (match_direction)
1098 *match_direction = dir;
1099 return q;
1100 }
1101
1102 static struct ip_fw *
1103 lookup_rule(struct ipfw_flow_id *pkt, int *match_direction, struct tcphdr *tcp,
1104 uint16_t len, int *deny)
1105 {
1106 struct ip_fw *rule = NULL;
1107 ipfw_dyn_rule *q;
1108 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1109 uint32_t gen;
1110
1111 *deny = 0;
1112 gen = ctx->ipfw_gen;
1113
1114 lockmgr(&dyn_lock, LK_SHARED);
1115
1116 if (ctx->ipfw_gen != gen) {
1117 /*
1118 * Static rules had been change when we were waiting
1119 * for the dynamic hash table lock; deny this packet,
1120 * since it is _not_ known whether it is safe to keep
1121 * iterating the static rules.
1122 */
1123 *deny = 1;
1124 goto back;
1125 }
1126
1127 q = lookup_dyn_rule(pkt, match_direction, tcp);
1128 if (q == NULL) {
1129 rule = NULL;
1130 } else {
1131 rule = q->stub->rule[mycpuid];
1132 KKASSERT(rule->stub == q->stub && rule->cpuid == mycpuid);
1133
1134 /* XXX */
1135 q->pcnt++;
1136 q->bcnt += len;
1137 }
1138 back:
1139 lockmgr(&dyn_lock, LK_RELEASE);
1140 return rule;
1141 }
1142
1143 static void
1144 realloc_dynamic_table(void)
1145 {
1146 ipfw_dyn_rule **old_dyn_v;
1147 uint32_t old_curr_dyn_buckets;
1148
1149 KASSERT(dyn_buckets <= 65536 && (dyn_buckets & (dyn_buckets - 1)) == 0,
1150 ("invalid dyn_buckets %d", dyn_buckets));
1151
1152 /* Save the current buckets array for later error recovery */
1153 old_dyn_v = ipfw_dyn_v;
1154 old_curr_dyn_buckets = curr_dyn_buckets;
1155
1156 curr_dyn_buckets = dyn_buckets;
1157 for (;;) {
1158 ipfw_dyn_v = kmalloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
1159 M_IPFW, M_NOWAIT | M_ZERO);
1160 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
1161 break;
1162
1163 curr_dyn_buckets /= 2;
1164 if (curr_dyn_buckets <= old_curr_dyn_buckets &&
1165 old_dyn_v != NULL) {
1166 /*
1167 * Don't try allocating smaller buckets array, reuse
1168 * the old one, which alreay contains enough buckets
1169 */
1170 break;
1171 }
1172 }
1173
1174 if (ipfw_dyn_v != NULL) {
1175 if (old_dyn_v != NULL)
1176 kfree(old_dyn_v, M_IPFW);
1177 } else {
1178 /* Allocation failed, restore old buckets array */
1179 ipfw_dyn_v = old_dyn_v;
1180 curr_dyn_buckets = old_curr_dyn_buckets;
1181 }
1182
1183 if (ipfw_dyn_v != NULL)
1184 ++dyn_buckets_gen;
1185 }
1186
1187 /**
1188 * Install state of type 'type' for a dynamic session.
1189 * The hash table contains two type of rules:
1190 * - regular rules (O_KEEP_STATE)
1191 * - rules for sessions with limited number of sess per user
1192 * (O_LIMIT). When they are created, the parent is
1193 * increased by 1, and decreased on delete. In this case,
1194 * the third parameter is the parent rule and not the chain.
1195 * - "parent" rules for the above (O_LIMIT_PARENT).
1196 */
1197 static ipfw_dyn_rule *
1198 add_dyn_rule(struct ipfw_flow_id *id, uint8_t dyn_type, struct ip_fw *rule)
1199 {
1200 ipfw_dyn_rule *r;
1201 int i;
1202
1203 if (ipfw_dyn_v == NULL ||
1204 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1205 realloc_dynamic_table();
1206 if (ipfw_dyn_v == NULL)
1207 return NULL; /* failed ! */
1208 }
1209 i = hash_packet(id);
1210
1211 r = kmalloc(sizeof(*r), M_IPFW, M_NOWAIT | M_ZERO);
1212 if (r == NULL)
1213 return NULL;
1214
1215 /* increase refcount on parent, and set pointer */
1216 if (dyn_type == O_LIMIT) {
1217 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1218
1219 if (parent->dyn_type != O_LIMIT_PARENT)
1220 panic("invalid parent");
1221 parent->count++;
1222 r->parent = parent;
1223 rule = parent->stub->rule[mycpuid];
1224 KKASSERT(rule->stub == parent->stub);
1225 }
1226 KKASSERT(rule->cpuid == mycpuid && rule->stub != NULL);
1227
1228 r->id = *id;
1229 r->expire = time_second + dyn_syn_lifetime;
1230 r->stub = rule->stub;
1231 r->dyn_type = dyn_type;
1232 r->pcnt = r->bcnt = 0;
1233 r->count = 0;
1234
1235 r->bucket = i;
1236 r->next = ipfw_dyn_v[i];
1237 ipfw_dyn_v[i] = r;
1238 dyn_count++;
1239 dyn_buckets_gen++;
1240 DPRINTF("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1241 dyn_type,
1242 r->id.src_ip, r->id.src_port,
1243 r->id.dst_ip, r->id.dst_port, dyn_count);
1244 return r;
1245 }
1246
1247 /**
1248 * lookup dynamic parent rule using pkt and rule as search keys.
1249 * If the lookup fails, then install one.
1250 */
1251 static ipfw_dyn_rule *
1252 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1253 {
1254 ipfw_dyn_rule *q;
1255 int i;
1256
1257 if (ipfw_dyn_v) {
1258 i = hash_packet(pkt);
1259 for (q = ipfw_dyn_v[i]; q != NULL; q = q->next) {
1260 if (q->dyn_type == O_LIMIT_PARENT &&
1261 rule->stub == q->stub &&
1262 pkt->proto == q->id.proto &&
1263 pkt->src_ip == q->id.src_ip &&
1264 pkt->dst_ip == q->id.dst_ip &&
1265 pkt->src_port == q->id.src_port &&
1266 pkt->dst_port == q->id.dst_port) {
1267 q->expire = time_second + dyn_short_lifetime;
1268 DPRINTF("lookup_dyn_parent found 0x%p\n", q);
1269 return q;
1270 }
1271 }
1272 }
1273 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1274 }
1275
1276 /**
1277 * Install dynamic state for rule type cmd->o.opcode
1278 *
1279 * Returns 1 (failure) if state is not installed because of errors or because
1280 * session limitations are enforced.
1281 */
1282 static int
1283 install_state_locked(struct ip_fw *rule, ipfw_insn_limit *cmd,
1284 struct ip_fw_args *args)
1285 {
1286 static int last_log; /* XXX */
1287
1288 ipfw_dyn_rule *q;
1289
1290 DPRINTF("-- install state type %d 0x%08x %u -> 0x%08x %u\n",
1291 cmd->o.opcode,
1292 args->f_id.src_ip, args->f_id.src_port,
1293 args->f_id.dst_ip, args->f_id.dst_port);
1294
1295 q = lookup_dyn_rule(&args->f_id, NULL, NULL);
1296 if (q != NULL) { /* should never occur */
1297 if (last_log != time_second) {
1298 last_log = time_second;
1299 kprintf(" install_state: entry already present, done\n");
1300 }
1301 return 0;
1302 }
1303
1304 if (dyn_count >= dyn_max) {
1305 /*
1306 * Run out of slots, try to remove any expired rule.
1307 */
1308 remove_dyn_rule_locked(NULL, (ipfw_dyn_rule *)1);
1309 if (dyn_count >= dyn_max) {
1310 if (last_log != time_second) {
1311 last_log = time_second;
1312 kprintf("install_state: "
1313 "Too many dynamic rules\n");
1314 }
1315 return 1; /* cannot install, notify caller */
1316 }
1317 }
1318
1319 switch (cmd->o.opcode) {
1320 case O_KEEP_STATE: /* bidir rule */
1321 if (add_dyn_rule(&args->f_id, O_KEEP_STATE, rule) == NULL)
1322 return 1;
1323 break;
1324
1325 case O_LIMIT: /* limit number of sessions */
1326 {
1327 uint16_t limit_mask = cmd->limit_mask;
1328 struct ipfw_flow_id id;
1329 ipfw_dyn_rule *parent;
1330
1331 DPRINTF("installing dyn-limit rule %d\n",
1332 cmd->conn_limit);
1333
1334 id.dst_ip = id.src_ip = 0;
1335 id.dst_port = id.src_port = 0;
1336 id.proto = args->f_id.proto;
1337
1338 if (limit_mask & DYN_SRC_ADDR)
1339 id.src_ip = args->f_id.src_ip;
1340 if (limit_mask & DYN_DST_ADDR)
1341 id.dst_ip = args->f_id.dst_ip;
1342 if (limit_mask & DYN_SRC_PORT)
1343 id.src_port = args->f_id.src_port;
1344 if (limit_mask & DYN_DST_PORT)
1345 id.dst_port = args->f_id.dst_port;
1346
1347 parent = lookup_dyn_parent(&id, rule);
1348 if (parent == NULL) {
1349 kprintf("add parent failed\n");
1350 return 1;
1351 }
1352
1353 if (parent->count >= cmd->conn_limit) {
1354 /*
1355 * See if we can remove some expired rule.
1356 */
1357 remove_dyn_rule_locked(rule, parent);
1358 if (parent->count >= cmd->conn_limit) {
1359 if (fw_verbose &&
1360 last_log != time_second) {
1361 last_log = time_second;
1362 log(LOG_SECURITY | LOG_DEBUG,
1363 "drop session, "
1364 "too many entries\n");
1365 }
1366 return 1;
1367 }
1368 }
1369 if (add_dyn_rule(&args->f_id, O_LIMIT,
1370 (struct ip_fw *)parent) == NULL)
1371 return 1;
1372 }
1373 break;
1374 default:
1375 kprintf("unknown dynamic rule type %u\n", cmd->o.opcode);
1376 return 1;
1377 }
1378 lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1379 return 0;
1380 }
1381
1382 static int
1383 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1384 struct ip_fw_args *args, int *deny)
1385 {
1386 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1387 uint32_t gen;
1388 int ret = 0;
1389
1390 *deny = 0;
1391 gen = ctx->ipfw_gen;
1392
1393 lockmgr(&dyn_lock, LK_EXCLUSIVE);
1394 if (ctx->ipfw_gen != gen) {
1395 /* See the comment in lookup_rule() */
1396 *deny = 1;
1397 } else {
1398 ret = install_state_locked(rule, cmd, args);
1399 }
1400 lockmgr(&dyn_lock, LK_RELEASE);
1401
1402 return ret;
1403 }
1404
1405 /*
1406 * Transmit a TCP packet, containing either a RST or a keepalive.
1407 * When flags & TH_RST, we are sending a RST packet, because of a
1408 * "reset" action matched the packet.
1409 * Otherwise we are sending a keepalive, and flags & TH_
1410 */
1411 static void
1412 send_pkt(struct ipfw_flow_id *id, uint32_t seq, uint32_t ack, int flags)
1413 {
1414 struct mbuf *m;
1415 struct ip *ip;
1416 struct tcphdr *tcp;
1417 struct route sro; /* fake route */
1418
1419 MGETHDR(m, M_NOWAIT, MT_HEADER);
1420 if (m == NULL)
1421 return;
1422 m->m_pkthdr.rcvif = NULL;
1423 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1424 m->m_data += max_linkhdr;
1425
1426 ip = mtod(m, struct ip *);
1427 bzero(ip, m->m_len);
1428 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1429 ip->ip_p = IPPROTO_TCP;
1430 tcp->th_off = 5;
1431
1432 /*
1433 * Assume we are sending a RST (or a keepalive in the reverse
1434 * direction), swap src and destination addresses and ports.
1435 */
1436 ip->ip_src.s_addr = htonl(id->dst_ip);
1437 ip->ip_dst.s_addr = htonl(id->src_ip);
1438 tcp->th_sport = htons(id->dst_port);
1439 tcp->th_dport = htons(id->src_port);
1440 if (flags & TH_RST) { /* we are sending a RST */
1441 if (flags & TH_ACK) {
1442 tcp->th_seq = htonl(ack);
1443 tcp->th_ack = htonl(0);
1444 tcp->th_flags = TH_RST;
1445 } else {
1446 if (flags & TH_SYN)
1447 seq++;
1448 tcp->th_seq = htonl(0);
1449 tcp->th_ack = htonl(seq);
1450 tcp->th_flags = TH_RST | TH_ACK;
1451 }
1452 } else {
1453 /*
1454 * We are sending a keepalive. flags & TH_SYN determines
1455 * the direction, forward if set, reverse if clear.
1456 * NOTE: seq and ack are always assumed to be correct
1457 * as set by the caller. This may be confusing...
1458 */
1459 if (flags & TH_SYN) {
1460 /*
1461 * we have to rewrite the correct addresses!
1462 */
1463 ip->ip_dst.s_addr = htonl(id->dst_ip);
1464 ip->ip_src.s_addr = htonl(id->src_ip);
1465 tcp->th_dport = htons(id->dst_port);
1466 tcp->th_sport = htons(id->src_port);
1467 }
1468 tcp->th_seq = htonl(seq);
1469 tcp->th_ack = htonl(ack);
1470 tcp->th_flags = TH_ACK;
1471 }
1472
1473 /*
1474 * set ip_len to the payload size so we can compute
1475 * the tcp checksum on the pseudoheader
1476 * XXX check this, could save a couple of words ?
1477 */
1478 ip->ip_len = htons(sizeof(struct tcphdr));
1479 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1480
1481 /*
1482 * now fill fields left out earlier
1483 */
1484 ip->ip_ttl = ip_defttl;
1485 ip->ip_len = m->m_pkthdr.len;
1486
1487 bzero(&sro, sizeof(sro));
1488 ip_rtaddr(ip->ip_dst, &sro);
1489
1490 m->m_pkthdr.fw_flags |= IPFW_MBUF_GENERATED;
1491 ip_output(m, NULL, &sro, 0, NULL, NULL);
1492 if (sro.ro_rt)
1493 RTFREE(sro.ro_rt);
1494 }
1495
1496 /*
1497 * sends a reject message, consuming the mbuf passed as an argument.
