AMD64 - Fix many compile-time warnings. int/ptr type mismatches, %llx, etc.
[dragonfly.git] / sys / net / ipfw / ip_fw2.c
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  * $DragonFly: src/sys/net/ipfw/ip_fw2.c,v 1.100 2008/11/22 11:03:35 sephe Exp $
27  */
28
29 /*
30  * Implement IP packet firewall (new version)
31  */
32
33 #include "opt_ipfw.h"
34 #include "opt_inet.h"
35 #ifndef INET
36 #error IPFIREWALL requires INET.
37 #endif /* INET */
38
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/kernel.h>
44 #include <sys/proc.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/sysctl.h>
48 #include <sys/syslog.h>
49 #include <sys/thread2.h>
50 #include <sys/ucred.h>
51 #include <sys/in_cksum.h>
52 #include <sys/lock.h>
53
54 #include <net/if.h>
55 #include <net/route.h>
56 #include <net/netmsg2.h>
57 #include <net/pfil.h>
58 #include <net/dummynet/ip_dummynet.h>
59
60 #include <netinet/in.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/in_var.h>
63 #include <netinet/in_pcb.h>
64 #include <netinet/ip.h>
65 #include <netinet/ip_var.h>
66 #include <netinet/ip_icmp.h>
67 #include <netinet/tcp.h>
68 #include <netinet/tcp_timer.h>
69 #include <netinet/tcp_var.h>
70 #include <netinet/tcpip.h>
71 #include <netinet/udp.h>
72 #include <netinet/udp_var.h>
73 #include <netinet/ip_divert.h>
74 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
75
76 #include <net/ipfw/ip_fw2.h>
77
78 #ifdef IPFIREWALL_DEBUG
79 #define DPRINTF(fmt, ...) \
80 do { \
81         if (fw_debug > 0) \
82                 kprintf(fmt, __VA_ARGS__); \
83 } while (0)
84 #else
85 #define DPRINTF(fmt, ...)       ((void)0)
86 #endif
87
88 /*
89  * Description about per-CPU rule duplication:
90  *
91  * Module loading/unloading and all ioctl operations are serialized
92  * by netisr0, so we don't have any ordering or locking problems.
93  *
94  * Following graph shows how operation on per-CPU rule list is
95  * performed [2 CPU case]:
96  *
97  *   CPU0                 CPU1
98  *
99  * netisr0 <------------------------------------+
100  *  domsg                                       |
101  *    |                                         |
102  *    | netmsg                                  |
103  *    |                                         |
104  *    V                                         |
105  *  ifnet0                                      |
106  *    :                                         | netmsg
107  *    :(delete/add...)                          |
108  *    :                                         |
109  *    :         netmsg                          |
110  *  forwardmsg---------->ifnet1                 |
111  *                          :                   |
112  *                          :(delete/add...)    |
113  *                          :                   |
114  *                          :                   |
115  *                        replymsg--------------+
116  *
117  *
118  *
119  *
120  * Rules which will not create states (dyn rules) [2 CPU case]
121  *
122  *    CPU0               CPU1
123  * layer3_chain       layer3_chain
124  *     |                  |
125  *     V                  V
126  * +-------+ sibling  +-------+ sibling
127  * | rule1 |--------->| rule1 |--------->NULL
128  * +-------+          +-------+
129  *     |                  |
130  *     |next              |next
131  *     V                  V
132  * +-------+ sibling  +-------+ sibling
133  * | rule2 |--------->| rule2 |--------->NULL
134  * +-------+          +-------+
135  *
136  * ip_fw.sibling:
137  * 1) Ease statistics calculation during IP_FW_GET.  We only need to
138  *    iterate layer3_chain on CPU0; the current rule's duplication on
139  *    the other CPUs could safely be read-only accessed by using
140  *    ip_fw.sibling
141  * 2) Accelerate rule insertion and deletion, e.g. rule insertion:
142  *    a) In netisr0 (on CPU0) rule3 is determined to be inserted between
143  *       rule1 and rule2.  To make this decision we need to iterate the
144  *       layer3_chain on CPU0.  The netmsg, which is used to insert the
145  *       rule, will contain rule1 on CPU0 as prev_rule and rule2 on CPU0
146  *       as next_rule
147  *    b) After the insertion on CPU0 is done, we will move on to CPU1.
148  *       But instead of relocating the rule3's position on CPU1 by
149  *       iterating the layer3_chain on CPU1, we set the netmsg's prev_rule
150  *       to rule1->sibling and next_rule to rule2->sibling before the
151  *       netmsg is forwarded to CPU1 from CPU0
152  *       
153  *    
154  *
155  * Rules which will create states (dyn rules) [2 CPU case]
156  * (unnecessary parts are omitted; they are same as in the previous figure)
157  *
158  *   CPU0                       CPU1
159  * 
160  * +-------+                  +-------+
161  * | rule1 |                  | rule1 |
162  * +-------+                  +-------+
163  *   ^   |                      |   ^
164  *   |   |stub              stub|   |
165  *   |   |                      |   |
166  *   |   +----+            +----+   |
167  *   |        |            |        |
168  *   |        V            V        |
169  *   |    +--------------------+    |
170  *   |    |     rule_stub      |    |
171  *   |    | (read-only shared) |    |
172  *   |    |                    |    |
173  *   |    | back pointer array |    |
174  *   |    | (indexed by cpuid) |    |
175  *   |    |                    |    |
176  *   +----|---------[0]        |    |
177  *        |         [1]--------|----+
178  *        |                    |
179  *        +--------------------+
180  *          ^            ^
181  *          |            |
182  *  ........|............|............
183  *  :       |            |           :
184  *  :       |stub        |stub       :
185  *  :       |            |           :
186  *  :  +---------+  +---------+      :
187  *  :  | state1a |  | state1b | .... :
188  *  :  +---------+  +---------+      :
189  *  :                                :
190  *  :           states table         :
191  *  :            (shared)            :
192  *  :      (protected by dyn_lock)   :
193  *  ..................................
194  * 
195  * [state1a and state1b are states created by rule1]
196  *
197  * ip_fw_stub:
198  * This structure is introduced so that shared (locked) state table could
199  * work with per-CPU (duplicated) static rules.  It mainly bridges states
200  * and static rules and serves as static rule's place holder (a read-only
201  * shared part of duplicated rules) from states point of view.
202  *
203  * IPFW_RULE_F_STATE (only for rules which create states):
204  * o  During rule installation, this flag is turned on after rule's
205  *    duplications reach all CPUs, to avoid at least following race:
206  *    1) rule1 is duplicated on CPU0 and is not duplicated on CPU1 yet
207  *    2) rule1 creates state1
208  *    3) state1 is located on CPU1 by check-state
209  *    But rule1 is not duplicated on CPU1 yet
210  * o  During rule deletion, this flag is turned off before deleting states
211  *    created by the rule and before deleting the rule itself, so no
212  *    more states will be created by the to-be-deleted rule even when its
213  *    duplication on certain CPUs are not eliminated yet.
214  */
215
216 #define IPFW_AUTOINC_STEP_MIN   1
217 #define IPFW_AUTOINC_STEP_MAX   1000
218 #define IPFW_AUTOINC_STEP_DEF   100
219
220 #define IPFW_DEFAULT_RULE       65535   /* rulenum for the default rule */
221 #define IPFW_DEFAULT_SET        31      /* set number for the default rule */
222
223 struct netmsg_ipfw {
224         struct netmsg   nmsg;
225         const struct ipfw_ioc_rule *ioc_rule;
226         struct ip_fw    *next_rule;
227         struct ip_fw    *prev_rule;
228         struct ip_fw    *sibling;
229         struct ip_fw_stub *stub;
230 };
231
232 struct netmsg_del {
233         struct netmsg   nmsg;
234         struct ip_fw    *start_rule;
235         struct ip_fw    *prev_rule;
236         uint16_t        rulenum;
237         uint8_t         from_set;
238         uint8_t         to_set;
239 };
240
241 struct netmsg_zent {
242         struct netmsg   nmsg;
243         struct ip_fw    *start_rule;
244         uint16_t        rulenum;
245         uint16_t        log_only;
246 };
247
248 struct ipfw_context {
249         struct ip_fw    *ipfw_layer3_chain;     /* list of rules for layer3 */
250         struct ip_fw    *ipfw_default_rule;     /* default rule */
251         uint64_t        ipfw_norule_counter;    /* counter for ipfw_log(NULL) */
252
253         /*
254          * ipfw_set_disable contains one bit per set value (0..31).
255          * If the bit is set, all rules with the corresponding set
256          * are disabled.  Set IPDW_DEFAULT_SET is reserved for the
257          * default rule and CANNOT be disabled.
258          */
259         uint32_t        ipfw_set_disable;
260         uint32_t        ipfw_gen;               /* generation of rule list */
261 };
262
263 static struct ipfw_context      *ipfw_ctx[MAXCPU];
264
265 #ifdef KLD_MODULE
266 /*
267  * Module can not be unloaded, if there are references to
268  * certains rules of ipfw(4), e.g. dummynet(4)
269  */
270 static int ipfw_refcnt;
271 #endif
272
273 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
274
275 /*
276  * Following two global variables are accessed and
277  * updated only on CPU0
278  */
279 static uint32_t static_count;   /* # of static rules */
280 static uint32_t static_ioc_len; /* bytes of static rules */
281
282 /*
283  * If 1, then ipfw static rules are being flushed,
284  * ipfw_chk() will skip to the default rule.
285  */
286 static int ipfw_flushing;
287
288 static int fw_verbose;
289 static int verbose_limit;
290
291 static int fw_debug;
292 static int autoinc_step = IPFW_AUTOINC_STEP_DEF;
293
294 static int      ipfw_sysctl_enable(SYSCTL_HANDLER_ARGS);
295 static int      ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS);
296 static int      ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS);
297 static int      ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS);
298 static int      ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS);
299
300 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
301 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, enable, CTLTYPE_INT | CTLFLAG_RW,
302     &fw_enable, 0, ipfw_sysctl_enable, "I", "Enable ipfw");
303 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLTYPE_INT | CTLFLAG_RW,
304     &autoinc_step, 0, ipfw_sysctl_autoinc_step, "I",
305     "Rule number autincrement step");
306 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW,
307     &fw_one_pass, 0,
308     "Only do a single pass through ipfw when using dummynet(4)");
309 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
310     &fw_debug, 0, "Enable printing of debug ip_fw statements");
311 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW,
312     &fw_verbose, 0, "Log matches to ipfw rules");
313 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
314     &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
315
316 /*
317  * Description of dynamic rules.
318  *
319  * Dynamic rules are stored in lists accessed through a hash table
320  * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
321  * be modified through the sysctl variable dyn_buckets which is
322  * updated when the table becomes empty.
323  *
324  * XXX currently there is only one list, ipfw_dyn.
325  *
326  * When a packet is received, its address fields are first masked
327  * with the mask defined for the rule, then hashed, then matched
328  * against the entries in the corresponding list.
329  * Dynamic rules can be used for different purposes:
330  *  + stateful rules;
331  *  + enforcing limits on the number of sessions;
332  *  + in-kernel NAT (not implemented yet)
333  *
334  * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
335  * measured in seconds and depending on the flags.
336  *
337  * The total number of dynamic rules is stored in dyn_count.
338  * The max number of dynamic rules is dyn_max. When we reach
339  * the maximum number of rules we do not create anymore. This is
340  * done to avoid consuming too much memory, but also too much
341  * time when searching on each packet (ideally, we should try instead
342  * to put a limit on the length of the list on each bucket...).
343  *
344  * Each dynamic rule holds a pointer to the parent ipfw rule so
345  * we know what action to perform. Dynamic rules are removed when
346  * the parent rule is deleted. XXX we should make them survive.
347  *
348  * There are some limitations with dynamic rules -- we do not
349  * obey the 'randomized match', and we do not do multiple
350  * passes through the firewall. XXX check the latter!!!
351  *
352  * NOTE about the SHARED LOCKMGR LOCK during dynamic rule looking up:
353  * Only TCP state transition will change dynamic rule's state and ack
354  * sequences, while all packets of one TCP connection only goes through
355  * one TCP thread, so it is safe to use shared lockmgr lock during dynamic
356  * rule looking up.  The keep alive callout uses exclusive lockmgr lock
357  * when it tries to find suitable dynamic rules to send keep alive, so
358  * it will not see half updated state and ack sequences.  Though the expire
359  * field updating looks racy for other protocols, the resolution (second)
360  * of expire field makes this kind of race harmless.
361  * XXX statistics' updating is _not_ MPsafe!!!
362  * XXX once UDP output path is fixed, we could use lockless dynamic rule
363  *     hash table
364  */
365 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
366 static uint32_t dyn_buckets = 256; /* must be power of 2 */
367 static uint32_t curr_dyn_buckets = 256; /* must be power of 2 */
368 static uint32_t dyn_buckets_gen; /* generation of dyn buckets array */
369 static struct lock dyn_lock; /* dynamic rules' hash table lock */
370
371 static struct netmsg ipfw_timeout_netmsg; /* schedule ipfw timeout */
372 static struct callout ipfw_timeout_h;
373
374 /*
375  * Timeouts for various events in handing dynamic rules.
376  */
377 static uint32_t dyn_ack_lifetime = 300;
378 static uint32_t dyn_syn_lifetime = 20;
379 static uint32_t dyn_fin_lifetime = 1;
380 static uint32_t dyn_rst_lifetime = 1;
381 static uint32_t dyn_udp_lifetime = 10;
382 static uint32_t dyn_short_lifetime = 5;
383
384 /*
385  * Keepalives are sent if dyn_keepalive is set. They are sent every
386  * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
387  * seconds of lifetime of a rule.
388  * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
389  * than dyn_keepalive_period.
390  */
391
392 static uint32_t dyn_keepalive_interval = 20;
393 static uint32_t dyn_keepalive_period = 5;
394 static uint32_t dyn_keepalive = 1;      /* do send keepalives */
395
396 static uint32_t dyn_count;              /* # of dynamic rules */
397 static uint32_t dyn_max = 4096;         /* max # of dynamic rules */
398
399 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLTYPE_INT | CTLFLAG_RW,
400     &dyn_buckets, 0, ipfw_sysctl_dyn_buckets, "I", "Number of dyn. buckets");
401 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
402     &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
403 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
404     &dyn_count, 0, "Number of dyn. rules");
405 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
406     &dyn_max, 0, "Max number of dyn. rules");
407 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
408     &static_count, 0, "Number of static rules");
409 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
410     &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
411 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
412     &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
413 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime,
414     CTLTYPE_INT | CTLFLAG_RW, &dyn_fin_lifetime, 0, ipfw_sysctl_dyn_fin, "I",
415     "Lifetime of dyn. rules for fin");
416 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime,
417     CTLTYPE_INT | CTLFLAG_RW, &dyn_rst_lifetime, 0, ipfw_sysctl_dyn_rst, "I",
418     "Lifetime of dyn. rules for rst");
419 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
420     &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
421 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
422     &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
423 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
424     &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
425
426 static ip_fw_chk_t      ipfw_chk;
427 static void             ipfw_tick(void *);
428
429 static __inline int
430 ipfw_free_rule(struct ip_fw *rule)
431 {
432         KASSERT(rule->cpuid == mycpuid, ("rule freed on cpu%d\n", mycpuid));
433         KASSERT(rule->refcnt > 0, ("invalid refcnt %u\n", rule->refcnt));
434         rule->refcnt--;
435         if (rule->refcnt == 0) {
436                 kfree(rule, M_IPFW);
437                 return 1;
438         }
439         return 0;
440 }
441
442 static void
443 ipfw_unref_rule(void *priv)
444 {
445         ipfw_free_rule(priv);
446 #ifdef KLD_MODULE
447         atomic_subtract_int(&ipfw_refcnt, 1);
448 #endif
449 }
450
451 static __inline void
452 ipfw_ref_rule(struct ip_fw *rule)
453 {
454         KASSERT(rule->cpuid == mycpuid, ("rule used on cpu%d\n", mycpuid));
455 #ifdef KLD_MODULE
456         atomic_add_int(&ipfw_refcnt, 1);
457 #endif
458         rule->refcnt++;
459 }
460
461 /*
462  * This macro maps an ip pointer into a layer3 header pointer of type T
463  */
464 #define L3HDR(T, ip) ((T *)((uint32_t *)(ip) + (ip)->ip_hl))
465
466 static __inline int
467 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd)
468 {
469         int type = L3HDR(struct icmp,ip)->icmp_type;
470
471         return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1 << type)));
472 }
473
474 #define TT      ((1 << ICMP_ECHO) | \
475                  (1 << ICMP_ROUTERSOLICIT) | \
476                  (1 << ICMP_TSTAMP) | \
477                  (1 << ICMP_IREQ) | \
478                  (1 << ICMP_MASKREQ))
479
480 static int
481 is_icmp_query(struct ip *ip)
482 {
483         int type = L3HDR(struct icmp, ip)->icmp_type;
484
485         return (type <= ICMP_MAXTYPE && (TT & (1 << type)));
486 }
487
488 #undef TT
489
490 /*
491  * The following checks use two arrays of 8 or 16 bits to store the
492  * bits that we want set or clear, respectively. They are in the
493  * low and high half of cmd->arg1 or cmd->d[0].
494  *
495  * We scan options and store the bits we find set. We succeed if
496  *
497  *      (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
498  *
499  * The code is sometimes optimized not to store additional variables.