1498 */
1499 static void
1500 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1501 {
1502 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1503 /* We need the IP header in host order for icmp_error(). */
1504 if (args->eh != NULL) {
1505 struct ip *ip = mtod(args->m, struct ip *);
1506
1507 ip->ip_len = ntohs(ip->ip_len);
1508 ip->ip_off = ntohs(ip->ip_off);
1509 }
1510 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1511 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1512 struct tcphdr *const tcp =
1513 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1514
1515 if ((tcp->th_flags & TH_RST) == 0) {
1516 send_pkt(&args->f_id, ntohl(tcp->th_seq),
1517 ntohl(tcp->th_ack), tcp->th_flags | TH_RST);
1518 }
1519 m_freem(args->m);
1520 } else {
1521 m_freem(args->m);
1522 }
1523 args->m = NULL;
1524 }
1525
1526 /**
1527 *
1528 * Given an ip_fw *, lookup_next_rule will return a pointer
1529 * to the next rule, which can be either the jump
1530 * target (for skipto instructions) or the next one in the list (in
1531 * all other cases including a missing jump target).
1532 * The result is also written in the "next_rule" field of the rule.
1533 * Backward jumps are not allowed, so start looking from the next
1534 * rule...
1535 *
1536 * This never returns NULL -- in case we do not have an exact match,
1537 * the next rule is returned. When the ruleset is changed,
1538 * pointers are flushed so we are always correct.
1539 */
1540
1541 static struct ip_fw *
1542 lookup_next_rule(struct ip_fw *me)
1543 {
1544 struct ip_fw *rule = NULL;
1545 ipfw_insn *cmd;
1546
1547 /* look for action, in case it is a skipto */
1548 cmd = ACTION_PTR(me);
1549 if (cmd->opcode == O_LOG)
1550 cmd += F_LEN(cmd);
1551 if (cmd->opcode == O_SKIPTO) {
1552 for (rule = me->next; rule; rule = rule->next) {
1553 if (rule->rulenum >= cmd->arg1)
1554 break;
1555 }
1556 }
1557 if (rule == NULL) /* failure or not a skipto */
1558 rule = me->next;
1559 me->next_rule = rule;
1560 return rule;
1561 }
1562
1563 static int
1564 _ipfw_match_uid(const struct ipfw_flow_id *fid, struct ifnet *oif,
1565 enum ipfw_opcodes opcode, uid_t uid)
1566 {
1567 struct in_addr src_ip, dst_ip;
1568 struct inpcbinfo *pi;
1569 boolean_t wildcard;
1570 struct inpcb *pcb;
1571
1572 if (fid->proto == IPPROTO_TCP) {
1573 wildcard = FALSE;
1574 pi = &tcbinfo[mycpuid];
1575 } else if (fid->proto == IPPROTO_UDP) {
1576 wildcard = TRUE;
1577 pi = &udbinfo[mycpuid];
1578 } else {
1579 return 0;
1580 }
1581
1582 /*
1583 * Values in 'fid' are in host byte order
1584 */
1585 dst_ip.s_addr = htonl(fid->dst_ip);
1586 src_ip.s_addr = htonl(fid->src_ip);
1587 if (oif) {
1588 pcb = in_pcblookup_hash(pi,
1589 dst_ip, htons(fid->dst_port),
1590 src_ip, htons(fid->src_port),
1591 wildcard, oif);
1592 } else {
1593 pcb = in_pcblookup_hash(pi,
1594 src_ip, htons(fid->src_port),
1595 dst_ip, htons(fid->dst_port),
1596 wildcard, NULL);
1597 }
1598 if (pcb == NULL || pcb->inp_socket == NULL)
1599 return 0;
1600
1601 if (opcode == O_UID) {
1602 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1603 return !socheckuid(pcb->inp_socket, uid);
1604 #undef socheckuid
1605 } else {
1606 return groupmember(uid, pcb->inp_socket->so_cred);
1607 }
1608 }
1609
1610 static int
1611 ipfw_match_uid(const struct ipfw_flow_id *fid, struct ifnet *oif,
1612 enum ipfw_opcodes opcode, uid_t uid, int *deny)
1613 {
1614 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1615 uint32_t gen;
1616 int match = 0;
1617
1618 *deny = 0;
1619 gen = ctx->ipfw_gen;
1620
1621 if (gen != ctx->ipfw_gen) {
1622 /* See the comment in lookup_rule() */
1623 *deny = 1;
1624 } else {
1625 match = _ipfw_match_uid(fid, oif, opcode, uid);
1626 }
1627 return match;
1628 }
1629
1630 /*
1631 * The main check routine for the firewall.
1632 *
1633 * All arguments are in args so we can modify them and return them
1634 * back to the caller.
1635 *
1636 * Parameters:
1637 *
1638 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
1639 * Starts with the IP header.
1640 * args->eh (in) Mac header if present, or NULL for layer3 packet.
1641 * args->oif Outgoing interface, or NULL if packet is incoming.
1642 * The incoming interface is in the mbuf. (in)
1643 *
1644 * args->rule Pointer to the last matching rule (in/out)
1645 * args->f_id Addresses grabbed from the packet (out)
1646 *
1647 * Return value:
1648 *
1649 * If the packet was denied/rejected and has been dropped, *m is equal
1650 * to NULL upon return.
1651 *
1652 * IP_FW_DENY the packet must be dropped.
1653 * IP_FW_PASS The packet is to be accepted and routed normally.
1654 * IP_FW_DIVERT Divert the packet to port (args->cookie)
1655 * IP_FW_TEE Tee the packet to port (args->cookie)
1656 * IP_FW_DUMMYNET Send the packet to pipe/queue (args->cookie)
1657 */
1658
1659 static int
1660 ipfw_chk(struct ip_fw_args *args)
1661 {
1662 /*
1663 * Local variables hold state during the processing of a packet.
1664 *
1665 * IMPORTANT NOTE: to speed up the processing of rules, there
1666 * are some assumption on the values of the variables, which
1667 * are documented here. Should you change them, please check
1668 * the implementation of the various instructions to make sure
1669 * that they still work.
1670 *
1671 * args->eh The MAC header. It is non-null for a layer2
1672 * packet, it is NULL for a layer-3 packet.
1673 *
1674 * m | args->m Pointer to the mbuf, as received from the caller.
1675 * It may change if ipfw_chk() does an m_pullup, or if it
1676 * consumes the packet because it calls send_reject().
1677 * XXX This has to change, so that ipfw_chk() never modifies
1678 * or consumes the buffer.
1679 * ip is simply an alias of the value of m, and it is kept
1680 * in sync with it (the packet is supposed to start with
1681 * the ip header).
1682 */
1683 struct mbuf *m = args->m;
1684 struct ip *ip = mtod(m, struct ip *);
1685
1686 /*
1687 * oif | args->oif If NULL, ipfw_chk has been called on the
1688 * inbound path (ether_input, ip_input).
1689 * If non-NULL, ipfw_chk has been called on the outbound path
1690 * (ether_output, ip_output).
1691 */
1692 struct ifnet *oif = args->oif;
1693
1694 struct ip_fw *f = NULL; /* matching rule */
1695 int retval = IP_FW_PASS;
1696 struct m_tag *mtag;
1697 struct divert_info *divinfo;
1698
1699 /*
1700 * hlen The length of the IPv4 header.
1701 * hlen >0 means we have an IPv4 packet.
1702 */
1703 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
1704
1705 /*
1706 * offset The offset of a fragment. offset != 0 means that
1707 * we have a fragment at this offset of an IPv4 packet.
1708 * offset == 0 means that (if this is an IPv4 packet)
1709 * this is the first or only fragment.
1710 */
1711 u_short offset = 0;
1712
1713 /*
1714 * Local copies of addresses. They are only valid if we have
1715 * an IP packet.
1716 *
1717 * proto The protocol. Set to 0 for non-ip packets,
1718 * or to the protocol read from the packet otherwise.
1719 * proto != 0 means that we have an IPv4 packet.
1720 *
1721 * src_port, dst_port port numbers, in HOST format. Only
1722 * valid for TCP and UDP packets.
1723 *
1724 * src_ip, dst_ip ip addresses, in NETWORK format.
1725 * Only valid for IPv4 packets.
1726 */
1727 uint8_t proto;
1728 uint16_t src_port = 0, dst_port = 0; /* NOTE: host format */
1729 struct in_addr src_ip, dst_ip; /* NOTE: network format */
1730 uint16_t ip_len = 0;
1731
1732 /*
1733 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1734 * MATCH_NONE when checked and not matched (dyn_f = NULL),
1735 * MATCH_FORWARD or MATCH_REVERSE otherwise (dyn_f != NULL)
1736 */
1737 int dyn_dir = MATCH_UNKNOWN;
1738 struct ip_fw *dyn_f = NULL;
1739 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1740
1741 if (m->m_pkthdr.fw_flags & IPFW_MBUF_GENERATED)
1742 return IP_FW_PASS; /* accept */
1743
1744 if (args->eh == NULL || /* layer 3 packet */
1745 (m->m_pkthdr.len >= sizeof(struct ip) &&
1746 ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1747 hlen = ip->ip_hl << 2;
1748
1749 /*
1750 * Collect parameters into local variables for faster matching.
1751 */
1752 if (hlen == 0) { /* do not grab addresses for non-ip pkts */
1753 proto = args->f_id.proto = 0; /* mark f_id invalid */
1754 goto after_ip_checks;
1755 }
1756
1757 proto = args->f_id.proto = ip->ip_p;
1758 src_ip = ip->ip_src;
1759 dst_ip = ip->ip_dst;
1760 if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1761 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1762 ip_len = ntohs(ip->ip_len);
1763 } else {
1764 offset = ip->ip_off & IP_OFFMASK;
1765 ip_len = ip->ip_len;
1766 }
1767
1768 #define PULLUP_TO(len) \
1769 do { \
1770 if (m->m_len < (len)) { \
1771 args->m = m = m_pullup(m, (len));\
1772 if (m == NULL) \
1773 goto pullup_failed; \
1774 ip = mtod(m, struct ip *); \
1775 } \
1776 } while (0)
1777
1778 if (offset == 0) {
1779 switch (proto) {
1780 case IPPROTO_TCP:
1781 {
1782 struct tcphdr *tcp;
1783
1784 PULLUP_TO(hlen + sizeof(struct tcphdr));
1785 tcp = L3HDR(struct tcphdr, ip);
1786 dst_port = tcp->th_dport;
1787 src_port = tcp->th_sport;
1788 args->f_id.flags = tcp->th_flags;
1789 }
1790 break;
1791
1792 case IPPROTO_UDP:
1793 {
1794 struct udphdr *udp;
1795
1796 PULLUP_TO(hlen + sizeof(struct udphdr));
1797 udp = L3HDR(struct udphdr, ip);
1798 dst_port = udp->uh_dport;
1799 src_port = udp->uh_sport;
1800 }
1801 break;
1802
1803 case IPPROTO_ICMP:
1804 PULLUP_TO(hlen + 4); /* type, code and checksum. */
1805 args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1806 break;
1807
1808 default:
1809 break;
1810 }
1811 }
1812
1813 #undef PULLUP_TO
1814
1815 args->f_id.src_ip = ntohl(src_ip.s_addr);
1816 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1817 args->f_id.src_port = src_port = ntohs(src_port);
1818 args->f_id.dst_port = dst_port = ntohs(dst_port);
1819
1820 after_ip_checks:
1821 if (args->rule) {
1822 /*
1823 * Packet has already been tagged. Look for the next rule
1824 * to restart processing.
1825 *
1826 * If fw_one_pass != 0 then just accept it.
1827 * XXX should not happen here, but optimized out in
1828 * the caller.
1829 */
1830 if (fw_one_pass)
1831 return IP_FW_PASS;
1832
1833 /* This rule is being/has been flushed */
1834 if (ipfw_flushing)
1835 return IP_FW_DENY;
1836
1837 KASSERT(args->rule->cpuid == mycpuid,
1838 ("rule used on cpu%d", mycpuid));
1839
1840 /* This rule was deleted */
1841 if (args->rule->rule_flags & IPFW_RULE_F_INVALID)
1842 return IP_FW_DENY;
1843
1844 f = args->rule->next_rule;
1845 if (f == NULL)
1846 f = lookup_next_rule(args->rule);
1847 } else {
1848 /*
1849 * Find the starting rule. It can be either the first
1850 * one, or the one after divert_rule if asked so.
1851 */
1852 int skipto;
1853
1854 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1855 if (mtag != NULL) {
1856 divinfo = m_tag_data(mtag);
1857 skipto = divinfo->skipto;
1858 } else {
1859 skipto = 0;
1860 }
1861
1862 f = ctx->ipfw_layer3_chain;
1863 if (args->eh == NULL && skipto != 0) {
1864 /* No skipto during rule flushing */
1865 if (ipfw_flushing)
1866 return IP_FW_DENY;
1867
1868 if (skipto >= IPFW_DEFAULT_RULE)
1869 return IP_FW_DENY; /* invalid */
1870
1871 while (f && f->rulenum <= skipto)
1872 f = f->next;
1873 if (f == NULL) /* drop packet */
1874 return IP_FW_DENY;
1875 } else if (ipfw_flushing) {
1876 /* Rules are being flushed; skip to default rule */
1877 f = ctx->ipfw_default_rule;
1878 }
1879 }
1880 if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
1881 m_tag_delete(m, mtag);
1882
1883 /*
1884 * Now scan the rules, and parse microinstructions for each rule.
1885 */
1886 for (; f; f = f->next) {
1887 int l, cmdlen;
1888 ipfw_insn *cmd;
1889 int skip_or; /* skip rest of OR block */
1890
1891 again:
1892 if (ctx->ipfw_set_disable & (1 << f->set))
1893 continue;
1894
1895 skip_or = 0;
1896 for (l = f->cmd_len, cmd = f->cmd; l > 0;
1897 l -= cmdlen, cmd += cmdlen) {
1898 int match, deny;
1899
1900 /*
1901 * check_body is a jump target used when we find a
1902 * CHECK_STATE, and need to jump to the body of
1903 * the target rule.
1904 */
1905
1906 check_body:
1907 cmdlen = F_LEN(cmd);
1908 /*
1909 * An OR block (insn_1 || .. || insn_n) has the
1910 * F_OR bit set in all but the last instruction.
1911 * The first match will set "skip_or", and cause
1912 * the following instructions to be skipped until
1913 * past the one with the F_OR bit clear.
1914 */
1915 if (skip_or) { /* skip this instruction */
1916 if ((cmd->len & F_OR) == 0)
1917 skip_or = 0; /* next one is good */
1918 continue;
1919 }
1920 match = 0; /* set to 1 if we succeed */
1921
1922 switch (cmd->opcode) {
1923 /*
1924 * The first set of opcodes compares the packet's
1925 * fields with some pattern, setting 'match' if a
1926 * match is found. At the end of the loop there is
1927 * logic to deal with F_NOT and F_OR flags associated
1928 * with the opcode.