500  */
501
502 static int
503 flags_match(ipfw_insn *cmd, uint8_t bits)
504 {
505         u_char want_clear;
506         bits = ~bits;
507
508         if (((cmd->arg1 & 0xff) & bits) != 0)
509                 return 0; /* some bits we want set were clear */
510
511         want_clear = (cmd->arg1 >> 8) & 0xff;
512         if ((want_clear & bits) != want_clear)
513                 return 0; /* some bits we want clear were set */
514         return 1;
515 }
516
517 static int
518 ipopts_match(struct ip *ip, ipfw_insn *cmd)
519 {
520         int optlen, bits = 0;
521         u_char *cp = (u_char *)(ip + 1);
522         int x = (ip->ip_hl << 2) - sizeof(struct ip);
523
524         for (; x > 0; x -= optlen, cp += optlen) {
525                 int opt = cp[IPOPT_OPTVAL];
526
527                 if (opt == IPOPT_EOL)
528                         break;
529
530                 if (opt == IPOPT_NOP) {
531                         optlen = 1;
532                 } else {
533                         optlen = cp[IPOPT_OLEN];
534                         if (optlen <= 0 || optlen > x)
535                                 return 0; /* invalid or truncated */
536                 }
537
538                 switch (opt) {
539                 case IPOPT_LSRR:
540                         bits |= IP_FW_IPOPT_LSRR;
541                         break;
542
543                 case IPOPT_SSRR:
544                         bits |= IP_FW_IPOPT_SSRR;
545                         break;
546
547                 case IPOPT_RR:
548                         bits |= IP_FW_IPOPT_RR;
549                         break;
550
551                 case IPOPT_TS:
552                         bits |= IP_FW_IPOPT_TS;
553                         break;
554
555                 default:
556                         break;
557                 }
558         }
559         return (flags_match(cmd, bits));
560 }
561
562 static int
563 tcpopts_match(struct ip *ip, ipfw_insn *cmd)
564 {
565         int optlen, bits = 0;
566         struct tcphdr *tcp = L3HDR(struct tcphdr,ip);
567         u_char *cp = (u_char *)(tcp + 1);
568         int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
569
570         for (; x > 0; x -= optlen, cp += optlen) {
571                 int opt = cp[0];
572
573                 if (opt == TCPOPT_EOL)
574                         break;
575
576                 if (opt == TCPOPT_NOP) {
577                         optlen = 1;
578                 } else {
579                         optlen = cp[1];
580                         if (optlen <= 0)
581                                 break;
582                 }
583
584                 switch (opt) {
585                 case TCPOPT_MAXSEG:
586                         bits |= IP_FW_TCPOPT_MSS;
587                         break;
588
589                 case TCPOPT_WINDOW:
590                         bits |= IP_FW_TCPOPT_WINDOW;
591                         break;
592
593                 case TCPOPT_SACK_PERMITTED:
594                 case TCPOPT_SACK:
595                         bits |= IP_FW_TCPOPT_SACK;
596                         break;
597
598                 case TCPOPT_TIMESTAMP:
599                         bits |= IP_FW_TCPOPT_TS;
600                         break;
601
602                 case TCPOPT_CC:
603                 case TCPOPT_CCNEW:
604                 case TCPOPT_CCECHO:
605                         bits |= IP_FW_TCPOPT_CC;
606                         break;
607
608                 default:
609                         break;
610                 }
611         }
612         return (flags_match(cmd, bits));
613 }
614
615 static int
616 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
617 {
618         if (ifp == NULL)        /* no iface with this packet, match fails */
619                 return 0;
620
621         /* Check by name or by IP address */
622         if (cmd->name[0] != '\0') { /* match by name */
623                 /* Check name */
624                 if (cmd->p.glob) {
625                         if (kfnmatch(cmd->name, ifp->if_xname, 0) == 0)
626                                 return(1);
627                 } else {
628                         if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
629                                 return(1);
630                 }
631         } else {
632                 struct ifaddr_container *ifac;
633
634                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
635                         struct ifaddr *ia = ifac->ifa;
636
637                         if (ia->ifa_addr == NULL)
638                                 continue;
639                         if (ia->ifa_addr->sa_family != AF_INET)
640                                 continue;
641                         if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
642                             (ia->ifa_addr))->sin_addr.s_addr)
643                                 return(1);      /* match */
644                 }
645         }
646         return(0);      /* no match, fail ... */
647 }
648
649 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
650
651 /*
652  * We enter here when we have a rule with O_LOG.
653  * XXX this function alone takes about 2Kbytes of code!
654  */
655 static void
656 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh,
657          struct mbuf *m, struct ifnet *oif)
658 {
659         char *action;
660         int limit_reached = 0;
661         char action2[40], proto[48], fragment[28];
662
663         fragment[0] = '\0';
664         proto[0] = '\0';
665
666         if (f == NULL) {        /* bogus pkt */
667                 struct ipfw_context *ctx = ipfw_ctx[mycpuid];
668
669                 if (verbose_limit != 0 &&
670                     ctx->ipfw_norule_counter >= verbose_limit)
671                         return;
672                 ctx->ipfw_norule_counter++;
673                 if (ctx->ipfw_norule_counter == verbose_limit)
674                         limit_reached = verbose_limit;
675                 action = "Refuse";
676         } else {        /* O_LOG is the first action, find the real one */
677                 ipfw_insn *cmd = ACTION_PTR(f);
678                 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
679
680                 if (l->max_log != 0 && l->log_left == 0)
681                         return;
682                 l->log_left--;
683                 if (l->log_left == 0)
684                         limit_reached = l->max_log;
685                 cmd += F_LEN(cmd);      /* point to first action */
686                 if (cmd->opcode == O_PROB)
687                         cmd += F_LEN(cmd);
688
689                 action = action2;
690                 switch (cmd->opcode) {
691                 case O_DENY:
692                         action = "Deny";
693                         break;
694
695                 case O_REJECT:
696                         if (cmd->arg1==ICMP_REJECT_RST) {
697                                 action = "Reset";
698                         } else if (cmd->arg1==ICMP_UNREACH_HOST) {
699                                 action = "Reject";
700                         } else {
701                                 ksnprintf(SNPARGS(action2, 0), "Unreach %d",
702                                           cmd->arg1);
703                         }
704                         break;
705
706                 case O_ACCEPT:
707                         action = "Accept";
708                         break;
709
710                 case O_COUNT:
711                         action = "Count";
712                         break;
713
714                 case O_DIVERT:
715                         ksnprintf(SNPARGS(action2, 0), "Divert %d", cmd->arg1);
716                         break;
717
718                 case O_TEE:
719                         ksnprintf(SNPARGS(action2, 0), "Tee %d", cmd->arg1);
720                         break;
721
722                 case O_SKIPTO:
723                         ksnprintf(SNPARGS(action2, 0), "SkipTo %d", cmd->arg1);
724                         break;
725
726                 case O_PIPE:
727                         ksnprintf(SNPARGS(action2, 0), "Pipe %d", cmd->arg1);
728                         break;
729
730                 case O_QUEUE:
731                         ksnprintf(SNPARGS(action2, 0), "Queue %d", cmd->arg1);
732                         break;
733
734                 case O_FORWARD_IP:
735                         {
736                                 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
737                                 int len;
738
739                                 len = ksnprintf(SNPARGS(action2, 0),
740                                                 "Forward to %s",
741                                                 inet_ntoa(sa->sa.sin_addr));
742                                 if (sa->sa.sin_port) {
743                                         ksnprintf(SNPARGS(action2, len), ":%d",
744                                                   sa->sa.sin_port);
745                                 }
746                         }
747                         break;
748
749                 default:
750                         action = "UNKNOWN";
751                         break;
752                 }
753         }
754
755         if (hlen == 0) {        /* non-ip */
756                 ksnprintf(SNPARGS(proto, 0), "MAC");
757         } else {
758                 struct ip *ip = mtod(m, struct ip *);
759                 /* these three are all aliases to the same thing */
760                 struct icmp *const icmp = L3HDR(struct icmp, ip);
761                 struct tcphdr *const tcp = (struct tcphdr *)icmp;
762                 struct udphdr *const udp = (struct udphdr *)icmp;
763
764                 int ip_off, offset, ip_len;
765                 int len;
766
767                 if (eh != NULL) { /* layer 2 packets are as on the wire */
768                         ip_off = ntohs(ip->ip_off);
769                         ip_len = ntohs(ip->ip_len);
770                 } else {
771                         ip_off = ip->ip_off;
772                         ip_len = ip->ip_len;
773                 }
774                 offset = ip_off & IP_OFFMASK;
775                 switch (ip->ip_p) {
776                 case IPPROTO_TCP:
777                         len = ksnprintf(SNPARGS(proto, 0), "TCP %s",
778                                         inet_ntoa(ip->ip_src));
779                         if (offset == 0) {
780                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
781                                           ntohs(tcp->th_sport),
782                                           inet_ntoa(ip->ip_dst),
783                                           ntohs(tcp->th_dport));
784                         } else {
785                                 ksnprintf(SNPARGS(proto, len), " %s",
786                                           inet_ntoa(ip->ip_dst));
787                         }
788                         break;
789
790                 case IPPROTO_UDP:
791                         len = ksnprintf(SNPARGS(proto, 0), "UDP %s",
792                                         inet_ntoa(ip->ip_src));
793                         if (offset == 0) {
794                                 ksnprintf(SNPARGS(proto, len), ":%d %s:%d",
795                                           ntohs(udp->uh_sport),
796                                           inet_ntoa(ip->ip_dst),
797                                           ntohs(udp->uh_dport));
798                         } else {
799                                 ksnprintf(SNPARGS(proto, len), " %s",
800                                           inet_ntoa(ip->ip_dst));
801                         }
802                         break;
803
804                 case IPPROTO_ICMP:
805                         if (offset == 0) {
806                                 len = ksnprintf(SNPARGS(proto, 0),
807                                                 "ICMP:%u.%u ",
808                                                 icmp->icmp_type,
809                                                 icmp->icmp_code);
810                         } else {
811                                 len = ksnprintf(SNPARGS(proto, 0), "ICMP ");
812                         }
813                         len += ksnprintf(SNPARGS(proto, len), "%s",
814                                          inet_ntoa(ip->ip_src));
815                         ksnprintf(SNPARGS(proto, len), " %s",
816                                   inet_ntoa(ip->ip_dst));
817                         break;
818
819                 default:
820                         len = ksnprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p,
821                                         inet_ntoa(ip->ip_src));
822                         ksnprintf(SNPARGS(proto, len), " %s",
823                                   inet_ntoa(ip->ip_dst));
824                         break;
825                 }
826
827                 if (ip_off & (IP_MF | IP_OFFMASK)) {
828                         ksnprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)",
829                                   ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
830                                   offset << 3, (ip_off & IP_MF) ? "+" : "");
831                 }
832         }
833
834         if (oif || m->m_pkthdr.rcvif) {
835                 log(LOG_SECURITY | LOG_INFO,
836                     "ipfw: %d %s %s %s via %s%s\n",
837                     f ? f->rulenum : -1,
838                     action, proto, oif ? "out" : "in",
839                     oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
840                     fragment);
841         } else {
842                 log(LOG_SECURITY | LOG_INFO,
843                     "ipfw: %d %s %s [no if info]%s\n",
844                     f ? f->rulenum : -1,
845                     action, proto, fragment);
846         }
847
848         if (limit_reached) {
849                 log(LOG_SECURITY | LOG_NOTICE,
850                     "ipfw: limit %d reached on entry %d\n",
851                     limit_reached, f ? f->rulenum : -1);
852         }
853 }
854
855 #undef SNPARGS
856
857 /*
858  * IMPORTANT: the hash function for dynamic rules must be commutative
859  * in source and destination (ip,port), because rules are bidirectional
860  * and we want to find both in the same bucket.
861  */
862 static __inline int
863 hash_packet(struct ipfw_flow_id *id)
864 {
865         uint32_t i;
866
867         i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
868         i &= (curr_dyn_buckets - 1);
869         return i;
870 }
871
872 /**
873  * unlink a dynamic rule from a chain. prev is a pointer to
874  * the previous one, q is a pointer to the rule to delete,
875  * head is a pointer to the head of the queue.
876  * Modifies q and potentially also head.
877  */
878 #define UNLINK_DYN_RULE(prev, head, q)                                  \
879 do {                                                                    \
880         ipfw_dyn_rule *old_q = q;                                       \
881                                                                         \
882         /* remove a refcount to the parent */                           \
883         if (q->dyn_type == O_LIMIT)                                     \
884                 q->parent->count--;                                     \
885         DPRINTF("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",    \
886                 q->id.src_ip, q->id.src_port,                           \
887                 q->id.dst_ip, q->id.dst_port, dyn_count - 1);           \
888         if (prev != NULL)                                               \
889                 prev->next = q = q->next;                               \
890         else                                                            \
891                 head = q = q->next;                                     \
892         KASSERT(dyn_count > 0, ("invalid dyn count %u\n", dyn_count));  \
893         dyn_count--;                                                    \
894         kfree(old_q, M_IPFW);                                           \
895 } while (0)
896
897 #define TIME_LEQ(a, b)  ((int)((a) - (b)) <= 0)
898
899 /**
900  * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
901  *
902  * If keep_me == NULL, rules are deleted even if not expired,
903  * otherwise only expired rules are removed.
904  *
905  * The value of the second parameter is also used to point to identify
906  * a rule we absolutely do not want to remove (e.g. because we are
907  * holding a reference to it -- this is the case with O_LIMIT_PARENT
908  * rules). The pointer is only used for comparison, so any non-null
909  * value will do.
910  */
911 static void
912 remove_dyn_rule_locked(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
913 {
914         static uint32_t last_remove = 0; /* XXX */
915
916 #define FORCE   (keep_me == NULL)
917
918         ipfw_dyn_rule *prev, *q;
919         int i, pass = 0, max_pass = 0, unlinked = 0;
920
921         if (ipfw_dyn_v == NULL || dyn_count == 0)
922                 return;
923         /* do not expire more than once per second, it is useless */
924         if (!FORCE && last_remove == time_second)
925                 return;
926         last_remove = time_second;
927
928         /*
929          * because O_LIMIT refer to parent rules, during the first pass only
930          * remove child and mark any pending LIMIT_PARENT, and remove
931          * them in a second pass.
932          */
933 next_pass:
934         for (i = 0; i < curr_dyn_buckets; i++) {
935                 for (prev = NULL, q = ipfw_dyn_v[i]; q;) {
936                         /*
937                          * Logic can become complex here, so we split tests.
938                          */
939                         if (q == keep_me)
940                                 goto next;
941                         if (rule != NULL && rule->stub != q->stub)
942                                 goto next; /* not the one we are looking for */
943                         if (q->dyn_type == O_LIMIT_PARENT) {
944                                 /*
945                                  * handle parent in the second pass,
946                                  * record we need one.
947                                  */
948                                 max_pass = 1;
949                                 if (pass == 0)
950                                         goto next;
951                                 if (FORCE && q->count != 0) {
952                                         /* XXX should not happen! */
953                                         kprintf("OUCH! cannot remove rule, "
954                                                 "count %d\n", q->count);
955                                 }
956                         } else {
957                                 if (!FORCE && !TIME_LEQ(q->expire, time_second))
958                                         goto next;
959                         }
960                         unlinked = 1;
961                         UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
962                         continue;
963 next:
964                         prev = q;
965                         q = q->next;
966                 }
967         }
968         if (pass++ < max_pass)
969                 goto next_pass;
970
971         if (unlinked)
972                 ++dyn_buckets_gen;
973
974 #undef FORCE
975 }
976
977 /**
978  * lookup a dynamic rule.
979  */
980 static ipfw_dyn_rule *
981 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
982                 struct tcphdr *tcp)
983 {
984         /*
985          * stateful ipfw extensions.
986          * Lookup into dynamic session queue
987          */
988 #define MATCH_REVERSE   0
989 #define MATCH_FORWARD   1
990 #define MATCH_NONE      2
991 #define MATCH_UNKNOWN   3
992         int i, dir = MATCH_NONE;
993         ipfw_dyn_rule *prev, *q=NULL;
994
995         if (ipfw_dyn_v == NULL)
996                 goto done;      /* not found */
997
998         i = hash_packet(pkt);
999         for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) {
1000                 if (q->dyn_type == O_LIMIT_PARENT)
1001                         goto next;
1002
1003                 if (TIME_LEQ(q->expire, time_second)) {
1004                         /*
1005                          * Entry expired; skip.
1006                          * Let ipfw_tick() take care of it
1007                          */
1008                         goto next;
1009                 }
1010
1011                 if (pkt->proto == q->id.proto) {
1012                         if (pkt->src_ip == q->id.src_ip &&
1013                             pkt->dst_ip == q->id.dst_ip &&
1014                             pkt->src_port == q->id.src_port &&
1015                             pkt->dst_port == q->id.dst_port) {
1016                                 dir = MATCH_FORWARD;
1017                                 break;
1018                         }
1019                         if (pkt->src_ip == q->id.dst_ip &&
1020                             pkt->dst_ip == q->id.src_ip &&
1021                             pkt->src_port == q->id.dst_port &&
1022                             pkt->dst_port == q->id.src_port) {
1023                                 dir = MATCH_REVERSE;
1024                                 break;
1025                         }
1026                 }
1027 next:
1028                 prev = q;
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\n", 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                 kprintf ("sorry cannot allocate state\n");
1214                 return NULL;
1215         }
1216
1217         /* increase refcount on parent, and set pointer */
1218         if (dyn_type == O_LIMIT) {
1219                 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1220
1221                 if (parent->dyn_type != O_LIMIT_PARENT)
1222                         panic("invalid parent");
1223                 parent->count++;
1224                 r->parent = parent;
1225                 rule = parent->stub->rule[mycpuid];
1226                 KKASSERT(rule->stub == parent->stub);
1227         }
1228         KKASSERT(rule->cpuid == mycpuid && rule->stub != NULL);
1229
1230         r->id = *id;
1231         r->expire = time_second + dyn_syn_lifetime;
1232         r->stub = rule->stub;
1233         r->dyn_type = dyn_type;
1234         r->pcnt = r->bcnt = 0;
1235         r->count = 0;
1236
1237         r->bucket = i;
1238         r->next = ipfw_dyn_v[i];
1239         ipfw_dyn_v[i] = r;
1240         dyn_count++;
1241         dyn_buckets_gen++;
1242         DPRINTF("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1243                 dyn_type,
1244                 r->id.src_ip, r->id.src_port,
1245                 r->id.dst_ip, r->id.dst_port, dyn_count);
1246         return r;
1247 }
1248
1249 /**
1250  * lookup dynamic parent rule using pkt and rule as search keys.
1251  * If the lookup fails, then install one.
1252  */
1253 static ipfw_dyn_rule *
1254 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1255 {
1256         ipfw_dyn_rule *q;
1257         int i;
1258
1259         if (ipfw_dyn_v) {
1260                 i = hash_packet(pkt);
1261                 for (q = ipfw_dyn_v[i]; q != NULL; q = q->next) {
1262                         if (q->dyn_type == O_LIMIT_PARENT &&
1263                             rule->stub == q->stub &&
1264                             pkt->proto == q->id.proto &&
1265                             pkt->src_ip == q->id.src_ip &&
1266                             pkt->dst_ip == q->id.dst_ip &&
1267                             pkt->src_port == q->id.src_port &&
1268                             pkt->dst_port == q->id.dst_port) {
1269                                 q->expire = time_second + dyn_short_lifetime;
1270                                 DPRINTF("lookup_dyn_parent found 0x%p\n", q);
1271                                 return q;
1272                         }
1273                 }
1274         }
1275         return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1276 }
1277
1278 /**
1279  * Install dynamic state for rule type cmd->o.opcode
1280  *
1281  * Returns 1 (failure) if state is not installed because of errors or because
1282  * session limitations are enforced.
1283  */
1284 static int
1285 install_state_locked(struct ip_fw *rule, ipfw_insn_limit *cmd,
1286                      struct ip_fw_args *args)
1287 {
1288         static int last_log; /* XXX */
1289
1290         ipfw_dyn_rule *q;
1291
1292         DPRINTF("-- install state type %d 0x%08x %u -> 0x%08x %u\n",
1293                 cmd->o.opcode,
1294                 args->f_id.src_ip, args->f_id.src_port,
1295                 args->f_id.dst_ip, args->f_id.dst_port);
1296
1297         q = lookup_dyn_rule(&args->f_id, NULL, NULL);
1298         if (q != NULL) { /* should never occur */
1299                 if (last_log != time_second) {
1300                         last_log = time_second;
1301                         kprintf(" install_state: entry already present, done\n");
1302                 }
1303                 return 0;
1304         }
1305
1306         if (dyn_count >= dyn_max) {
1307                 /*
1308                  * Run out of slots, try to remove any expired rule.