1929 */
1930 case O_NOP:
1931 match = 1;
1932 break;
1933
1934 case O_FORWARD_MAC:
1935 kprintf("ipfw: opcode %d unimplemented\n",
1936 cmd->opcode);
1937 break;
1938
1939 case O_GID:
1940 case O_UID:
1941 /*
1942 * We only check offset == 0 && proto != 0,
1943 * as this ensures that we have an IPv4
1944 * packet with the ports info.
1945 */
1946 if (offset!=0)
1947 break;
1948
1949 match = ipfw_match_uid(&args->f_id, oif,
1950 cmd->opcode,
1951 (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1952 &deny);
1953 if (deny)
1954 return IP_FW_DENY;
1955 break;
1956
1957 case O_RECV:
1958 match = iface_match(m->m_pkthdr.rcvif,
1959 (ipfw_insn_if *)cmd);
1960 break;
1961
1962 case O_XMIT:
1963 match = iface_match(oif, (ipfw_insn_if *)cmd);
1964 break;
1965
1966 case O_VIA:
1967 match = iface_match(oif ? oif :
1968 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1969 break;
1970
1971 case O_MACADDR2:
1972 if (args->eh != NULL) { /* have MAC header */
1973 uint32_t *want = (uint32_t *)
1974 ((ipfw_insn_mac *)cmd)->addr;
1975 uint32_t *mask = (uint32_t *)
1976 ((ipfw_insn_mac *)cmd)->mask;
1977 uint32_t *hdr = (uint32_t *)args->eh;
1978
1979 match =
1980 (want[0] == (hdr[0] & mask[0]) &&
1981 want[1] == (hdr[1] & mask[1]) &&
1982 want[2] == (hdr[2] & mask[2]));
1983 }
1984 break;
1985
1986 case O_MAC_TYPE:
1987 if (args->eh != NULL) {
1988 uint16_t t =
1989 ntohs(args->eh->ether_type);
1990 uint16_t *p =
1991 ((ipfw_insn_u16 *)cmd)->ports;
1992 int i;
1993
1994 /* Special vlan handling */
1995 if (m->m_flags & M_VLANTAG)
1996 t = ETHERTYPE_VLAN;
1997
1998 for (i = cmdlen - 1; !match && i > 0;
1999 i--, p += 2) {
2000 match =
2001 (t >= p[0] && t <= p[1]);
2002 }
2003 }
2004 break;
2005
2006 case O_FRAG:
2007 match = (hlen > 0 && offset != 0);
2008 break;
2009
2010 case O_IN: /* "out" is "not in" */
2011 match = (oif == NULL);
2012 break;
2013
2014 case O_LAYER2:
2015 match = (args->eh != NULL);
2016 break;
2017
2018 case O_PROTO:
2019 /*
2020 * We do not allow an arg of 0 so the
2021 * check of "proto" only suffices.
2022 */
2023 match = (proto == cmd->arg1);
2024 break;
2025
2026 case O_IP_SRC:
2027 match = (hlen > 0 &&
2028 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2029 src_ip.s_addr);
2030 break;
2031
2032 case O_IP_SRC_MASK:
2033 match = (hlen > 0 &&
2034 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2035 (src_ip.s_addr &
2036 ((ipfw_insn_ip *)cmd)->mask.s_addr));
2037 break;
2038
2039 case O_IP_SRC_ME:
2040 if (hlen > 0) {
2041 struct ifnet *tif;
2042
2043 tif = INADDR_TO_IFP(&src_ip);
2044 match = (tif != NULL);
2045 }
2046 break;
2047
2048 case O_IP_DST_SET:
2049 case O_IP_SRC_SET:
2050 if (hlen > 0) {
2051 uint32_t *d = (uint32_t *)(cmd + 1);
2052 uint32_t addr =
2053 cmd->opcode == O_IP_DST_SET ?
2054 args->f_id.dst_ip :
2055 args->f_id.src_ip;
2056
2057 if (addr < d[0])
2058 break;
2059 addr -= d[0]; /* subtract base */
2060 match =
2061 (addr < cmd->arg1) &&
2062 (d[1 + (addr >> 5)] &
2063 (1 << (addr & 0x1f)));
2064 }
2065 break;
2066
2067 case O_IP_DST:
2068 match = (hlen > 0 &&
2069 ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2070 dst_ip.s_addr);
2071 break;
2072
2073 case O_IP_DST_MASK:
2074 match = (hlen > 0) &&
2075 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2076 (dst_ip.s_addr &
2077 ((ipfw_insn_ip *)cmd)->mask.s_addr));
2078 break;
2079
2080 case O_IP_DST_ME:
2081 if (hlen > 0) {
2082 struct ifnet *tif;
2083
2084 tif = INADDR_TO_IFP(&dst_ip);
2085 match = (tif != NULL);
2086 }
2087 break;
2088
2089 case O_IP_SRCPORT:
2090 case O_IP_DSTPORT:
2091 /*
2092 * offset == 0 && proto != 0 is enough
2093 * to guarantee that we have an IPv4
2094 * packet with port info.
2095 */
2096 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2097 && offset == 0) {
2098 uint16_t x =
2099 (cmd->opcode == O_IP_SRCPORT) ?
2100 src_port : dst_port ;
2101 uint16_t *p =
2102 ((ipfw_insn_u16 *)cmd)->ports;
2103 int i;
2104
2105 for (i = cmdlen - 1; !match && i > 0;
2106 i--, p += 2) {
2107 match =
2108 (x >= p[0] && x <= p[1]);
2109 }
2110 }
2111 break;
2112
2113 case O_ICMPTYPE:
2114 match = (offset == 0 && proto==IPPROTO_ICMP &&
2115 icmptype_match(ip, (ipfw_insn_u32 *)cmd));
2116 break;
2117
2118 case O_IPOPT:
2119 match = (hlen > 0 && ipopts_match(ip, cmd));
2120 break;
2121
2122 case O_IPVER:
2123 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
2124 break;
2125
2126 case O_IPTTL:
2127 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl);
2128 break;
2129
2130 case O_IPID:
2131 match = (hlen > 0 &&
2132 cmd->arg1 == ntohs(ip->ip_id));
2133 break;
2134
2135 case O_IPLEN:
2136 match = (hlen > 0 && cmd->arg1 == ip_len);
2137 break;
2138
2139 case O_IPPRECEDENCE:
2140 match = (hlen > 0 &&
2141 (cmd->arg1 == (ip->ip_tos & 0xe0)));
2142 break;
2143
2144 case O_IPTOS:
2145 match = (hlen > 0 &&
2146 flags_match(cmd, ip->ip_tos));
2147 break;
2148
2149 case O_TCPFLAGS:
2150 match = (proto == IPPROTO_TCP && offset == 0 &&
2151 flags_match(cmd,
2152 L3HDR(struct tcphdr,ip)->th_flags));
2153 break;
2154
2155 case O_TCPOPTS:
2156 match = (proto == IPPROTO_TCP && offset == 0 &&
2157 tcpopts_match(ip, cmd));
2158 break;
2159
2160 case O_TCPSEQ:
2161 match = (proto == IPPROTO_TCP && offset == 0 &&
2162 ((ipfw_insn_u32 *)cmd)->d[0] ==
2163 L3HDR(struct tcphdr,ip)->th_seq);
2164 break;
2165
2166 case O_TCPACK:
2167 match = (proto == IPPROTO_TCP && offset == 0 &&
2168 ((ipfw_insn_u32 *)cmd)->d[0] ==
2169 L3HDR(struct tcphdr,ip)->th_ack);
2170 break;
2171
2172 case O_TCPWIN:
2173 match = (proto == IPPROTO_TCP && offset == 0 &&
2174 cmd->arg1 ==
2175 L3HDR(struct tcphdr,ip)->th_win);
2176 break;
2177
2178 case O_ESTAB:
2179 /* reject packets which have SYN only */
2180 /* XXX should i also check for TH_ACK ? */
2181 match = (proto == IPPROTO_TCP && offset == 0 &&
2182 (L3HDR(struct tcphdr,ip)->th_flags &
2183 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2184 break;
2185
2186 case O_LOG:
2187 if (fw_verbose)
2188 ipfw_log(f, hlen, args->eh, m, oif);
2189 match = 1;
2190 break;
2191
2192 case O_PROB:
2193 match = (krandom() <
2194 ((ipfw_insn_u32 *)cmd)->d[0]);
2195 break;
2196
2197 /*
2198 * The second set of opcodes represents 'actions',
2199 * i.e. the terminal part of a rule once the packet
2200 * matches all previous patterns.
2201 * Typically there is only one action for each rule,
2202 * and the opcode is stored at the end of the rule
2203 * (but there are exceptions -- see below).
2204 *
2205 * In general, here we set retval and terminate the
2206 * outer loop (would be a 'break 3' in some language,
2207 * but we need to do a 'goto done').
2208 *
2209 * Exceptions:
2210 * O_COUNT and O_SKIPTO actions:
2211 * instead of terminating, we jump to the next rule
2212 * ('goto next_rule', equivalent to a 'break 2'),
2213 * or to the SKIPTO target ('goto again' after
2214 * having set f, cmd and l), respectively.
2215 *
2216 * O_LIMIT and O_KEEP_STATE: these opcodes are
2217 * not real 'actions', and are stored right
2218 * before the 'action' part of the rule.
2219 * These opcodes try to install an entry in the
2220 * state tables; if successful, we continue with
2221 * the next opcode (match=1; break;), otherwise
2222 * the packet must be dropped ('goto done' after
2223 * setting retval). If static rules are changed
2224 * during the state installation, the packet will
2225 * be dropped and rule's stats will not beupdated
2226 * ('return IP_FW_DENY').
2227 *
2228 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
2229 * cause a lookup of the state table, and a jump
2230 * to the 'action' part of the parent rule
2231 * ('goto check_body') if an entry is found, or
2232 * (CHECK_STATE only) a jump to the next rule if
2233 * the entry is not found ('goto next_rule').
2234 * The result of the lookup is cached to make
2235 * further instances of these opcodes are
2236 * effectively NOPs. If static rules are changed
2237 * during the state looking up, the packet will
2238 * be dropped and rule's stats will not be updated
2239 * ('return IP_FW_DENY').
2240 */
2241 case O_LIMIT:
2242 case O_KEEP_STATE:
2243 if (!(f->rule_flags & IPFW_RULE_F_STATE)) {
2244 kprintf("%s rule (%d) is not ready "
2245 "on cpu%d\n",
2246 cmd->opcode == O_LIMIT ?
2247 "limit" : "keep state",
2248 f->rulenum, f->cpuid);
2249 goto next_rule;
2250 }
2251 if (install_state(f,
2252 (ipfw_insn_limit *)cmd, args, &deny)) {
2253 if (deny)
2254 return IP_FW_DENY;
2255
2256 retval = IP_FW_DENY;
2257 goto done; /* error/limit violation */
2258 }
2259 if (deny)
2260 return IP_FW_DENY;
2261 match = 1;
2262 break;
2263
2264 case O_PROBE_STATE:
2265 case O_CHECK_STATE:
2266 /*
2267 * dynamic rules are checked at the first
2268 * keep-state or check-state occurrence,
2269 * with the result being stored in dyn_dir.
2270 * The compiler introduces a PROBE_STATE
2271 * instruction for us when we have a
2272 * KEEP_STATE (because PROBE_STATE needs
2273 * to be run first).
2274 */
2275 if (dyn_dir == MATCH_UNKNOWN) {
2276 dyn_f = lookup_rule(&args->f_id,
2277 &dyn_dir,
2278 proto == IPPROTO_TCP ?
2279 L3HDR(struct tcphdr, ip) : NULL,
2280 ip_len, &deny);
2281 if (deny)
2282 return IP_FW_DENY;
2283 if (dyn_f != NULL) {
2284 /*
2285 * Found a rule from a dynamic
2286 * entry; jump to the 'action'
2287 * part of the rule.
2288 */
2289 f = dyn_f;
2290 cmd = ACTION_PTR(f);
2291 l = f->cmd_len - f->act_ofs;
2292 goto check_body;
2293 }
2294 }
2295 /*
2296 * Dynamic entry not found. If CHECK_STATE,
2297 * skip to next rule, if PROBE_STATE just
2298 * ignore and continue with next opcode.
2299 */
2300 if (cmd->opcode == O_CHECK_STATE)
2301 goto next_rule;
2302 else if (!(f->rule_flags & IPFW_RULE_F_STATE))
2303 goto next_rule; /* not ready yet */
2304 match = 1;
2305 break;
2306
2307 case O_ACCEPT:
2308 retval = IP_FW_PASS; /* accept */
2309 goto done;
2310
2311 case O_PIPE:
2312 case O_QUEUE:
2313 args->rule = f; /* report matching rule */
2314 args->cookie = cmd->arg1;
2315 retval = IP_FW_DUMMYNET;
2316 goto done;
2317
2318 case O_DIVERT:
2319 case O_TEE:
2320 if (args->eh) /* not on layer 2 */
2321 break;
2322
2323 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT,
2324 sizeof(*divinfo), M_NOWAIT);
2325 if (mtag == NULL) {
2326 retval = IP_FW_DENY;
2327 goto done;
2328 }
2329 divinfo = m_tag_data(mtag);
2330
2331 divinfo->skipto = f->rulenum;
2332 divinfo->port = cmd->arg1;
2333 divinfo->tee = (cmd->opcode == O_TEE);
2334 m_tag_prepend(m, mtag);
2335
2336 args->cookie = cmd->arg1;
2337 retval = (cmd->opcode == O_DIVERT) ?
2338 IP_FW_DIVERT : IP_FW_TEE;
2339 goto done;
2340
2341 case O_COUNT:
2342 case O_SKIPTO:
2343 f->pcnt++; /* update stats */
2344 f->bcnt += ip_len;
2345 f->timestamp = time_second;
2346 if (cmd->opcode == O_COUNT)
2347 goto next_rule;
2348 /* handle skipto */
2349 if (f->next_rule == NULL)
2350 lookup_next_rule(f);
2351 f = f->next_rule;
2352 goto again;
2353
2354 case O_REJECT:
2355 /*
2356 * Drop the packet and send a reject notice
2357 * if the packet is not ICMP (or is an ICMP
2358 * query), and it is not multicast/broadcast.