1309                  */
1310                 remove_dyn_rule_locked(NULL, (ipfw_dyn_rule *)1);
1311                 if (dyn_count >= dyn_max) {
1312                         if (last_log != time_second) {
1313                                 last_log = time_second;
1314                                 kprintf("install_state: "
1315                                         "Too many dynamic rules\n");
1316                         }
1317                         return 1; /* cannot install, notify caller */
1318                 }
1319         }
1320
1321         switch (cmd->o.opcode) {
1322         case O_KEEP_STATE: /* bidir rule */
1323                 if (add_dyn_rule(&args->f_id, O_KEEP_STATE, rule) == NULL)
1324                         return 1;
1325                 break;
1326
1327         case O_LIMIT: /* limit number of sessions */
1328                 {
1329                         uint16_t limit_mask = cmd->limit_mask;
1330                         struct ipfw_flow_id id;
1331                         ipfw_dyn_rule *parent;
1332
1333                         DPRINTF("installing dyn-limit rule %d\n",
1334                                 cmd->conn_limit);
1335
1336                         id.dst_ip = id.src_ip = 0;
1337                         id.dst_port = id.src_port = 0;
1338                         id.proto = args->f_id.proto;
1339
1340                         if (limit_mask & DYN_SRC_ADDR)
1341                                 id.src_ip = args->f_id.src_ip;
1342                         if (limit_mask & DYN_DST_ADDR)
1343                                 id.dst_ip = args->f_id.dst_ip;
1344                         if (limit_mask & DYN_SRC_PORT)
1345                                 id.src_port = args->f_id.src_port;
1346                         if (limit_mask & DYN_DST_PORT)
1347                                 id.dst_port = args->f_id.dst_port;
1348
1349                         parent = lookup_dyn_parent(&id, rule);
1350                         if (parent == NULL) {
1351                                 kprintf("add parent failed\n");
1352                                 return 1;
1353                         }
1354
1355                         if (parent->count >= cmd->conn_limit) {
1356                                 /*
1357                                  * See if we can remove some expired rule.
1358                                  */
1359                                 remove_dyn_rule_locked(rule, parent);
1360                                 if (parent->count >= cmd->conn_limit) {
1361                                         if (fw_verbose &&
1362                                             last_log != time_second) {
1363                                                 last_log = time_second;
1364                                                 log(LOG_SECURITY | LOG_DEBUG,
1365                                                     "drop session, "
1366                                                     "too many entries\n");
1367                                         }
1368                                         return 1;
1369                                 }
1370                         }
1371                         if (add_dyn_rule(&args->f_id, O_LIMIT,
1372                                          (struct ip_fw *)parent) == NULL)
1373                                 return 1;
1374                 }
1375                 break;
1376         default:
1377                 kprintf("unknown dynamic rule type %u\n", cmd->o.opcode);
1378                 return 1;
1379         }
1380         lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1381         return 0;
1382 }
1383
1384 static int
1385 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1386               struct ip_fw_args *args, int *deny)
1387 {
1388         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1389         uint32_t gen;
1390         int ret = 0;
1391
1392         *deny = 0;
1393         gen = ctx->ipfw_gen;
1394
1395         lockmgr(&dyn_lock, LK_EXCLUSIVE);
1396         if (ctx->ipfw_gen != gen) {
1397                 /* See the comment in lookup_rule() */
1398                 *deny = 1;
1399         } else {
1400                 ret = install_state_locked(rule, cmd, args);
1401         }
1402         lockmgr(&dyn_lock, LK_RELEASE);
1403
1404         return ret;
1405 }
1406
1407 /*
1408  * Transmit a TCP packet, containing either a RST or a keepalive.
1409  * When flags & TH_RST, we are sending a RST packet, because of a
1410  * "reset" action matched the packet.
1411  * Otherwise we are sending a keepalive, and flags & TH_
1412  */
1413 static void
1414 send_pkt(struct ipfw_flow_id *id, uint32_t seq, uint32_t ack, int flags)
1415 {
1416         struct mbuf *m;
1417         struct ip *ip;
1418         struct tcphdr *tcp;
1419         struct route sro;       /* fake route */
1420
1421         MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1422         if (m == NULL)
1423                 return;
1424         m->m_pkthdr.rcvif = NULL;
1425         m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1426         m->m_data += max_linkhdr;
1427
1428         ip = mtod(m, struct ip *);
1429         bzero(ip, m->m_len);
1430         tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1431         ip->ip_p = IPPROTO_TCP;
1432         tcp->th_off = 5;
1433
1434         /*
1435          * Assume we are sending a RST (or a keepalive in the reverse
1436          * direction), swap src and destination addresses and ports.
1437          */
1438         ip->ip_src.s_addr = htonl(id->dst_ip);
1439         ip->ip_dst.s_addr = htonl(id->src_ip);
1440         tcp->th_sport = htons(id->dst_port);
1441         tcp->th_dport = htons(id->src_port);
1442         if (flags & TH_RST) {   /* we are sending a RST */
1443                 if (flags & TH_ACK) {
1444                         tcp->th_seq = htonl(ack);
1445                         tcp->th_ack = htonl(0);
1446                         tcp->th_flags = TH_RST;
1447                 } else {
1448                         if (flags & TH_SYN)
1449                                 seq++;
1450                         tcp->th_seq = htonl(0);
1451                         tcp->th_ack = htonl(seq);
1452                         tcp->th_flags = TH_RST | TH_ACK;
1453                 }
1454         } else {
1455                 /*
1456                  * We are sending a keepalive. flags & TH_SYN determines
1457                  * the direction, forward if set, reverse if clear.
1458                  * NOTE: seq and ack are always assumed to be correct
1459                  * as set by the caller. This may be confusing...
1460                  */
1461                 if (flags & TH_SYN) {
1462                         /*
1463                          * we have to rewrite the correct addresses!
1464                          */
1465                         ip->ip_dst.s_addr = htonl(id->dst_ip);
1466                         ip->ip_src.s_addr = htonl(id->src_ip);
1467                         tcp->th_dport = htons(id->dst_port);
1468                         tcp->th_sport = htons(id->src_port);
1469                 }
1470                 tcp->th_seq = htonl(seq);
1471                 tcp->th_ack = htonl(ack);
1472                 tcp->th_flags = TH_ACK;
1473         }
1474
1475         /*
1476          * set ip_len to the payload size so we can compute
1477          * the tcp checksum on the pseudoheader
1478          * XXX check this, could save a couple of words ?
1479          */
1480         ip->ip_len = htons(sizeof(struct tcphdr));
1481         tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1482
1483         /*
1484          * now fill fields left out earlier
1485          */
1486         ip->ip_ttl = ip_defttl;
1487         ip->ip_len = m->m_pkthdr.len;
1488
1489         bzero(&sro, sizeof(sro));
1490         ip_rtaddr(ip->ip_dst, &sro);
1491
1492         m->m_pkthdr.fw_flags |= IPFW_MBUF_GENERATED;
1493         ip_output(m, NULL, &sro, 0, NULL, NULL);
1494         if (sro.ro_rt)
1495                 RTFREE(sro.ro_rt);
1496 }
1497
1498 /*
1499  * sends a reject message, consuming the mbuf passed as an argument.
1500  */
1501 static void
1502 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len)
1503 {
1504         if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1505                 /* We need the IP header in host order for icmp_error(). */
1506                 if (args->eh != NULL) {
1507                         struct ip *ip = mtod(args->m, struct ip *);
1508
1509                         ip->ip_len = ntohs(ip->ip_len);
1510                         ip->ip_off = ntohs(ip->ip_off);
1511                 }
1512                 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1513         } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1514                 struct tcphdr *const tcp =
1515                     L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1516
1517                 if ((tcp->th_flags & TH_RST) == 0) {
1518                         send_pkt(&args->f_id, ntohl(tcp->th_seq),
1519                                  ntohl(tcp->th_ack), tcp->th_flags | TH_RST);
1520                 }
1521                 m_freem(args->m);
1522         } else {
1523                 m_freem(args->m);
1524         }
1525         args->m = NULL;
1526 }
1527
1528 /**
1529  *
1530  * Given an ip_fw *, lookup_next_rule will return a pointer
1531  * to the next rule, which can be either the jump
1532  * target (for skipto instructions) or the next one in the list (in
1533  * all other cases including a missing jump target).
1534  * The result is also written in the "next_rule" field of the rule.
1535  * Backward jumps are not allowed, so start looking from the next
1536  * rule...
1537  *
1538  * This never returns NULL -- in case we do not have an exact match,
1539  * the next rule is returned. When the ruleset is changed,
1540  * pointers are flushed so we are always correct.
1541  */
1542
1543 static struct ip_fw *
1544 lookup_next_rule(struct ip_fw *me)
1545 {
1546         struct ip_fw *rule = NULL;
1547         ipfw_insn *cmd;
1548
1549         /* look for action, in case it is a skipto */
1550         cmd = ACTION_PTR(me);
1551         if (cmd->opcode == O_LOG)
1552                 cmd += F_LEN(cmd);
1553         if (cmd->opcode == O_SKIPTO) {
1554                 for (rule = me->next; rule; rule = rule->next) {
1555                         if (rule->rulenum >= cmd->arg1)
1556                                 break;
1557                 }
1558         }
1559         if (rule == NULL)                       /* failure or not a skipto */
1560                 rule = me->next;
1561         me->next_rule = rule;
1562         return rule;
1563 }
1564
1565 static int
1566 _ipfw_match_uid(const struct ipfw_flow_id *fid, struct ifnet *oif,
1567                 enum ipfw_opcodes opcode, uid_t uid)
1568 {
1569         struct in_addr src_ip, dst_ip;
1570         struct inpcbinfo *pi;
1571         int wildcard;
1572         struct inpcb *pcb;
1573
1574         if (fid->proto == IPPROTO_TCP) {
1575                 wildcard = 0;
1576                 pi = &tcbinfo[mycpuid];
1577         } else if (fid->proto == IPPROTO_UDP) {
1578                 wildcard = 1;
1579                 pi = &udbinfo;
1580         } else {
1581                 return 0;
1582         }
1583
1584         /*
1585          * Values in 'fid' are in host byte order
1586          */
1587         dst_ip.s_addr = htonl(fid->dst_ip);
1588         src_ip.s_addr = htonl(fid->src_ip);
1589         if (oif) {
1590                 pcb = in_pcblookup_hash(pi,
1591                         dst_ip, htons(fid->dst_port),
1592                         src_ip, htons(fid->src_port),
1593                         wildcard, oif);
1594         } else {
1595                 pcb = in_pcblookup_hash(pi,
1596                         src_ip, htons(fid->src_port),
1597                         dst_ip, htons(fid->dst_port),
1598                         wildcard, NULL);
1599         }
1600         if (pcb == NULL || pcb->inp_socket == NULL)
1601                 return 0;
1602
1603         if (opcode == O_UID) {
1604 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b))
1605                 return !socheckuid(pcb->inp_socket, uid);
1606 #undef socheckuid
1607         } else  {
1608                 return groupmember(uid, pcb->inp_socket->so_cred);
1609         }
1610 }
1611
1612 static int
1613 ipfw_match_uid(const struct ipfw_flow_id *fid, struct ifnet *oif,
1614                enum ipfw_opcodes opcode, uid_t uid, int *deny)
1615 {
1616         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1617         uint32_t gen;
1618         int match = 0;
1619
1620         *deny = 0;
1621         gen = ctx->ipfw_gen;
1622
1623         get_mplock();
1624         if (gen != ctx->ipfw_gen) {
1625                 /* See the comment in lookup_rule() */
1626                 *deny = 1;
1627         } else {
1628                 match = _ipfw_match_uid(fid, oif, opcode, uid);
1629         }
1630         rel_mplock();
1631         return match;
1632 }
1633
1634 /*
1635  * The main check routine for the firewall.