2359 */
2360 if (hlen > 0 &&
2361 (proto != IPPROTO_ICMP ||
2362 is_icmp_query(ip)) &&
2363 !(m->m_flags & (M_BCAST|M_MCAST)) &&
2364 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
2365 /*
2366 * Update statistics before the possible
2367 * blocking 'send_reject'
2368 */
2369 f->pcnt++;
2370 f->bcnt += ip_len;
2371 f->timestamp = time_second;
2372
2373 send_reject(args, cmd->arg1,
2374 offset,ip_len);
2375 m = args->m;
2376
2377 /*
2378 * Return directly here, rule stats
2379 * have been updated above.
2380 */
2381 return IP_FW_DENY;
2382 }
2383 /* FALLTHROUGH */
2384 case O_DENY:
2385 retval = IP_FW_DENY;
2386 goto done;
2387
2388 case O_FORWARD_IP:
2389 if (args->eh) /* not valid on layer2 pkts */
2390 break;
2391 if (!dyn_f || dyn_dir == MATCH_FORWARD) {
2392 struct sockaddr_in *sin;
2393
2394 mtag = m_tag_get(PACKET_TAG_IPFORWARD,
2395 sizeof(*sin), M_NOWAIT);
2396 if (mtag == NULL) {
2397 retval = IP_FW_DENY;
2398 goto done;
2399 }
2400 sin = m_tag_data(mtag);
2401
2402 /* Structure copy */
2403 *sin = ((ipfw_insn_sa *)cmd)->sa;
2404
2405 m_tag_prepend(m, mtag);
2406 m->m_pkthdr.fw_flags |=
2407 IPFORWARD_MBUF_TAGGED;
2408 m->m_pkthdr.fw_flags &=
2409 ~BRIDGE_MBUF_TAGGED;
2410 }
2411 retval = IP_FW_PASS;
2412 goto done;
2413
2414 default:
2415 panic("-- unknown opcode %d", cmd->opcode);
2416 } /* end of switch() on opcodes */
2417
2418 if (cmd->len & F_NOT)
2419 match = !match;
2420
2421 if (match) {
2422 if (cmd->len & F_OR)
2423 skip_or = 1;
2424 } else {
2425 if (!(cmd->len & F_OR)) /* not an OR block, */
2426 break; /* try next rule */
2427 }
2428
2429 } /* end of inner for, scan opcodes */
2430
2431 next_rule:; /* try next rule */
2432
2433 } /* end of outer for, scan rules */
2434 kprintf("+++ ipfw: ouch!, skip past end of rules, denying packet\n");
2435 return IP_FW_DENY;
2436
2437 done:
2438 /* Update statistics */
2439 f->pcnt++;
2440 f->bcnt += ip_len;
2441 f->timestamp = time_second;
2442 return retval;
2443
2444 pullup_failed:
2445 if (fw_verbose)
2446 kprintf("pullup failed\n");
2447 return IP_FW_DENY;
2448 }
2449
2450 static void
2451 ipfw_dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa)
2452 {
2453 struct m_tag *mtag;
2454 struct dn_pkt *pkt;
2455 ipfw_insn *cmd;
2456 const struct ipfw_flow_id *id;
2457 struct dn_flow_id *fid;
2458
2459 M_ASSERTPKTHDR(m);
2460
2461 mtag = m_tag_get(PACKET_TAG_DUMMYNET, sizeof(*pkt), M_NOWAIT);
2462 if (mtag == NULL) {
2463 m_freem(m);
2464 return;
2465 }
2466 m_tag_prepend(m, mtag);
2467
2468 pkt = m_tag_data(mtag);
2469 bzero(pkt, sizeof(*pkt));
2470
2471 cmd = fwa->rule->cmd + fwa->rule->act_ofs;
2472 if (cmd->opcode == O_LOG)
2473 cmd += F_LEN(cmd);
2474 KASSERT(cmd->opcode == O_PIPE || cmd->opcode == O_QUEUE,
2475 ("Rule is not PIPE or QUEUE, opcode %d", cmd->opcode));
2476
2477 pkt->dn_m = m;
2478 pkt->dn_flags = (dir & DN_FLAGS_DIR_MASK);
2479 pkt->ifp = fwa->oif;
2480 pkt->pipe_nr = pipe_nr;
2481
2482 pkt->cpuid = mycpuid;
2483 pkt->msgport = netisr_curport();
2484
2485 id = &fwa->f_id;
2486 fid = &pkt->id;
2487 fid->fid_dst_ip = id->dst_ip;
2488 fid->fid_src_ip = id->src_ip;
2489 fid->fid_dst_port = id->dst_port;
2490 fid->fid_src_port = id->src_port;
2491 fid->fid_proto = id->proto;
2492 fid->fid_flags = id->flags;
2493
2494 ipfw_ref_rule(fwa->rule);
2495 pkt->dn_priv = fwa->rule;
2496 pkt->dn_unref_priv = ipfw_unref_rule;
2497
2498 if (cmd->opcode == O_PIPE)
2499 pkt->dn_flags |= DN_FLAGS_IS_PIPE;
2500
2501 m->m_pkthdr.fw_flags |= DUMMYNET_MBUF_TAGGED;
2502 }
2503
2504 /*
2505 * When a rule is added/deleted, clear the next_rule pointers in all rules.
2506 * These will be reconstructed on the fly as packets are matched.
2507 * Must be called at splimp().
2508 */
2509 static void
2510 ipfw_flush_rule_ptrs(struct ipfw_context *ctx)
2511 {
2512 struct ip_fw *rule;
2513
2514 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
2515 rule->next_rule = NULL;
2516 }
2517
2518 static __inline void
2519 ipfw_inc_static_count(struct ip_fw *rule)
2520 {
2521 /* Static rule's counts are updated only on CPU0 */
2522 KKASSERT(mycpuid == 0);
2523
2524 static_count++;
2525 static_ioc_len += IOC_RULESIZE(rule);
2526 }
2527
2528 static __inline void
2529 ipfw_dec_static_count(struct ip_fw *rule)
2530 {
2531 int l = IOC_RULESIZE(rule);
2532
2533 /* Static rule's counts are updated only on CPU0 */
2534 KKASSERT(mycpuid == 0);
2535
2536 KASSERT(static_count > 0, ("invalid static count %u", static_count));
2537 static_count--;
2538
2539 KASSERT(static_ioc_len >= l,
2540 ("invalid static len %u", static_ioc_len));
2541 static_ioc_len -= l;
2542 }
2543
2544 static void
2545 ipfw_link_sibling(struct netmsg_ipfw *fwmsg, struct ip_fw *rule)
2546 {
2547 if (fwmsg->sibling != NULL) {
2548 KKASSERT(mycpuid > 0 && fwmsg->sibling->cpuid == mycpuid - 1);
2549 fwmsg->sibling->sibling = rule;
2550 }
2551 fwmsg->sibling = rule;
2552 }
2553
2554 static struct ip_fw *
2555 ipfw_create_rule(const struct ipfw_ioc_rule *ioc_rule, struct ip_fw_stub *stub)
2556 {
2557 struct ip_fw *rule;
2558
2559 rule = kmalloc(RULESIZE(ioc_rule), M_IPFW, M_WAITOK | M_ZERO);
2560
2561 rule->act_ofs = ioc_rule->act_ofs;
2562 rule->cmd_len = ioc_rule->cmd_len;
2563 rule->rulenum = ioc_rule->rulenum;
2564 rule->set = ioc_rule->set;
2565 rule->usr_flags = ioc_rule->usr_flags;
2566
2567 bcopy(ioc_rule->cmd, rule->cmd, rule->cmd_len * 4 /* XXX */);
2568
2569 rule->refcnt = 1;
2570 rule->cpuid = mycpuid;
2571
2572 rule->stub = stub;
2573 if (stub != NULL)
2574 stub->rule[mycpuid] = rule;
2575
2576 return rule;
2577 }
2578
2579 static void
2580 ipfw_add_rule_dispatch(netmsg_t nmsg)
2581 {
2582 struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg;
2583 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2584 struct ip_fw *rule;
2585
2586 rule = ipfw_create_rule(fwmsg->ioc_rule, fwmsg->stub);
2587
2588 /*
2589 * Bump generation after ipfw_create_rule(),
2590 * since this function is blocking
2591 */
2592 ctx->ipfw_gen++;
2593
2594 /*
2595 * Insert rule into the pre-determined position
2596 */
2597 if (fwmsg->prev_rule != NULL) {
2598 struct ip_fw *prev, *next;
2599
2600 prev = fwmsg->prev_rule;
2601 KKASSERT(prev->cpuid == mycpuid);
2602
2603 next = fwmsg->next_rule;
2604 KKASSERT(next->cpuid == mycpuid);
2605
2606 rule->next = next;
2607 prev->next = rule;
2608
2609 /*
2610 * Move to the position on the next CPU
2611 * before the msg is forwarded.
2612 */
2613 fwmsg->prev_rule = prev->sibling;
2614 fwmsg->next_rule = next->sibling;
2615 } else {
2616 KKASSERT(fwmsg->next_rule == NULL);
2617 rule->next = ctx->ipfw_layer3_chain;
2618 ctx->ipfw_layer3_chain = rule;
2619 }
2620
2621 /* Link rule CPU sibling */
2622 ipfw_link_sibling(fwmsg, rule);
2623
2624 ipfw_flush_rule_ptrs(ctx);
2625
2626 if (mycpuid == 0) {
2627 /* Statistics only need to be updated once */
2628 ipfw_inc_static_count(rule);
2629
2630 /* Return the rule on CPU0 */
2631 nmsg->lmsg.u.ms_resultp = rule;
2632 }
2633
2634 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2635 }
2636
2637 static void
2638 ipfw_enable_state_dispatch(netmsg_t nmsg)
2639 {
2640 struct lwkt_msg *lmsg = &nmsg->lmsg;
2641 struct ip_fw *rule = lmsg->u.ms_resultp;
2642 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2643
2644 ctx->ipfw_gen++;
2645
2646 KKASSERT(rule->cpuid == mycpuid);
2647 KKASSERT(rule->stub != NULL && rule->stub->rule[mycpuid] == rule);
2648 KKASSERT(!(rule->rule_flags & IPFW_RULE_F_STATE));
2649 rule->rule_flags |= IPFW_RULE_F_STATE;
2650 lmsg->u.ms_resultp = rule->sibling;
2651
2652 ifnet_forwardmsg(lmsg, mycpuid + 1);
2653 }
2654
2655 /*
2656 * Add a new rule to the list. Copy the rule into a malloc'ed area,
2657 * then possibly create a rule number and add the rule to the list.
2658 * Update the rule_number in the input struct so the caller knows
2659 * it as well.
2660 */
2661 static void
2662 ipfw_add_rule(struct ipfw_ioc_rule *ioc_rule, uint32_t rule_flags)
2663 {
2664 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2665 struct netmsg_ipfw fwmsg;
2666 struct netmsg_base *nmsg;
2667 struct ip_fw *f, *prev, *rule;
2668 struct ip_fw_stub *stub;
2669
2670 IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2671
2672 /*
2673 * If rulenum is 0, find highest numbered rule before the
2674 * default rule, and add rule number incremental step.
2675 */
2676 if (ioc_rule->rulenum == 0) {
2677 int step = autoinc_step;
2678
2679 KKASSERT(step >= IPFW_AUTOINC_STEP_MIN &&
2680 step <= IPFW_AUTOINC_STEP_MAX);
2681
2682 /*
2683 * Locate the highest numbered rule before default
2684 */
2685 for (f = ctx->ipfw_layer3_chain; f; f = f->next) {
2686 if (f->rulenum == IPFW_DEFAULT_RULE)
2687 break;
2688 ioc_rule->rulenum = f->rulenum;
2689 }
2690 if (ioc_rule->rulenum < IPFW_DEFAULT_RULE - step)
2691 ioc_rule->rulenum += step;
2692 }
2693 KASSERT(ioc_rule->rulenum != IPFW_DEFAULT_RULE &&
2694 ioc_rule->rulenum != 0,
2695 ("invalid rule num %d", ioc_rule->rulenum));
2696
2697 /*
2698 * Now find the right place for the new rule in the sorted list.
2699 */
2700 for (prev = NULL, f = ctx->ipfw_layer3_chain; f;
2701 prev = f, f = f->next) {
2702 if (f->rulenum > ioc_rule->rulenum) {
2703 /* Found the location */
2704 break;
2705 }
2706 }
2707 KASSERT(f != NULL, ("no default rule?!"));
2708
2709 if (rule_flags & IPFW_RULE_F_STATE) {
2710 int size;
2711
2712 /*
2713 * If the new rule will create states, then allocate
2714 * a rule stub, which will be referenced by states
2715 * (dyn rules)
2716 */
2717 size = sizeof(*stub) + ((ncpus - 1) * sizeof(struct ip_fw *));
2718 stub = kmalloc(size, M_IPFW, M_WAITOK | M_ZERO);
2719 } else {
2720 stub = NULL;
2721 }
2722
2723 /*
2724 * Duplicate the rule onto each CPU.
2725 * The rule duplicated on CPU0 will be returned.
2726 */
2727 bzero(&fwmsg, sizeof(fwmsg));
2728 nmsg = &fwmsg.base;
2729 netmsg_init(nmsg, NULL, &curthread->td_msgport,
2730 0, ipfw_add_rule_dispatch);
2731 fwmsg.ioc_rule = ioc_rule;
2732 fwmsg.prev_rule = prev;
2733 fwmsg.next_rule = prev == NULL ? NULL : f;
2734 fwmsg.stub = stub;
2735
2736 ifnet_domsg(&nmsg->lmsg, 0);
2737 KKASSERT(fwmsg.prev_rule == NULL && fwmsg.next_rule == NULL);
2738
2739 rule = nmsg->lmsg.u.ms_resultp;
2740 KKASSERT(rule != NULL && rule->cpuid == mycpuid);
2741
2742 if (rule_flags & IPFW_RULE_F_STATE) {
2743 /*
2744 * Turn on state flag, _after_ everything on all
2745 * CPUs have been setup.
2746 */
2747 bzero(nmsg, sizeof(*nmsg));
2748 netmsg_init(nmsg, NULL, &curthread->td_msgport,
2749 0, ipfw_enable_state_dispatch);
2750 nmsg->lmsg.u.ms_resultp = rule;
2751
2752 ifnet_domsg(&nmsg->lmsg, 0);
2753 KKASSERT(nmsg->lmsg.u.ms_resultp == NULL);
2754 }
2755
2756 DPRINTF("++ installed rule %d, static count now %d\n",
2757 rule->rulenum, static_count);
2758 }
2759
2760 /**
2761 * Free storage associated with a static rule (including derived
2762 * dynamic rules).
2763 * The caller is in charge of clearing rule pointers to avoid
2764 * dangling pointers.
2765 * @return a pointer to the next entry.
2766 * Arguments are not checked, so they better be correct.
2767 * Must be called at splimp().
2768 */
2769 static struct ip_fw *
2770 ipfw_delete_rule(struct ipfw_context *ctx,
2771 struct ip_fw *prev, struct ip_fw *rule)
2772 {
2773 struct ip_fw *n;
2774 struct ip_fw_stub *stub;
2775
2776 ctx->ipfw_gen++;
2777
2778 /* STATE flag should have been cleared before we reach here */
2779 KKASSERT((rule->rule_flags & IPFW_RULE_F_STATE) == 0);
2780
2781 stub = rule->stub;
2782 n = rule->next;
2783 if (prev == NULL)
2784 ctx->ipfw_layer3_chain = n;
2785 else
2786 prev->next = n;
2787
2788 /* Mark the rule as invalid */
2789 rule->rule_flags |= IPFW_RULE_F_INVALID;
2790 rule->next_rule = NULL;
2791 rule->sibling = NULL;
2792 rule->stub = NULL;
2793 #ifdef foo
2794 /* Don't reset cpuid here; keep various assertion working */
2795 rule->cpuid = -1;
2796 #endif
2797
2798 /* Statistics only need to be updated once */
2799 if (mycpuid == 0)
2800 ipfw_dec_static_count(rule);
2801
2802 /* Free 'stub' on the last CPU */
2803 if (stub != NULL && mycpuid == ncpus - 1)
2804 kfree(stub, M_IPFW);
2805
2806 /* Try to free this rule */
2807 ipfw_free_rule(rule);
2808
2809 /* Return the next rule */
2810 return n;
2811 }
2812
2813 static void
2814 ipfw_flush_dispatch(netmsg_t nmsg)
2815 {
2816 struct lwkt_msg *lmsg = &nmsg->lmsg;
2817 int kill_default = lmsg->u.ms_result;
2818 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2819 struct ip_fw *rule;
2820
2821 ipfw_flush_rule_ptrs(ctx); /* more efficient to do outside the loop */
2822
2823 while ((rule = ctx->ipfw_layer3_chain) != NULL &&
2824 (kill_default || rule->rulenum != IPFW_DEFAULT_RULE))
2825 ipfw_delete_rule(ctx, NULL, rule);
2826
2827 ifnet_forwardmsg(lmsg, mycpuid + 1);
2828 }
2829
2830 static void
2831 ipfw_disable_rule_state_dispatch(netmsg_t nmsg)
2832 {
2833 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
2834 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2835 struct ip_fw *rule;
2836
2837 ctx->ipfw_gen++;
2838
2839 rule = dmsg->start_rule;
2840 if (rule != NULL) {
2841 KKASSERT(rule->cpuid == mycpuid);
2842
2843 /*
2844 * Move to the position on the next CPU
2845 * before the msg is forwarded.
2846 */
2847 dmsg->start_rule = rule->sibling;
2848 } else {
2849 KKASSERT(dmsg->rulenum == 0);
2850 rule = ctx->ipfw_layer3_chain;
2851 }
2852
2853 while (rule != NULL) {
2854 if (dmsg->rulenum && rule->rulenum != dmsg->rulenum)
2855 break;
2856 rule->rule_flags &= ~IPFW_RULE_F_STATE;
2857 rule = rule->next;
2858 }
2859
2860 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2861 }
2862
2863 /*
2864 * Deletes all rules from a chain (including the default rule
2865 * if the second argument is set).
2866 * Must be called at splimp().
2867 */
2868 static void
2869 ipfw_flush(int kill_default)
2870 {
2871 struct netmsg_del dmsg;
2872 struct netmsg_base nmsg;
2873 struct lwkt_msg *lmsg;
2874 struct ip_fw *rule;
2875 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2876
2877 IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2878
2879 /*
2880 * If 'kill_default' then caller has done the necessary
2881 * msgport syncing; unnecessary to do it again.
2882 */
2883 if (!kill_default) {
2884 /*
2885 * Let ipfw_chk() know the rules are going to
2886 * be flushed, so it could jump directly to
2887 * the default rule.
2888 */
2889 ipfw_flushing = 1;
2890 netmsg_service_sync();
2891 }
2892
2893 /*
2894 * Clear STATE flag on rules, so no more states (dyn rules)
2895 * will be created.
2896 */
2897 bzero(&dmsg, sizeof(dmsg));
2898 netmsg_init(&dmsg.base, NULL, &curthread->td_msgport,
2899 0, ipfw_disable_rule_state_dispatch);
2900 ifnet_domsg(&dmsg.base.lmsg, 0);
2901
2902 /*
2903 * This actually nukes all states (dyn rules)
2904 */
2905 lockmgr(&dyn_lock, LK_EXCLUSIVE);
2906 for (rule = ctx->ipfw_layer3_chain; rule != NULL; rule = rule->next) {
2907 /*
2908 * Can't check IPFW_RULE_F_STATE here,
2909 * since it has been cleared previously.
2910 * Check 'stub' instead.
2911 */
2912 if (rule->stub != NULL) {
2913 /* Force removal */
2914 remove_dyn_rule_locked(rule, NULL);
2915 }
2916 }
2917 lockmgr(&dyn_lock, LK_RELEASE);
2918
2919 /*
2920 * Press the 'flush' button
2921 */
2922 bzero(&nmsg, sizeof(nmsg));
2923 netmsg_init(&nmsg, NULL, &curthread->td_msgport,
2924 0, ipfw_flush_dispatch);
2925 lmsg = &nmsg.lmsg;
2926 lmsg->u.ms_result = kill_default;
2927 ifnet_domsg(lmsg, 0);
2928
2929 KASSERT(dyn_count == 0, ("%u dyn rule remains", dyn_count));
2930
2931 if (kill_default) {
2932 if (ipfw_dyn_v != NULL) {
2933 /*
2934 * Free dynamic rules(state) hash table
2935 */
2936 kfree(ipfw_dyn_v, M_IPFW);
2937 ipfw_dyn_v = NULL;
2938 }
2939
2940 KASSERT(static_count == 0,
2941 ("%u static rules remain", static_count));
2942 KASSERT(static_ioc_len == 0,
2943 ("%u bytes of static rules remain", static_ioc_len));
2944 } else {
2945 KASSERT(static_count == 1,
2946 ("%u static rules remain", static_count));
2947 KASSERT(static_ioc_len == IOC_RULESIZE(ctx->ipfw_default_rule),
2948 ("%u bytes of static rules remain, should be %lu",
2949 static_ioc_len,
2950 (u_long)IOC_RULESIZE(ctx->ipfw_default_rule)));
2951 }
2952
2953 /* Flush is done */
2954 ipfw_flushing = 0;
2955 }
2956
2957 static void
2958 ipfw_alt_delete_rule_dispatch(netmsg_t nmsg)
2959 {
2960 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
2961 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2962 struct ip_fw *rule, *prev;
2963
2964 rule = dmsg->start_rule;
2965 KKASSERT(rule->cpuid == mycpuid);
2966 dmsg->start_rule = rule->sibling;
2967
2968 prev = dmsg->prev_rule;
2969 if (prev != NULL) {
2970 KKASSERT(prev->cpuid == mycpuid);
2971
2972 /*
2973 * Move to the position on the next CPU
2974 * before the msg is forwarded.
2975 */
2976 dmsg->prev_rule = prev->sibling;
2977 }
2978
2979 /*
2980 * flush pointers outside the loop, then delete all matching
2981 * rules. 'prev' remains the same throughout the cycle.
2982 */
2983 ipfw_flush_rule_ptrs(ctx);
2984 while (rule && rule->rulenum == dmsg->rulenum)
2985 rule = ipfw_delete_rule(ctx, prev, rule);
2986
2987 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
2988 }
2989
2990 static int
2991 ipfw_alt_delete_rule(uint16_t rulenum)
2992 {
2993 struct ip_fw *prev, *rule, *f;
2994 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2995 struct netmsg_del dmsg;
2996 struct netmsg_base *nmsg;
2997 int state;
2998
2999 /*
3000 * Locate first rule to delete
3001 */
3002 for (prev = NULL, rule = ctx->ipfw_layer3_chain;
3003 rule && rule->rulenum < rulenum;
3004 prev = rule, rule = rule->next)
3005 ; /* EMPTY */
3006 if (rule->rulenum != rulenum)
3007 return EINVAL;
3008
3009 /*
3010 * Check whether any rules with the given number will
3011 * create states.
3012 */
3013 state = 0;
3014 for (f = rule; f && f->rulenum == rulenum; f = f->next) {
3015 if (f->rule_flags & IPFW_RULE_F_STATE) {
3016 state = 1;
3017 break;
3018 }
3019 }
3020
3021 if (state) {
3022 /*
3023 * Clear the STATE flag, so no more states will be
3024 * created based the rules numbered 'rulenum'.
3025 */
3026 bzero(&dmsg, sizeof(dmsg));
3027 nmsg = &dmsg.base;
3028 netmsg_init(nmsg, NULL, &curthread->td_msgport,
3029 0, ipfw_disable_rule_state_dispatch);
3030 dmsg.start_rule = rule;
3031 dmsg.rulenum = rulenum;
3032
3033 ifnet_domsg(&nmsg->lmsg, 0);
3034 KKASSERT(dmsg.start_rule == NULL);
3035
3036 /*
3037 * Nuke all related states
3038 */
3039 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3040 for (f = rule; f && f->rulenum == rulenum; f = f->next) {
3041 /*
3042 * Can't check IPFW_RULE_F_STATE here,
3043 * since it has been cleared previously.
3044 * Check 'stub' instead.
3045 */
3046 if (f->stub != NULL) {
3047 /* Force removal */
3048 remove_dyn_rule_locked(f, NULL);
3049 }
3050 }
3051 lockmgr(&dyn_lock, LK_RELEASE);
3052 }
3053
3054 /*
3055 * Get rid of the rule duplications on all CPUs
3056 */
3057 bzero(&dmsg, sizeof(dmsg));
3058 nmsg = &dmsg.base;
3059 netmsg_init(nmsg, NULL, &curthread->td_msgport,
3060 0, ipfw_alt_delete_rule_dispatch);
3061 dmsg.prev_rule = prev;
3062 dmsg.start_rule = rule;
3063 dmsg.rulenum = rulenum;
3064
3065 ifnet_domsg(&nmsg->lmsg, 0);
3066 KKASSERT(dmsg.prev_rule == NULL && dmsg.start_rule == NULL);
3067 return 0;
3068 }
3069
3070 static void
3071 ipfw_alt_delete_ruleset_dispatch(netmsg_t nmsg)
3072 {
3073 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3074 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3075 struct ip_fw *prev, *rule;
3076 #ifdef INVARIANTS
3077 int del = 0;
3078 #endif
3079
3080 ipfw_flush_rule_ptrs(ctx);
3081
3082 prev = NULL;
3083 rule = ctx->ipfw_layer3_chain;
3084 while (rule != NULL) {
3085 if (rule->set == dmsg->from_set) {
3086 rule = ipfw_delete_rule(ctx, prev, rule);
3087 #ifdef INVARIANTS
3088 del = 1;
3089 #endif
3090 } else {
3091 prev = rule;
3092 rule = rule->next;
3093 }
3094 }
3095 KASSERT(del, ("no match set?!"));
3096
3097 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3098 }
3099
3100 static void
3101 ipfw_disable_ruleset_state_dispatch(netmsg_t nmsg)
3102 {
3103 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3104 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3105 struct ip_fw *rule;
3106 #ifdef INVARIANTS
3107 int cleared = 0;
3108 #endif
3109
3110 ctx->ipfw_gen++;
3111
3112 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3113 if (rule->set == dmsg->from_set) {
3114 #ifdef INVARIANTS
3115 cleared = 1;
3116 #endif
3117 rule->rule_flags &= ~IPFW_RULE_F_STATE;
3118 }
3119 }
3120 KASSERT(cleared, ("no match set?!"));
3121
3122 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3123 }
3124
3125 static int
3126 ipfw_alt_delete_ruleset(uint8_t set)
3127 {
3128 struct netmsg_del dmsg;
3129 struct netmsg_base *nmsg;
3130 int state, del;
3131 struct ip_fw *rule;
3132 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3133
3134 /*
3135 * Check whether the 'set' exists. If it exists,
3136 * then check whether any rules within the set will
3137 * try to create states.
3138 */
3139 state = 0;
3140 del = 0;
3141 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3142 if (rule->set == set) {
3143 del = 1;
3144 if (rule->rule_flags & IPFW_RULE_F_STATE) {
3145 state = 1;
3146 break;
3147 }
3148 }
3149 }
3150 if (!del)
3151 return 0; /* XXX EINVAL? */
3152
3153 if (state) {
3154 /*
3155 * Clear the STATE flag, so no more states will be
3156 * created based the rules in this set.
3157 */
3158 bzero(&dmsg, sizeof(dmsg));
3159 nmsg = &dmsg.base;
3160 netmsg_init(nmsg, NULL, &curthread->td_msgport,
3161 0, ipfw_disable_ruleset_state_dispatch);
3162 dmsg.from_set = set;
3163
3164 ifnet_domsg(&nmsg->lmsg, 0);
3165
3166 /*
3167 * Nuke all related states
3168 */
3169 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3170 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3171 if (rule->set != set)
3172 continue;
3173
3174 /*
3175 * Can't check IPFW_RULE_F_STATE here,
3176 * since it has been cleared previously.
3177 * Check 'stub' instead.
3178 */
3179 if (rule->stub != NULL) {
3180 /* Force removal */
3181 remove_dyn_rule_locked(rule, NULL);
3182 }
3183 }
3184 lockmgr(&dyn_lock, LK_RELEASE);
3185 }
3186
3187 /*
3188 * Delete this set
3189 */
3190 bzero(&dmsg, sizeof(dmsg));
3191 nmsg = &dmsg.base;
3192 netmsg_init(nmsg, NULL, &curthread->td_msgport,
3193 0, ipfw_alt_delete_ruleset_dispatch);
3194 dmsg.from_set = set;
3195
3196 ifnet_domsg(&nmsg->lmsg, 0);
3197 return 0;
3198 }
3199
3200 static void
3201 ipfw_alt_move_rule_dispatch(netmsg_t nmsg)
3202 {
3203 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3204 struct ip_fw *rule;
3205
3206 rule = dmsg->start_rule;
3207 KKASSERT(rule->cpuid == mycpuid);
3208
3209 /*
3210 * Move to the position on the next CPU
3211 * before the msg is forwarded.