1636  *
1637  * All arguments are in args so we can modify them and return them
1638  * back to the caller.
1639  *
1640  * Parameters:
1641  *
1642  *      args->m (in/out) The packet; we set to NULL when/if we nuke it.
1643  *              Starts with the IP header.
1644  *      args->eh (in)   Mac header if present, or NULL for layer3 packet.
1645  *      args->oif       Outgoing interface, or NULL if packet is incoming.
1646  *              The incoming interface is in the mbuf. (in)
1647  *
1648  *      args->rule      Pointer to the last matching rule (in/out)
1649  *      args->f_id      Addresses grabbed from the packet (out)
1650  *
1651  * Return value:
1652  *
1653  *      If the packet was denied/rejected and has been dropped, *m is equal
1654  *      to NULL upon return.
1655  *
1656  *      IP_FW_DENY      the packet must be dropped.
1657  *      IP_FW_PASS      The packet is to be accepted and routed normally.
1658  *      IP_FW_DIVERT    Divert the packet to port (args->cookie)
1659  *      IP_FW_TEE       Tee the packet to port (args->cookie)
1660  *      IP_FW_DUMMYNET  Send the packet to pipe/queue (args->cookie)
1661  */
1662
1663 static int
1664 ipfw_chk(struct ip_fw_args *args)
1665 {
1666         /*
1667          * Local variables hold state during the processing of a packet.
1668          *
1669          * IMPORTANT NOTE: to speed up the processing of rules, there
1670          * are some assumption on the values of the variables, which
1671          * are documented here. Should you change them, please check
1672          * the implementation of the various instructions to make sure
1673          * that they still work.
1674          *
1675          * args->eh     The MAC header. It is non-null for a layer2
1676          *      packet, it is NULL for a layer-3 packet.
1677          *
1678          * m | args->m  Pointer to the mbuf, as received from the caller.
1679          *      It may change if ipfw_chk() does an m_pullup, or if it
1680          *      consumes the packet because it calls send_reject().
1681          *      XXX This has to change, so that ipfw_chk() never modifies
1682          *      or consumes the buffer.
1683          * ip   is simply an alias of the value of m, and it is kept
1684          *      in sync with it (the packet is  supposed to start with
1685          *      the ip header).
1686          */
1687         struct mbuf *m = args->m;
1688         struct ip *ip = mtod(m, struct ip *);
1689
1690         /*
1691          * oif | args->oif      If NULL, ipfw_chk has been called on the
1692          *      inbound path (ether_input, ip_input).
1693          *      If non-NULL, ipfw_chk has been called on the outbound path
1694          *      (ether_output, ip_output).
1695          */
1696         struct ifnet *oif = args->oif;
1697
1698         struct ip_fw *f = NULL;         /* matching rule */
1699         int retval = IP_FW_PASS;
1700         struct m_tag *mtag;
1701         struct divert_info *divinfo;
1702
1703         /*
1704          * hlen The length of the IPv4 header.
1705          *      hlen >0 means we have an IPv4 packet.
1706          */
1707         u_int hlen = 0;         /* hlen >0 means we have an IP pkt */
1708
1709         /*
1710          * offset       The offset of a fragment. offset != 0 means that
1711          *      we have a fragment at this offset of an IPv4 packet.
1712          *      offset == 0 means that (if this is an IPv4 packet)
1713          *      this is the first or only fragment.
1714          */
1715         u_short offset = 0;
1716
1717         /*
1718          * Local copies of addresses. They are only valid if we have
1719          * an IP packet.
1720          *
1721          * proto        The protocol. Set to 0 for non-ip packets,
1722          *      or to the protocol read from the packet otherwise.
1723          *      proto != 0 means that we have an IPv4 packet.
1724          *
1725          * src_port, dst_port   port numbers, in HOST format. Only
1726          *      valid for TCP and UDP packets.
1727          *
1728          * src_ip, dst_ip       ip addresses, in NETWORK format.
1729          *      Only valid for IPv4 packets.
1730          */
1731         uint8_t proto;
1732         uint16_t src_port = 0, dst_port = 0;    /* NOTE: host format    */
1733         struct in_addr src_ip, dst_ip;          /* NOTE: network format */
1734         uint16_t ip_len = 0;
1735
1736         /*
1737          * dyn_dir = MATCH_UNKNOWN when rules unchecked,
1738          *      MATCH_NONE when checked and not matched (dyn_f = NULL),
1739          *      MATCH_FORWARD or MATCH_REVERSE otherwise (dyn_f != NULL)
1740          */
1741         int dyn_dir = MATCH_UNKNOWN;
1742         struct ip_fw *dyn_f = NULL;
1743         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
1744
1745         if (m->m_pkthdr.fw_flags & IPFW_MBUF_GENERATED)
1746                 return IP_FW_PASS;      /* accept */
1747
1748         if (args->eh == NULL ||         /* layer 3 packet */
1749             (m->m_pkthdr.len >= sizeof(struct ip) &&
1750              ntohs(args->eh->ether_type) == ETHERTYPE_IP))
1751                 hlen = ip->ip_hl << 2;
1752
1753         /*
1754          * Collect parameters into local variables for faster matching.
1755          */
1756         if (hlen == 0) {        /* do not grab addresses for non-ip pkts */
1757                 proto = args->f_id.proto = 0;   /* mark f_id invalid */
1758                 goto after_ip_checks;
1759         }
1760
1761         proto = args->f_id.proto = ip->ip_p;
1762         src_ip = ip->ip_src;
1763         dst_ip = ip->ip_dst;
1764         if (args->eh != NULL) { /* layer 2 packets are as on the wire */
1765                 offset = ntohs(ip->ip_off) & IP_OFFMASK;
1766                 ip_len = ntohs(ip->ip_len);
1767         } else {
1768                 offset = ip->ip_off & IP_OFFMASK;
1769                 ip_len = ip->ip_len;
1770         }
1771
1772 #define PULLUP_TO(len)                          \
1773 do {                                            \
1774         if (m->m_len < (len)) {                 \
1775                 args->m = m = m_pullup(m, (len));\
1776                 if (m == NULL)                  \
1777                         goto pullup_failed;     \
1778                 ip = mtod(m, struct ip *);      \
1779         }                                       \
1780 } while (0)
1781
1782         if (offset == 0) {
1783                 switch (proto) {
1784                 case IPPROTO_TCP:
1785                         {
1786                                 struct tcphdr *tcp;
1787
1788                                 PULLUP_TO(hlen + sizeof(struct tcphdr));
1789                                 tcp = L3HDR(struct tcphdr, ip);
1790                                 dst_port = tcp->th_dport;
1791                                 src_port = tcp->th_sport;
1792                                 args->f_id.flags = tcp->th_flags;
1793                         }
1794                         break;
1795
1796                 case IPPROTO_UDP:
1797                         {
1798                                 struct udphdr *udp;
1799
1800                                 PULLUP_TO(hlen + sizeof(struct udphdr));
1801                                 udp = L3HDR(struct udphdr, ip);
1802                                 dst_port = udp->uh_dport;
1803                                 src_port = udp->uh_sport;
1804                         }
1805                         break;
1806
1807                 case IPPROTO_ICMP:
1808                         PULLUP_TO(hlen + 4);    /* type, code and checksum. */
1809                         args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type;
1810                         break;
1811
1812                 default:
1813                         break;
1814                 }
1815         }
1816
1817 #undef PULLUP_TO
1818
1819         args->f_id.src_ip = ntohl(src_ip.s_addr);
1820         args->f_id.dst_ip = ntohl(dst_ip.s_addr);
1821         args->f_id.src_port = src_port = ntohs(src_port);
1822         args->f_id.dst_port = dst_port = ntohs(dst_port);
1823
1824 after_ip_checks:
1825         if (args->rule) {
1826                 /*
1827                  * Packet has already been tagged. Look for the next rule
1828                  * to restart processing.
1829                  *
1830                  * If fw_one_pass != 0 then just accept it.
1831                  * XXX should not happen here, but optimized out in
1832                  * the caller.