3212 */
3213 dmsg->start_rule = rule->sibling;
3214
3215 while (rule && rule->rulenum <= dmsg->rulenum) {
3216 if (rule->rulenum == dmsg->rulenum)
3217 rule->set = dmsg->to_set;
3218 rule = rule->next;
3219 }
3220 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3221 }
3222
3223 static int
3224 ipfw_alt_move_rule(uint16_t rulenum, uint8_t set)
3225 {
3226 struct netmsg_del dmsg;
3227 struct netmsg_base *nmsg;
3228 struct ip_fw *rule;
3229 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3230
3231 /*
3232 * Locate first rule to move
3233 */
3234 for (rule = ctx->ipfw_layer3_chain; rule && rule->rulenum <= rulenum;
3235 rule = rule->next) {
3236 if (rule->rulenum == rulenum && rule->set != set)
3237 break;
3238 }
3239 if (rule == NULL || rule->rulenum > rulenum)
3240 return 0; /* XXX error? */
3241
3242 bzero(&dmsg, sizeof(dmsg));
3243 nmsg = &dmsg.base;
3244 netmsg_init(nmsg, NULL, &curthread->td_msgport,
3245 0, ipfw_alt_move_rule_dispatch);
3246 dmsg.start_rule = rule;
3247 dmsg.rulenum = rulenum;
3248 dmsg.to_set = set;
3249
3250 ifnet_domsg(&nmsg->lmsg, 0);
3251 KKASSERT(dmsg.start_rule == NULL);
3252 return 0;
3253 }
3254
3255 static void
3256 ipfw_alt_move_ruleset_dispatch(netmsg_t nmsg)
3257 {
3258 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3259 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3260 struct ip_fw *rule;
3261
3262 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3263 if (rule->set == dmsg->from_set)
3264 rule->set = dmsg->to_set;
3265 }
3266 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3267 }
3268
3269 static int
3270 ipfw_alt_move_ruleset(uint8_t from_set, uint8_t to_set)
3271 {
3272 struct netmsg_del dmsg;
3273 struct netmsg_base *nmsg;
3274
3275 bzero(&dmsg, sizeof(dmsg));
3276 nmsg = &dmsg.base;
3277 netmsg_init(nmsg, NULL, &curthread->td_msgport,
3278 0, ipfw_alt_move_ruleset_dispatch);
3279 dmsg.from_set = from_set;
3280 dmsg.to_set = to_set;
3281
3282 ifnet_domsg(&nmsg->lmsg, 0);
3283 return 0;
3284 }
3285
3286 static void
3287 ipfw_alt_swap_ruleset_dispatch(netmsg_t nmsg)
3288 {
3289 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3290 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3291 struct ip_fw *rule;
3292
3293 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3294 if (rule->set == dmsg->from_set)
3295 rule->set = dmsg->to_set;
3296 else if (rule->set == dmsg->to_set)
3297 rule->set = dmsg->from_set;
3298 }
3299 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3300 }
3301
3302 static int
3303 ipfw_alt_swap_ruleset(uint8_t set1, uint8_t set2)
3304 {
3305 struct netmsg_del dmsg;
3306 struct netmsg_base *nmsg;
3307
3308 bzero(&dmsg, sizeof(dmsg));
3309 nmsg = &dmsg.base;
3310 netmsg_init(nmsg, NULL, &curthread->td_msgport,
3311 0, ipfw_alt_swap_ruleset_dispatch);
3312 dmsg.from_set = set1;
3313 dmsg.to_set = set2;
3314
3315 ifnet_domsg(&nmsg->lmsg, 0);
3316 return 0;
3317 }
3318
3319 /**
3320 * Remove all rules with given number, and also do set manipulation.
3321 *
3322 * The argument is an uint32_t. The low 16 bit are the rule or set number,
3323 * the next 8 bits are the new set, the top 8 bits are the command:
3324 *
3325 * 0 delete rules with given number
3326 * 1 delete rules with given set number
3327 * 2 move rules with given number to new set
3328 * 3 move rules with given set number to new set
3329 * 4 swap sets with given numbers
3330 */
3331 static int
3332 ipfw_ctl_alter(uint32_t arg)
3333 {
3334 uint16_t rulenum;
3335 uint8_t cmd, new_set;
3336 int error = 0;
3337
3338 rulenum = arg & 0xffff;
3339 cmd = (arg >> 24) & 0xff;
3340 new_set = (arg >> 16) & 0xff;
3341
3342 if (cmd > 4)
3343 return EINVAL;
3344 if (new_set >= IPFW_DEFAULT_SET)
3345 return EINVAL;
3346 if (cmd == 0 || cmd == 2) {
3347 if (rulenum == IPFW_DEFAULT_RULE)
3348 return EINVAL;
3349 } else {
3350 if (rulenum >= IPFW_DEFAULT_SET)
3351 return EINVAL;
3352 }
3353
3354 switch (cmd) {
3355 case 0: /* delete rules with given number */
3356 error = ipfw_alt_delete_rule(rulenum);
3357 break;
3358
3359 case 1: /* delete all rules with given set number */
3360 error = ipfw_alt_delete_ruleset(rulenum);
3361 break;
3362
3363 case 2: /* move rules with given number to new set */
3364 error = ipfw_alt_move_rule(rulenum, new_set);
3365 break;
3366
3367 case 3: /* move rules with given set number to new set */
3368 error = ipfw_alt_move_ruleset(rulenum, new_set);
3369 break;
3370
3371 case 4: /* swap two sets */
3372 error = ipfw_alt_swap_ruleset(rulenum, new_set);
3373 break;
3374 }
3375 return error;
3376 }
3377
3378 /*
3379 * Clear counters for a specific rule.
3380 */
3381 static void
3382 clear_counters(struct ip_fw *rule, int log_only)
3383 {
3384 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3385
3386 if (log_only == 0) {
3387 rule->bcnt = rule->pcnt = 0;
3388 rule->timestamp = 0;
3389 }
3390 if (l->o.opcode == O_LOG)
3391 l->log_left = l->max_log;
3392 }
3393
3394 static void
3395 ipfw_zero_entry_dispatch(netmsg_t nmsg)
3396 {
3397 struct netmsg_zent *zmsg = (struct netmsg_zent *)nmsg;
3398 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3399 struct ip_fw *rule;
3400
3401 if (zmsg->rulenum == 0) {
3402 KKASSERT(zmsg->start_rule == NULL);
3403
3404 ctx->ipfw_norule_counter = 0;
3405 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
3406 clear_counters(rule, zmsg->log_only);
3407 } else {
3408 struct ip_fw *start = zmsg->start_rule;
3409
3410 KKASSERT(start->cpuid == mycpuid);
3411 KKASSERT(start->rulenum == zmsg->rulenum);
3412
3413 /*
3414 * We can have multiple rules with the same number, so we
3415 * need to clear them all.
3416 */
3417 for (rule = start; rule && rule->rulenum == zmsg->rulenum;
3418 rule = rule->next)
3419 clear_counters(rule, zmsg->log_only);
3420
3421 /*
3422 * Move to the position on the next CPU
3423 * before the msg is forwarded.
3424 */
3425 zmsg->start_rule = start->sibling;
3426 }
3427 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
3428 }
3429
3430 /**
3431 * Reset some or all counters on firewall rules.
3432 * @arg frwl is null to clear all entries, or contains a specific
3433 * rule number.
3434 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3435 */
3436 static int
3437 ipfw_ctl_zero_entry(int rulenum, int log_only)
3438 {
3439 struct netmsg_zent zmsg;
3440 struct netmsg_base *nmsg;
3441 const char *msg;
3442 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3443
3444 bzero(&zmsg, sizeof(zmsg));
3445 nmsg = &zmsg.base;
3446 netmsg_init(nmsg, NULL, &curthread->td_msgport,
3447 0, ipfw_zero_entry_dispatch);
3448 zmsg.log_only = log_only;
3449
3450 if (rulenum == 0) {
3451 msg = log_only ? "ipfw: All logging counts reset.\n"
3452 : "ipfw: Accounting cleared.\n";
3453 } else {
3454 struct ip_fw *rule;
3455
3456 /*
3457 * Locate the first rule with 'rulenum'
3458 */
3459 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3460 if (rule->rulenum == rulenum)
3461 break;
3462 }
3463 if (rule == NULL) /* we did not find any matching rules */
3464 return (EINVAL);
3465 zmsg.start_rule = rule;
3466 zmsg.rulenum = rulenum;
3467
3468 msg = log_only ? "ipfw: Entry %d logging count reset.\n"
3469 : "ipfw: Entry %d cleared.\n";
3470 }
3471 ifnet_domsg(&nmsg->lmsg, 0);
3472 KKASSERT(zmsg.start_rule == NULL);
3473
3474 if (fw_verbose)
3475 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3476 return (0);
3477 }
3478
3479 /*
3480 * Check validity of the structure before insert.
3481 * Fortunately rules are simple, so this mostly need to check rule sizes.
3482 */
3483 static int
3484 ipfw_check_ioc_rule(struct ipfw_ioc_rule *rule, int size, uint32_t *rule_flags)
3485 {
3486 int l, cmdlen = 0;
3487 int have_action = 0;
3488 ipfw_insn *cmd;
3489
3490 *rule_flags = 0;
3491
3492 /* Check for valid size */
3493 if (size < sizeof(*rule)) {
3494 kprintf("ipfw: rule too short\n");
3495 return EINVAL;
3496 }
3497 l = IOC_RULESIZE(rule);
3498 if (l != size) {
3499 kprintf("ipfw: size mismatch (have %d want %d)\n", size, l);
3500 return EINVAL;
3501 }
3502
3503 /* Check rule number */
3504 if (rule->rulenum == IPFW_DEFAULT_RULE) {
3505 kprintf("ipfw: invalid rule number\n");
3506 return EINVAL;
3507 }
3508
3509 /*
3510 * Now go for the individual checks. Very simple ones, basically only
3511 * instruction sizes.
3512 */
3513 for (l = rule->cmd_len, cmd = rule->cmd; l > 0;
3514 l -= cmdlen, cmd += cmdlen) {
3515 cmdlen = F_LEN(cmd);
3516 if (cmdlen > l) {
3517 kprintf("ipfw: opcode %d size truncated\n",
3518 cmd->opcode);
3519 return EINVAL;
3520 }
3521
3522 DPRINTF("ipfw: opcode %d\n", cmd->opcode);
3523
3524 if (cmd->opcode == O_KEEP_STATE || cmd->opcode == O_LIMIT) {
3525 /* This rule will create states */
3526 *rule_flags |= IPFW_RULE_F_STATE;
3527 }
3528
3529 switch (cmd->opcode) {
3530 case O_NOP:
3531 case O_PROBE_STATE:
3532 case O_KEEP_STATE:
3533 case O_PROTO:
3534 case O_IP_SRC_ME:
3535 case O_IP_DST_ME:
3536 case O_LAYER2:
3537 case O_IN:
3538 case O_FRAG:
3539 case O_IPOPT:
3540 case O_IPLEN:
3541 case O_IPID:
3542 case O_IPTOS:
3543 case O_IPPRECEDENCE:
3544 case O_IPTTL:
3545 case O_IPVER:
3546 case O_TCPWIN:
3547 case O_TCPFLAGS:
3548 case O_TCPOPTS:
3549 case O_ESTAB:
3550 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3551 goto bad_size;
3552 break;
3553
3554 case O_UID:
3555 case O_GID:
3556 case O_IP_SRC:
3557 case O_IP_DST:
3558 case O_TCPSEQ:
3559 case O_TCPACK:
3560 case O_PROB:
3561 case O_ICMPTYPE:
3562 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3563 goto bad_size;
3564 break;
3565
3566 case O_LIMIT:
3567 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3568 goto bad_size;
3569 break;
3570
3571 case O_LOG:
3572 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3573 goto bad_size;
3574
3575 ((ipfw_insn_log *)cmd)->log_left =
3576 ((ipfw_insn_log *)cmd)->max_log;
3577
3578 break;
3579
3580 case O_IP_SRC_MASK:
3581 case O_IP_DST_MASK:
3582 if (cmdlen != F_INSN_SIZE(ipfw_insn_ip))
3583 goto bad_size;
3584 if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) {
3585 kprintf("ipfw: opcode %d, useless rule\n",
3586 cmd->opcode);
3587 return EINVAL;
3588 }
3589 break;
3590
3591 case O_IP_SRC_SET:
3592 case O_IP_DST_SET:
3593 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3594 kprintf("ipfw: invalid set size %d\n",
3595 cmd->arg1);
3596 return EINVAL;
3597 }
3598 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3599 (cmd->arg1+31)/32 )
3600 goto bad_size;
3601 break;
3602
3603 case O_MACADDR2:
3604 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3605 goto bad_size;
3606 break;
3607
3608 case O_MAC_TYPE:
3609 case O_IP_SRCPORT:
3610 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3611 if (cmdlen < 2 || cmdlen > 31)
3612 goto bad_size;
3613 break;
3614
3615 case O_RECV:
3616 case O_XMIT:
3617 case O_VIA:
3618 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3619 goto bad_size;
3620 break;
3621
3622 case O_PIPE:
3623 case O_QUEUE:
3624 if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
3625 goto bad_size;
3626 goto check_action;
3627
3628 case O_FORWARD_IP:
3629 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa)) {
3630 goto bad_size;
3631 } else {
3632 in_addr_t fwd_addr;
3633
3634 fwd_addr = ((ipfw_insn_sa *)cmd)->
3635 sa.sin_addr.s_addr;
3636 if (IN_MULTICAST(ntohl(fwd_addr))) {
3637 kprintf("ipfw: try forwarding to "
3638 "multicast address\n");
3639 return EINVAL;
3640 }
3641 }
3642 goto check_action;
3643
3644 case O_FORWARD_MAC: /* XXX not implemented yet */
3645 case O_CHECK_STATE:
3646 case O_COUNT:
3647 case O_ACCEPT:
3648 case O_DENY:
3649 case O_REJECT:
3650 case O_SKIPTO:
3651 case O_DIVERT:
3652 case O_TEE:
3653 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3654 goto bad_size;
3655 check_action:
3656 if (have_action) {
3657 kprintf("ipfw: opcode %d, multiple actions"
3658 " not allowed\n",
3659 cmd->opcode);
3660 return EINVAL;
3661 }
3662 have_action = 1;
3663 if (l != cmdlen) {
3664 kprintf("ipfw: opcode %d, action must be"
3665 " last opcode\n",
3666 cmd->opcode);
3667 return EINVAL;
3668 }
3669 break;
3670 default:
3671 kprintf("ipfw: opcode %d, unknown opcode\n",
3672 cmd->opcode);
3673 return EINVAL;
3674 }
3675 }
3676 if (have_action == 0) {
3677 kprintf("ipfw: missing action\n");
3678 return EINVAL;
3679 }
3680 return 0;
3681
3682 bad_size:
3683 kprintf("ipfw: opcode %d size %d wrong\n",
3684 cmd->opcode, cmdlen);
3685 return EINVAL;
3686 }
3687
3688 static int
3689 ipfw_ctl_add_rule(struct sockopt *sopt)
3690 {
3691 struct ipfw_ioc_rule *ioc_rule;
3692 size_t size;
3693 uint32_t rule_flags;
3694 int error;
3695
3696 size = sopt->sopt_valsize;
3697 if (size > (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX) ||
3698 size < sizeof(*ioc_rule)) {
3699 return EINVAL;
3700 }
3701 if (size != (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX)) {
3702 sopt->sopt_val = krealloc(sopt->sopt_val, sizeof(uint32_t) *
3703 IPFW_RULE_SIZE_MAX, M_TEMP, M_WAITOK);
3704 }
3705 ioc_rule = sopt->sopt_val;
3706
3707 error = ipfw_check_ioc_rule(ioc_rule, size, &rule_flags);
3708 if (error)
3709 return error;
3710
3711 ipfw_add_rule(ioc_rule, rule_flags);
3712
3713 if (sopt->sopt_dir == SOPT_GET)
3714 sopt->sopt_valsize = IOC_RULESIZE(ioc_rule);
3715 return 0;
3716 }
3717
3718 static void *
3719 ipfw_copy_rule(const struct ip_fw *rule, struct ipfw_ioc_rule *ioc_rule)
3720 {
3721 const struct ip_fw *sibling;
3722 #ifdef INVARIANTS
3723 int i;
3724 #endif
3725
3726 KKASSERT(rule->cpuid == IPFW_CFGCPUID);
3727
3728 ioc_rule->act_ofs = rule->act_ofs;
3729 ioc_rule->cmd_len = rule->cmd_len;
3730 ioc_rule->rulenum = rule->rulenum;
3731 ioc_rule->set = rule->set;
3732 ioc_rule->usr_flags = rule->usr_flags;
3733
3734 ioc_rule->set_disable = ipfw_ctx[mycpuid]->ipfw_set_disable;
3735 ioc_rule->static_count = static_count;
3736 ioc_rule->static_len = static_ioc_len;
3737
3738 /*
3739 * Visit (read-only) all of the rule's duplications to get
3740 * the necessary statistics
3741 */
3742 #ifdef INVARIANTS
3743 i = 0;
3744 #endif
3745 ioc_rule->pcnt = 0;
3746 ioc_rule->bcnt = 0;
3747 ioc_rule->timestamp = 0;
3748 for (sibling = rule; sibling != NULL; sibling = sibling->sibling) {
3749 ioc_rule->pcnt += sibling->pcnt;
3750 ioc_rule->bcnt += sibling->bcnt;
3751 if (sibling->timestamp > ioc_rule->timestamp)
3752 ioc_rule->timestamp = sibling->timestamp;
3753 #ifdef INVARIANTS
3754 ++i;
3755 #endif
3756 }
3757 KASSERT(i == ncpus, ("static rule is not duplicated on every cpu"));
3758
3759 bcopy(rule->cmd, ioc_rule->cmd, ioc_rule->cmd_len * 4 /* XXX */);
3760
3761 return ((uint8_t *)ioc_rule + IOC_RULESIZE(ioc_rule));
3762 }
3763
3764 static void
3765 ipfw_copy_state(const ipfw_dyn_rule *dyn_rule,
3766 struct ipfw_ioc_state *ioc_state)
3767 {
3768 const struct ipfw_flow_id *id;
3769 struct ipfw_ioc_flowid *ioc_id;
3770
3771 ioc_state->expire = TIME_LEQ(dyn_rule->expire, time_second) ?
3772 0 : dyn_rule->expire - time_second;
3773 ioc_state->pcnt = dyn_rule->pcnt;
3774 ioc_state->bcnt = dyn_rule->bcnt;
3775
3776 ioc_state->dyn_type = dyn_rule->dyn_type;
3777 ioc_state->count = dyn_rule->count;
3778
3779 ioc_state->rulenum = dyn_rule->stub->rule[mycpuid]->rulenum;
3780
3781 id = &dyn_rule->id;
3782 ioc_id = &ioc_state->id;
3783
3784 ioc_id->type = ETHERTYPE_IP;
3785 ioc_id->u.ip.dst_ip = id->dst_ip;
3786 ioc_id->u.ip.src_ip = id->src_ip;
3787 ioc_id->u.ip.dst_port = id->dst_port;
3788 ioc_id->u.ip.src_port = id->src_port;
3789 ioc_id->u.ip.proto = id->proto;
3790 }
3791
3792 static int
3793 ipfw_ctl_get_rules(struct sockopt *sopt)
3794 {
3795 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3796 struct ip_fw *rule;
3797 void *bp;
3798 size_t size;
3799 uint32_t dcount = 0;
3800
3801 /*
3802 * pass up a copy of the current rules. Static rules
3803 * come first (the last of which has number IPFW_DEFAULT_RULE),
3804 * followed by a possibly empty list of dynamic rule.
3805 */
3806
3807 size = static_ioc_len; /* size of static rules */
3808 if (ipfw_dyn_v) { /* add size of dyn.rules */
3809 dcount = dyn_count;
3810 size += dcount * sizeof(struct ipfw_ioc_state);
3811 }
3812
3813 if (sopt->sopt_valsize < size) {
3814 /* short length, no need to return incomplete rules */
3815 /* XXX: if superuser, no need to zero buffer */
3816 bzero(sopt->sopt_val, sopt->sopt_valsize);
3817 return 0;
3818 }
3819 bp = sopt->sopt_val;
3820
3821 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
3822 bp = ipfw_copy_rule(rule, bp);
3823
3824 if (ipfw_dyn_v && dcount != 0) {
3825 struct ipfw_ioc_state *ioc_state = bp;
3826 uint32_t dcount2 = 0;
3827 #ifdef INVARIANTS
3828 size_t old_size = size;
3829 #endif
3830 int i;
3831
3832 lockmgr(&dyn_lock, LK_SHARED);
3833
3834 /* Check 'ipfw_dyn_v' again with lock held */
3835 if (ipfw_dyn_v == NULL)
3836 goto skip;
3837
3838 for (i = 0; i < curr_dyn_buckets; i++) {
3839 ipfw_dyn_rule *p;
3840
3841 /*
3842 * The # of dynamic rules may have grown after the
3843 * snapshot of 'dyn_count' was taken, so we will have
3844 * to check 'dcount' (snapshot of dyn_count) here to
3845 * make sure that we don't overflow the pre-allocated
3846 * buffer.
3847 */
3848 for (p = ipfw_dyn_v[i]; p != NULL && dcount != 0;
3849 p = p->next, ioc_state++, dcount--, dcount2++)
3850 ipfw_copy_state(p, ioc_state);
3851 }
3852 skip:
3853 lockmgr(&dyn_lock, LK_RELEASE);
3854
3855 /*
3856 * The # of dynamic rules may be shrinked after the
3857 * snapshot of 'dyn_count' was taken. To give user a
3858 * correct dynamic rule count, we use the 'dcount2'
3859 * calculated above (with shared lockmgr lock held).
3860 */
3861 size = static_ioc_len +
3862 (dcount2 * sizeof(struct ipfw_ioc_state));
3863 KKASSERT(size <= old_size);
3864 }
3865
3866 sopt->sopt_valsize = size;
3867 return 0;
3868 }
3869
3870 static void
3871 ipfw_set_disable_dispatch(netmsg_t nmsg)
3872 {
3873 struct lwkt_msg *lmsg = &nmsg->lmsg;
3874 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3875
3876 ctx->ipfw_gen++;
3877 ctx->ipfw_set_disable = lmsg->u.ms_result32;
3878
3879 ifnet_forwardmsg(lmsg, mycpuid + 1);
3880 }
3881
3882 static void
3883 ipfw_ctl_set_disable(uint32_t disable, uint32_t enable)
3884 {
3885 struct netmsg_base nmsg;
3886 struct lwkt_msg *lmsg;
3887 uint32_t set_disable;
3888
3889 /* IPFW_DEFAULT_SET is always enabled */
3890 enable |= (1 << IPFW_DEFAULT_SET);
3891 set_disable = (ipfw_ctx[mycpuid]->ipfw_set_disable | disable) & ~enable;
3892
3893 bzero(&nmsg, sizeof(nmsg));
3894 netmsg_init(&nmsg, NULL, &curthread->td_msgport,
3895 0, ipfw_set_disable_dispatch);
3896 lmsg = &nmsg.lmsg;
3897 lmsg->u.ms_result32 = set_disable;
3898
3899 ifnet_domsg(lmsg, 0);
3900 }
3901
3902 /**
3903 * {set|get}sockopt parser.
3904 */
3905 static int
3906 ipfw_ctl(struct sockopt *sopt)
3907 {
3908 int error, rulenum;
3909 uint32_t *masks;
3910 size_t size;
3911
3912 error = 0;
3913
3914 switch (sopt->sopt_name) {
3915 case IP_FW_GET:
3916 error = ipfw_ctl_get_rules(sopt);
3917 break;
3918
3919 case IP_FW_FLUSH:
3920 ipfw_flush(0 /* keep default rule */);
3921 break;
3922
3923 case IP_FW_ADD:
3924 error = ipfw_ctl_add_rule(sopt);
3925 break;
3926
3927 case IP_FW_DEL:
3928 /*
3929 * IP_FW_DEL is used for deleting single rules or sets,
3930 * and (ab)used to atomically manipulate sets.
3931 * Argument size is used to distinguish between the two:
3932 * sizeof(uint32_t)
3933 * delete single rule or set of rules,
3934 * or reassign rules (or sets) to a different set.
3935 * 2 * sizeof(uint32_t)
3936 * atomic disable/enable sets.
3937 * first uint32_t contains sets to be disabled,
3938 * second uint32_t contains sets to be enabled.
3939 */
3940 masks = sopt->sopt_val;
3941 size = sopt->sopt_valsize;
3942 if (size == sizeof(*masks)) {
3943 /*
3944 * Delete or reassign static rule
3945 */
3946 error = ipfw_ctl_alter(masks[0]);
3947 } else if (size == (2 * sizeof(*masks))) {
3948 /*
3949 * Set enable/disable
3950 */
3951 ipfw_ctl_set_disable(masks[0], masks[1]);
3952 } else {
3953 error = EINVAL;
3954 }
3955 break;
3956
3957 case IP_FW_ZERO:
3958 case IP_FW_RESETLOG: /* argument is an int, the rule number */
3959 rulenum = 0;
3960
3961 if (sopt->sopt_val != 0) {
3962 error = soopt_to_kbuf(sopt, &rulenum,
3963 sizeof(int), sizeof(int));
3964 if (error)
3965 break;
3966 }
3967 error = ipfw_ctl_zero_entry(rulenum,
3968 sopt->sopt_name == IP_FW_RESETLOG);
3969 break;
3970
3971 default:
3972 kprintf("ipfw_ctl invalid option %d\n", sopt->sopt_name);
3973 error = EINVAL;
3974 }
3975 return error;
3976 }
3977
3978 /*
3979 * This procedure is only used to handle keepalives. It is invoked
3980 * every dyn_keepalive_period
3981 */
3982 static void
3983 ipfw_tick_dispatch(netmsg_t nmsg)
3984 {
3985 time_t keep_alive;
3986 uint32_t gen;
3987 int i;
3988
3989 IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
3990 KKASSERT(IPFW_LOADED);
3991
3992 /* Reply ASAP */
3993 crit_enter();
3994 lwkt_replymsg(&nmsg->lmsg, 0);
3995 crit_exit();
3996
3997 if (ipfw_dyn_v == NULL || dyn_count == 0)
3998 goto done;
3999
4000 keep_alive = time_second;
4001
4002 lockmgr(&dyn_lock, LK_EXCLUSIVE);
4003 again:
4004 if (ipfw_dyn_v == NULL || dyn_count == 0) {
4005 lockmgr(&dyn_lock, LK_RELEASE);
4006 goto done;
4007 }
4008 gen = dyn_buckets_gen;
4009
4010 for (i = 0; i < curr_dyn_buckets; i++) {
4011 ipfw_dyn_rule *q, *prev;
4012
4013 for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) {
4014 uint32_t ack_rev, ack_fwd;
4015 struct ipfw_flow_id id;
4016
4017 if (q->dyn_type == O_LIMIT_PARENT)
4018 goto next;
4019
4020 if (TIME_LEQ(q->expire, time_second)) {
4021 /* State expired */
4022 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
4023 continue;
4024 }
4025
4026 /*
4027 * Keep alive processing
4028 */
4029
4030 if (!dyn_keepalive)
4031 goto next;
4032 if (q->id.proto != IPPROTO_TCP)
4033 goto next;
4034 if ((q->state & BOTH_SYN) != BOTH_SYN)
4035 goto next;
4036 if (TIME_LEQ(time_second + dyn_keepalive_interval,
4037 q->expire))
4038 goto next; /* too early */
4039 if (q->keep_alive == keep_alive)
4040 goto next; /* alreay done */
4041
4042 /*
4043 * Save necessary information, so that they could
4044 * survive after possible blocking in send_pkt()
4045 */
4046 id = q->id;
4047 ack_rev = q->ack_rev;
4048 ack_fwd = q->ack_fwd;
4049
4050 /* Sending has been started */
4051 q->keep_alive = keep_alive;
4052
4053 /* Release lock to avoid possible dead lock */
4054 lockmgr(&dyn_lock, LK_RELEASE);
4055 send_pkt(&id, ack_rev - 1, ack_fwd, TH_SYN);
4056 send_pkt(&id, ack_fwd - 1, ack_rev, 0);
4057 lockmgr(&dyn_lock, LK_EXCLUSIVE);
4058
4059 if (gen != dyn_buckets_gen) {
4060 /*
4061 * Dyn bucket array has been changed during
4062 * the above two sending; reiterate.