1833                  */
1834                 if (fw_one_pass)
1835                         return IP_FW_PASS;
1836
1837                 /* This rule is being/has been flushed */
1838                 if (ipfw_flushing)
1839                         return IP_FW_DENY;
1840
1841                 KASSERT(args->rule->cpuid == mycpuid,
1842                         ("rule used on cpu%d\n", mycpuid));
1843
1844                 /* This rule was deleted */
1845                 if (args->rule->rule_flags & IPFW_RULE_F_INVALID)
1846                         return IP_FW_DENY;
1847
1848                 f = args->rule->next_rule;
1849                 if (f == NULL)
1850                         f = lookup_next_rule(args->rule);
1851         } else {
1852                 /*
1853                  * Find the starting rule. It can be either the first
1854                  * one, or the one after divert_rule if asked so.
1855                  */
1856                 int skipto;
1857
1858                 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL);
1859                 if (mtag != NULL) {
1860                         divinfo = m_tag_data(mtag);
1861                         skipto = divinfo->skipto;
1862                 } else {
1863                         skipto = 0;
1864                 }
1865
1866                 f = ctx->ipfw_layer3_chain;
1867                 if (args->eh == NULL && skipto != 0) {
1868                         /* No skipto during rule flushing */
1869                         if (ipfw_flushing)
1870                                 return IP_FW_DENY;
1871
1872                         if (skipto >= IPFW_DEFAULT_RULE)
1873                                 return IP_FW_DENY; /* invalid */
1874
1875                         while (f && f->rulenum <= skipto)
1876                                 f = f->next;
1877                         if (f == NULL)  /* drop packet */
1878                                 return IP_FW_DENY;
1879                 } else if (ipfw_flushing) {
1880                         /* Rules are being flushed; skip to default rule */
1881                         f = ctx->ipfw_default_rule;
1882                 }
1883         }
1884         if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL)
1885                 m_tag_delete(m, mtag);
1886
1887         /*
1888          * Now scan the rules, and parse microinstructions for each rule.
1889          */
1890         for (; f; f = f->next) {
1891                 int l, cmdlen;
1892                 ipfw_insn *cmd;
1893                 int skip_or; /* skip rest of OR block */
1894
1895 again:
1896                 if (ctx->ipfw_set_disable & (1 << f->set))
1897                         continue;
1898
1899                 skip_or = 0;
1900                 for (l = f->cmd_len, cmd = f->cmd; l > 0;
1901                      l -= cmdlen, cmd += cmdlen) {
1902                         int match, deny;
1903
1904                         /*
1905                          * check_body is a jump target used when we find a
1906                          * CHECK_STATE, and need to jump to the body of
1907                          * the target rule.
1908                          */
1909
1910 check_body:
1911                         cmdlen = F_LEN(cmd);
1912                         /*
1913                          * An OR block (insn_1 || .. || insn_n) has the
1914                          * F_OR bit set in all but the last instruction.
1915                          * The first match will set "skip_or", and cause
1916                          * the following instructions to be skipped until
1917                          * past the one with the F_OR bit clear.
1918                          */
1919                         if (skip_or) {          /* skip this instruction */
1920                                 if ((cmd->len & F_OR) == 0)
1921                                         skip_or = 0;    /* next one is good */
1922                                 continue;
1923                         }
1924                         match = 0; /* set to 1 if we succeed */
1925
1926                         switch (cmd->opcode) {
1927                         /*
1928                          * The first set of opcodes compares the packet's
1929                          * fields with some pattern, setting 'match' if a
1930                          * match is found. At the end of the loop there is
1931                          * logic to deal with F_NOT and F_OR flags associated
1932                          * with the opcode.
1933                          */
1934                         case O_NOP:
1935                                 match = 1;
1936                                 break;
1937
1938                         case O_FORWARD_MAC:
1939                                 kprintf("ipfw: opcode %d unimplemented\n",
1940                                         cmd->opcode);
1941                                 break;
1942
1943                         case O_GID:
1944                         case O_UID:
1945                                 /*
1946                                  * We only check offset == 0 && proto != 0,
1947                                  * as this ensures that we have an IPv4
1948                                  * packet with the ports info.
1949                                  */
1950                                 if (offset!=0)
1951                                         break;
1952
1953                                 match = ipfw_match_uid(&args->f_id, oif,
1954                                         cmd->opcode,
1955                                         (uid_t)((ipfw_insn_u32 *)cmd)->d[0],
1956                                         &deny);
1957                                 if (deny)
1958                                         return IP_FW_DENY;
1959                                 break;
1960
1961                         case O_RECV:
1962                                 match = iface_match(m->m_pkthdr.rcvif,
1963                                     (ipfw_insn_if *)cmd);
1964                                 break;
1965
1966                         case O_XMIT:
1967                                 match = iface_match(oif, (ipfw_insn_if *)cmd);
1968                                 break;
1969
1970                         case O_VIA:
1971                                 match = iface_match(oif ? oif :
1972                                     m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
1973                                 break;
1974
1975                         case O_MACADDR2:
1976                                 if (args->eh != NULL) { /* have MAC header */
1977                                         uint32_t *want = (uint32_t *)
1978                                                 ((ipfw_insn_mac *)cmd)->addr;
1979                                         uint32_t *mask = (uint32_t *)
1980                                                 ((ipfw_insn_mac *)cmd)->mask;
1981                                         uint32_t *hdr = (uint32_t *)args->eh;
1982
1983                                         match =
1984                                         (want[0] == (hdr[0] & mask[0]) &&
1985                                          want[1] == (hdr[1] & mask[1]) &&
1986                                          want[2] == (hdr[2] & mask[2]));
1987                                 }
1988                                 break;
1989
1990                         case O_MAC_TYPE:
1991                                 if (args->eh != NULL) {
1992                                         uint16_t t =
1993                                             ntohs(args->eh->ether_type);
1994                                         uint16_t *p =
1995                                             ((ipfw_insn_u16 *)cmd)->ports;
1996                                         int i;
1997
1998                                         /* Special vlan handling */
1999                                         if (m->m_flags & M_VLANTAG)
2000                                                 t = ETHERTYPE_VLAN;
2001
2002                                         for (i = cmdlen - 1; !match && i > 0;
2003                                              i--, p += 2) {
2004                                                 match =
2005                                                 (t >= p[0] && t <= p[1]);
2006                                         }
2007                                 }
2008                                 break;
2009
2010                         case O_FRAG:
2011                                 match = (hlen > 0 && offset != 0);
2012                                 break;
2013
2014                         case O_IN:      /* "out" is "not in" */
2015                                 match = (oif == NULL);
2016                                 break;
2017
2018                         case O_LAYER2:
2019                                 match = (args->eh != NULL);
2020                                 break;
2021
2022                         case O_PROTO:
2023                                 /*
2024                                  * We do not allow an arg of 0 so the
2025                                  * check of "proto" only suffices.
2026                                  */
2027                                 match = (proto == cmd->arg1);
2028                                 break;
2029
2030                         case O_IP_SRC:
2031                                 match = (hlen > 0 &&
2032                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2033                                     src_ip.s_addr);
2034                                 break;
2035
2036                         case O_IP_SRC_MASK:
2037                                 match = (hlen > 0 &&
2038                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2039                                      (src_ip.s_addr &
2040                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
2041                                 break;
2042
2043                         case O_IP_SRC_ME:
2044                                 if (hlen > 0) {
2045                                         struct ifnet *tif;
2046
2047                                         tif = INADDR_TO_IFP(&src_ip);
2048                                         match = (tif != NULL);
2049                                 }
2050                                 break;
2051
2052                         case O_IP_DST_SET:
2053                         case O_IP_SRC_SET:
2054                                 if (hlen > 0) {
2055                                         uint32_t *d = (uint32_t *)(cmd + 1);
2056                                         uint32_t addr =
2057                                             cmd->opcode == O_IP_DST_SET ?
2058                                                 args->f_id.dst_ip :
2059                                                 args->f_id.src_ip;
2060
2061                                         if (addr < d[0])
2062                                                 break;
2063                                         addr -= d[0]; /* subtract base */
2064                                         match =
2065                                         (addr < cmd->arg1) &&
2066                                          (d[1 + (addr >> 5)] &
2067                                           (1 << (addr & 0x1f)));
2068                                 }
2069                                 break;
2070
2071                         case O_IP_DST:
2072                                 match = (hlen > 0 &&
2073                                     ((ipfw_insn_ip *)cmd)->addr.s_addr ==
2074                                     dst_ip.s_addr);
2075                                 break;
2076
2077                         case O_IP_DST_MASK:
2078                                 match = (hlen > 0) &&
2079                                     (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2080                                      (dst_ip.s_addr &
2081                                      ((ipfw_insn_ip *)cmd)->mask.s_addr));
2082                                 break;
2083
2084                         case O_IP_DST_ME:
2085                                 if (hlen > 0) {
2086                                         struct ifnet *tif;
2087
2088                                         tif = INADDR_TO_IFP(&dst_ip);
2089                                         match = (tif != NULL);
2090                                 }
2091                                 break;
2092
2093                         case O_IP_SRCPORT:
2094                         case O_IP_DSTPORT:
2095                                 /*
2096                                  * offset == 0 && proto != 0 is enough
2097                                  * to guarantee that we have an IPv4
2098                                  * packet with port info.
2099                                  */
2100                                 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2101                                     && offset == 0) {
2102                                         uint16_t x =
2103                                             (cmd->opcode == O_IP_SRCPORT) ?