4063 */
4064 goto again;
4065 }
4066 next:
4067 prev = q;
4068 q = q->next;
4069 }
4070 }
4071 lockmgr(&dyn_lock, LK_RELEASE);
4072 done:
4073 callout_reset(&ipfw_timeout_h, dyn_keepalive_period * hz,
4074 ipfw_tick, NULL);
4075 }
4076
4077 /*
4078 * This procedure is only used to handle keepalives. It is invoked
4079 * every dyn_keepalive_period
4080 */
4081 static void
4082 ipfw_tick(void *dummy __unused)
4083 {
4084 struct lwkt_msg *lmsg = &ipfw_timeout_netmsg.lmsg;
4085
4086 KKASSERT(mycpuid == IPFW_CFGCPUID);
4087
4088 crit_enter();
4089
4090 KKASSERT(lmsg->ms_flags & MSGF_DONE);
4091 if (IPFW_LOADED) {
4092 lwkt_sendmsg_oncpu(IPFW_CFGPORT, lmsg);
4093 /* ipfw_timeout_netmsg's handler reset this callout */
4094 }
4095
4096 crit_exit();
4097 }
4098
4099 static int
4100 ipfw_check_in(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir)
4101 {
4102 struct ip_fw_args args;
4103 struct mbuf *m = *m0;
4104 struct m_tag *mtag;
4105 int tee = 0, error = 0, ret;
4106
4107 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
4108 /* Extract info from dummynet tag */
4109 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
4110 KKASSERT(mtag != NULL);
4111 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
4112 KKASSERT(args.rule != NULL);
4113
4114 m_tag_delete(m, mtag);
4115 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
4116 } else {
4117 args.rule = NULL;
4118 }
4119
4120 args.eh = NULL;
4121 args.oif = NULL;
4122 args.m = m;
4123 ret = ipfw_chk(&args);
4124 m = args.m;
4125
4126 if (m == NULL) {
4127 error = EACCES;
4128 goto back;
4129 }
4130
4131 switch (ret) {
4132 case IP_FW_PASS:
4133 break;
4134
4135 case IP_FW_DENY:
4136 m_freem(m);
4137 m = NULL;
4138 error = EACCES;
4139 break;
4140
4141 case IP_FW_DUMMYNET:
4142 /* Send packet to the appropriate pipe */
4143 ipfw_dummynet_io(m, args.cookie, DN_TO_IP_IN, &args);
4144 break;
4145
4146 case IP_FW_TEE:
4147 tee = 1;
4148 /* FALL THROUGH */
4149
4150 case IP_FW_DIVERT:
4151 /*
4152 * Must clear bridge tag when changing
4153 */
4154 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
4155 if (ip_divert_p != NULL) {
4156 m = ip_divert_p(m, tee, 1);
4157 } else {
4158 m_freem(m);
4159 m = NULL;
4160 /* not sure this is the right error msg */
4161 error = EACCES;
4162 }
4163 break;
4164
4165 default:
4166 panic("unknown ipfw return value: %d", ret);
4167 }
4168 back:
4169 *m0 = m;
4170 return error;
4171 }
4172
4173 static int
4174 ipfw_check_out(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir)
4175 {
4176 struct ip_fw_args args;
4177 struct mbuf *m = *m0;
4178 struct m_tag *mtag;
4179 int tee = 0, error = 0, ret;
4180
4181 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) {
4182 /* Extract info from dummynet tag */
4183 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
4184 KKASSERT(mtag != NULL);
4185 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv;
4186 KKASSERT(args.rule != NULL);
4187
4188 m_tag_delete(m, mtag);
4189 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED;
4190 } else {
4191 args.rule = NULL;
4192 }
4193
4194 args.eh = NULL;
4195 args.m = m;
4196 args.oif = ifp;
4197 ret = ipfw_chk(&args);
4198 m = args.m;
4199
4200 if (m == NULL) {
4201 error = EACCES;
4202 goto back;
4203 }
4204
4205 switch (ret) {
4206 case IP_FW_PASS:
4207 break;
4208
4209 case IP_FW_DENY:
4210 m_freem(m);
4211 m = NULL;
4212 error = EACCES;
4213 break;
4214
4215 case IP_FW_DUMMYNET:
4216 ipfw_dummynet_io(m, args.cookie, DN_TO_IP_OUT, &args);
4217 break;
4218
4219 case IP_FW_TEE:
4220 tee = 1;
4221 /* FALL THROUGH */
4222
4223 case IP_FW_DIVERT:
4224 if (ip_divert_p != NULL) {
4225 m = ip_divert_p(m, tee, 0);
4226 } else {
4227 m_freem(m);
4228 m = NULL;
4229 /* not sure this is the right error msg */
4230 error = EACCES;
4231 }
4232 break;
4233
4234 default:
4235 panic("unknown ipfw return value: %d", ret);
4236 }
4237 back:
4238 *m0 = m;
4239 return error;
4240 }
4241
4242 static void
4243 ipfw_hook(void)
4244 {
4245 struct pfil_head *pfh;
4246
4247 IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
4248
4249 pfh = pfil_head_get(PFIL_TYPE_AF, AF_INET);
4250 if (pfh == NULL)
4251 return;
4252
4253 pfil_add_hook(ipfw_check_in, NULL, PFIL_IN | PFIL_MPSAFE, pfh);
4254 pfil_add_hook(ipfw_check_out, NULL, PFIL_OUT | PFIL_MPSAFE, pfh);
4255 }
4256
4257 static void
4258 ipfw_dehook(void)
4259 {
4260 struct pfil_head *pfh;
4261
4262 IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
4263
4264 pfh = pfil_head_get(PFIL_TYPE_AF, AF_INET);
4265 if (pfh == NULL)
4266 return;
4267
4268 pfil_remove_hook(ipfw_check_in, NULL, PFIL_IN, pfh);
4269 pfil_remove_hook(ipfw_check_out, NULL, PFIL_OUT, pfh);
4270 }
4271
4272 static void
4273 ipfw_sysctl_enable_dispatch(netmsg_t nmsg)
4274 {
4275 struct lwkt_msg *lmsg = &nmsg->lmsg;
4276 int enable = lmsg->u.ms_result;
4277
4278 if (fw_enable == enable)
4279 goto reply;
4280
4281 fw_enable = enable;
4282 if (fw_enable)
4283 ipfw_hook();
4284 else
4285 ipfw_dehook();
4286 reply:
4287 lwkt_replymsg(lmsg, 0);
4288 }
4289
4290 static int
4291 ipfw_sysctl_enable(SYSCTL_HANDLER_ARGS)
4292 {
4293 struct netmsg_base nmsg;
4294 struct lwkt_msg *lmsg;
4295 int enable, error;
4296
4297 enable = fw_enable;
4298 error = sysctl_handle_int(oidp, &enable, 0, req);
4299 if (error || req->newptr == NULL)
4300 return error;
4301
4302 netmsg_init(&nmsg, NULL, &curthread->td_msgport,
4303 0, ipfw_sysctl_enable_dispatch);
4304 lmsg = &nmsg.lmsg;
4305 lmsg->u.ms_result = enable;
4306
4307 return lwkt_domsg(IPFW_CFGPORT, lmsg, 0);
4308 }
4309
4310 static int
4311 ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS)
4312 {
4313 return sysctl_int_range(oidp, arg1, arg2, req,
4314 IPFW_AUTOINC_STEP_MIN, IPFW_AUTOINC_STEP_MAX);
4315 }
4316
4317 static int
4318 ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS)
4319 {
4320 int error, value;
4321
4322 lockmgr(&dyn_lock, LK_EXCLUSIVE);
4323
4324 value = dyn_buckets;
4325 error = sysctl_handle_int(oidp, &value, 0, req);
4326 if (error || !req->newptr)
4327 goto back;
4328
4329 /*
4330 * Make sure we have a power of 2 and
4331 * do not allow more than 64k entries.
4332 */
4333 error = EINVAL;
4334 if (value <= 1 || value > 65536)
4335 goto back;
4336 if ((value & (value - 1)) != 0)
4337 goto back;
4338
4339 error = 0;
4340 dyn_buckets = value;
4341 back:
4342 lockmgr(&dyn_lock, LK_RELEASE);
4343 return error;
4344 }
4345
4346 static int
4347 ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS)
4348 {
4349 return sysctl_int_range(oidp, arg1, arg2, req,
4350 1, dyn_keepalive_period - 1);
4351 }
4352
4353 static int
4354 ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS)
4355 {
4356 return sysctl_int_range(oidp, arg1, arg2, req,
4357 1, dyn_keepalive_period - 1);
4358 }
4359
4360 static void
4361 ipfw_ctx_init_dispatch(netmsg_t nmsg)
4362 {
4363 struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg;
4364 struct ipfw_context *ctx;
4365 struct ip_fw *def_rule;
4366
4367 ctx = kmalloc(sizeof(*ctx), M_IPFW, M_WAITOK | M_ZERO);
4368 ipfw_ctx[mycpuid] = ctx;
4369
4370 def_rule = kmalloc(sizeof(*def_rule), M_IPFW, M_WAITOK | M_ZERO);
4371
4372 def_rule->act_ofs = 0;
4373 def_rule->rulenum = IPFW_DEFAULT_RULE;
4374 def_rule->cmd_len = 1;
4375 def_rule->set = IPFW_DEFAULT_SET;
4376
4377 def_rule->cmd[0].len = 1;
4378 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
4379 def_rule->cmd[0].opcode = O_ACCEPT;
4380 #else
4381 if (filters_default_to_accept)
4382 def_rule->cmd[0].opcode = O_ACCEPT;
4383 else
4384 def_rule->cmd[0].opcode = O_DENY;
4385 #endif
4386
4387 def_rule->refcnt = 1;
4388 def_rule->cpuid = mycpuid;
4389
4390 /* Install the default rule */
4391 ctx->ipfw_default_rule = def_rule;
4392 ctx->ipfw_layer3_chain = def_rule;
4393
4394 /* Link rule CPU sibling */
4395 ipfw_link_sibling(fwmsg, def_rule);
4396
4397 /* Statistics only need to be updated once */
4398 if (mycpuid == 0)
4399 ipfw_inc_static_count(def_rule);
4400
4401 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1);
4402 }
4403
4404 static void
4405 ipfw_init_dispatch(netmsg_t nmsg)
4406 {
4407 struct netmsg_ipfw fwmsg;
4408 int error = 0;
4409
4410 if (IPFW_LOADED) {
4411 kprintf("IP firewall already loaded\n");
4412 error = EEXIST;
4413 goto reply;
4414 }
4415
4416 bzero(&fwmsg, sizeof(fwmsg));
4417 netmsg_init(&fwmsg.base, NULL, &curthread->td_msgport,
4418 0, ipfw_ctx_init_dispatch);
4419 ifnet_domsg(&fwmsg.base.lmsg, 0);
4420
4421 ip_fw_chk_ptr = ipfw_chk;
4422 ip_fw_ctl_ptr = ipfw_ctl;
4423 ip_fw_dn_io_ptr = ipfw_dummynet_io;
4424
4425 kprintf("ipfw2 initialized, default to %s, logging ",
4426 ipfw_ctx[mycpuid]->ipfw_default_rule->cmd[0].opcode ==
4427 O_ACCEPT ? "accept" : "deny");
4428
4429 #ifdef IPFIREWALL_VERBOSE
4430 fw_verbose = 1;
4431 #endif
4432 #ifdef IPFIREWALL_VERBOSE_LIMIT
4433 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
4434 #endif
4435 if (fw_verbose == 0) {
4436 kprintf("disabled\n");
4437 } else if (verbose_limit == 0) {
4438 kprintf("unlimited\n");
4439 } else {
4440 kprintf("limited to %d packets/entry by default\n",
4441 verbose_limit);
4442 }
4443
4444 callout_init_mp(&ipfw_timeout_h);
4445 netmsg_init(&ipfw_timeout_netmsg, NULL, &netisr_adone_rport,
4446 MSGF_DROPABLE | MSGF_PRIORITY,
4447 ipfw_tick_dispatch);
4448 lockinit(&dyn_lock, "ipfw_dyn", 0, 0);
4449
4450 ip_fw_loaded = 1;
4451 callout_reset(&ipfw_timeout_h, hz, ipfw_tick, NULL);
4452
4453 if (fw_enable)
4454 ipfw_hook();
4455 reply:
4456 lwkt_replymsg(&nmsg->lmsg, error);
4457 }
4458
4459 static int
4460 ipfw_init(void)
4461 {
4462 struct netmsg_base smsg;
4463
4464 netmsg_init(&smsg, NULL, &curthread->td_msgport,
4465 0, ipfw_init_dispatch);
4466 return lwkt_domsg(IPFW_CFGPORT, &smsg.lmsg, 0);
4467 }
4468
4469 #ifdef KLD_MODULE
4470
4471 static void
4472 ipfw_fini_dispatch(netmsg_t nmsg)
4473 {
4474 int error = 0, cpu;
4475
4476 if (ipfw_refcnt != 0) {
4477 error = EBUSY;
4478 goto reply;
4479 }
4480
4481 ip_fw_loaded = 0;
4482
4483 ipfw_dehook();
4484 callout_stop(&ipfw_timeout_h);
4485
4486 netmsg_service_sync();
4487
4488 crit_enter();
4489 lwkt_dropmsg(&ipfw_timeout_netmsg.lmsg);
4490 crit_exit();
4491
4492 ip_fw_chk_ptr = NULL;
4493 ip_fw_ctl_ptr = NULL;
4494 ip_fw_dn_io_ptr = NULL;
4495 ipfw_flush(1 /* kill default rule */);
4496
4497 /* Free pre-cpu context */
4498 for (cpu = 0; cpu < ncpus; ++cpu)
4499 kfree(ipfw_ctx[cpu], M_IPFW);
4500
4501 kprintf("IP firewall unloaded\n");
4502 reply:
4503 lwkt_replymsg(&nmsg->lmsg, error);
4504 }
4505
4506 static int
4507 ipfw_fini(void)
4508 {
4509 struct netmsg_base smsg;
4510
4511 netmsg_init(&smsg, NULL, &curthread->td_msgport,
4512 0, ipfw_fini_dispatch);
4513 return lwkt_domsg(IPFW_CFGPORT, &smsg.lmsg, 0);
4514 }
4515
4516 #endif /* KLD_MODULE */
4517
4518 static int
4519 ipfw_modevent(module_t mod, int type, void *unused)
4520 {
4521 int err = 0;
4522
4523 switch (type) {
4524 case MOD_LOAD:
4525 err = ipfw_init();
4526 break;
4527
4528 case MOD_UNLOAD:
4529 #ifndef KLD_MODULE
4530 kprintf("ipfw statically compiled, cannot unload\n");
4531 err = EBUSY;
4532 #else
4533 err = ipfw_fini();
4534 #endif
4535 break;
4536 default:
4537 break;
4538 }
4539 return err;
4540 }
4541
4542 static moduledata_t ipfwmod = {
4543 "ipfw",
4544 ipfw_modevent,
4545 0
4546 };
4547 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PROTO_END, SI_ORDER_ANY);
4548 MODULE_VERSION(ipfw, 1);
DragonFly BSD