2104                                                 src_port : dst_port ;
2105                                         uint16_t *p =
2106                                             ((ipfw_insn_u16 *)cmd)->ports;
2107                                         int i;
2108
2109                                         for (i = cmdlen - 1; !match && i > 0;
2110                                              i--, p += 2) {
2111                                                 match =
2112                                                 (x >= p[0] && x <= p[1]);
2113                                         }
2114                                 }
2115                                 break;
2116
2117                         case O_ICMPTYPE:
2118                                 match = (offset == 0 && proto==IPPROTO_ICMP &&
2119                                     icmptype_match(ip, (ipfw_insn_u32 *)cmd));
2120                                 break;
2121
2122                         case O_IPOPT:
2123                                 match = (hlen > 0 && ipopts_match(ip, cmd));
2124                                 break;
2125
2126                         case O_IPVER:
2127                                 match = (hlen > 0 && cmd->arg1 == ip->ip_v);
2128                                 break;
2129
2130                         case O_IPTTL:
2131                                 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl);
2132                                 break;
2133
2134                         case O_IPID:
2135                                 match = (hlen > 0 &&
2136                                     cmd->arg1 == ntohs(ip->ip_id));
2137                                 break;
2138
2139                         case O_IPLEN:
2140                                 match = (hlen > 0 && cmd->arg1 == ip_len);
2141                                 break;
2142
2143                         case O_IPPRECEDENCE:
2144                                 match = (hlen > 0 &&
2145                                     (cmd->arg1 == (ip->ip_tos & 0xe0)));
2146                                 break;
2147
2148                         case O_IPTOS:
2149                                 match = (hlen > 0 &&
2150                                     flags_match(cmd, ip->ip_tos));
2151                                 break;
2152
2153                         case O_TCPFLAGS:
2154                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2155                                     flags_match(cmd,
2156                                         L3HDR(struct tcphdr,ip)->th_flags));
2157                                 break;
2158
2159                         case O_TCPOPTS:
2160                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2161                                     tcpopts_match(ip, cmd));
2162                                 break;
2163
2164                         case O_TCPSEQ:
2165                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2166                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
2167                                         L3HDR(struct tcphdr,ip)->th_seq);
2168                                 break;
2169
2170                         case O_TCPACK:
2171                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2172                                     ((ipfw_insn_u32 *)cmd)->d[0] ==
2173                                         L3HDR(struct tcphdr,ip)->th_ack);
2174                                 break;
2175
2176                         case O_TCPWIN:
2177                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2178                                     cmd->arg1 ==
2179                                         L3HDR(struct tcphdr,ip)->th_win);
2180                                 break;
2181
2182                         case O_ESTAB:
2183                                 /* reject packets which have SYN only */
2184                                 /* XXX should i also check for TH_ACK ? */
2185                                 match = (proto == IPPROTO_TCP && offset == 0 &&
2186                                     (L3HDR(struct tcphdr,ip)->th_flags &
2187                                      (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2188                                 break;
2189
2190                         case O_LOG:
2191                                 if (fw_verbose)
2192                                         ipfw_log(f, hlen, args->eh, m, oif);
2193                                 match = 1;
2194                                 break;
2195
2196                         case O_PROB:
2197                                 match = (krandom() <
2198                                         ((ipfw_insn_u32 *)cmd)->d[0]);
2199                                 break;
2200
2201                         /*
2202                          * The second set of opcodes represents 'actions',
2203                          * i.e. the terminal part of a rule once the packet
2204                          * matches all previous patterns.
2205                          * Typically there is only one action for each rule,
2206                          * and the opcode is stored at the end of the rule
2207                          * (but there are exceptions -- see below).
2208                          *
2209                          * In general, here we set retval and terminate the
2210                          * outer loop (would be a 'break 3' in some language,
2211                          * but we need to do a 'goto done').
2212                          *
2213                          * Exceptions:
2214                          * O_COUNT and O_SKIPTO actions:
2215                          *   instead of terminating, we jump to the next rule
2216                          *   ('goto next_rule', equivalent to a 'break 2'),
2217                          *   or to the SKIPTO target ('goto again' after
2218                          *   having set f, cmd and l), respectively.
2219                          *
2220                          * O_LIMIT and O_KEEP_STATE: these opcodes are
2221                          *   not real 'actions', and are stored right
2222                          *   before the 'action' part of the rule.
2223                          *   These opcodes try to install an entry in the
2224                          *   state tables; if successful, we continue with
2225                          *   the next opcode (match=1; break;), otherwise
2226                          *   the packet must be dropped ('goto done' after
2227                          *   setting retval).  If static rules are changed
2228                          *   during the state installation, the packet will
2229                          *   be dropped and rule's stats will not beupdated
2230                          *   ('return IP_FW_DENY').
2231                          *
2232                          * O_PROBE_STATE and O_CHECK_STATE: these opcodes
2233                          *   cause a lookup of the state table, and a jump
2234                          *   to the 'action' part of the parent rule
2235                          *   ('goto check_body') if an entry is found, or
2236                          *   (CHECK_STATE only) a jump to the next rule if
2237                          *   the entry is not found ('goto next_rule').
2238                          *   The result of the lookup is cached to make
2239                          *   further instances of these opcodes are
2240                          *   effectively NOPs.  If static rules are changed
2241                          *   during the state looking up, the packet will
2242                          *   be dropped and rule's stats will not be updated
2243                          *   ('return IP_FW_DENY').
2244                          */
2245                         case O_LIMIT:
2246                         case O_KEEP_STATE:
2247                                 if (!(f->rule_flags & IPFW_RULE_F_STATE)) {
2248                                         kprintf("%s rule (%d) is not ready "
2249                                                 "on cpu%d\n",
2250                                                 cmd->opcode == O_LIMIT ?
2251                                                 "limit" : "keep state",
2252                                                 f->rulenum, f->cpuid);
2253                                         goto next_rule;
2254                                 }
2255                                 if (install_state(f,
2256                                     (ipfw_insn_limit *)cmd, args, &deny)) {
2257                                         if (deny)
2258                                                 return IP_FW_DENY;
2259
2260                                         retval = IP_FW_DENY;
2261                                         goto done; /* error/limit violation */
2262                                 }
2263                                 if (deny)
2264                                         return IP_FW_DENY;
2265                                 match = 1;
2266                                 break;
2267
2268                         case O_PROBE_STATE:
2269                         case O_CHECK_STATE:
2270                                 /*
2271                                  * dynamic rules are checked at the first
2272                                  * keep-state or check-state occurrence,
2273                                  * with the result being stored in dyn_dir.
2274                                  * The compiler introduces a PROBE_STATE
2275                                  * instruction for us when we have a
2276                                  * KEEP_STATE (because PROBE_STATE needs
2277                                  * to be run first).
2278                                  */
2279                                 if (dyn_dir == MATCH_UNKNOWN) {
2280                                         dyn_f = lookup_rule(&args->f_id,
2281                                                 &dyn_dir,
2282                                                 proto == IPPROTO_TCP ?
2283                                                 L3HDR(struct tcphdr, ip) : NULL,
2284                                                 ip_len, &deny);
2285                                         if (deny)
2286                                                 return IP_FW_DENY;
2287                                         if (dyn_f != NULL) {
2288                                                 /*
2289                                                  * Found a rule from a dynamic
2290                                                  * entry; jump to the 'action'
2291                                                  * part of the rule.
2292                                                  */
2293                                                 f = dyn_f;
2294                                                 cmd = ACTION_PTR(f);
2295                                                 l = f->cmd_len - f->act_ofs;
2296                                                 goto check_body;
2297                                         }
2298                                 }
2299                                 /*
2300                                  * Dynamic entry not found. If CHECK_STATE,
2301                                  * skip to next rule, if PROBE_STATE just
2302                                  * ignore and continue with next opcode.
2303                                  */
2304                                 if (cmd->opcode == O_CHECK_STATE)
2305                                         goto next_rule;
2306                                 else if (!(f->rule_flags & IPFW_RULE_F_STATE))
2307                                         goto next_rule; /* not ready yet */
2308                                 match = 1;
2309                                 break;
2310
2311                         case O_ACCEPT:
2312                                 retval = IP_FW_PASS;    /* accept */
2313                                 goto done;
2314
2315                         case O_PIPE:
2316                         case O_QUEUE:
2317                                 args->rule = f; /* report matching rule */
2318                                 args->cookie = cmd->arg1;
2319                                 retval = IP_FW_DUMMYNET;
2320                                 goto done;
2321
2322                         case O_DIVERT:
2323                         case O_TEE:
2324                                 if (args->eh) /* not on layer 2 */
2325                                         break;
2326
2327                                 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT,
2328                                                  sizeof(*divinfo), MB_DONTWAIT);
2329                                 if (mtag == NULL) {
2330                                         retval = IP_FW_DENY;
2331                                         goto done;
2332                                 }
2333                                 divinfo = m_tag_data(mtag);
2334
2335                                 divinfo->skipto = f->rulenum;
2336                                 divinfo->port = cmd->arg1;
2337                                 divinfo->tee = (cmd->opcode == O_TEE);
2338                                 m_tag_prepend(m, mtag);
2339
2340                                 args->cookie = cmd->arg1;
2341                                 retval = (cmd->opcode == O_DIVERT) ?
2342                                          IP_FW_DIVERT : IP_FW_TEE;
2343                                 goto done;
2344
2345                         case O_COUNT:
2346                         case O_SKIPTO:
2347                                 f->pcnt++;      /* update stats */
2348                                 f->bcnt += ip_len;
2349                                 f->timestamp = time_second;
2350                                 if (cmd->opcode == O_COUNT)
2351                                         goto next_rule;
2352                                 /* handle skipto */
2353                                 if (f->next_rule == NULL)
2354                                         lookup_next_rule(f);
2355                                 f = f->next_rule;
2356                                 goto again;
2357
2358                         case O_REJECT:
2359                                 /*
2360                                  * Drop the packet and send a reject notice
2361                                  * if the packet is not ICMP (or is an ICMP
2362                                  * query), and it is not multicast/broadcast.
2363                                  */
2364                                 if (hlen > 0 &&
2365                                     (proto != IPPROTO_ICMP ||
2366                                      is_icmp_query(ip)) &&
2367                                     !(m->m_flags & (M_BCAST|M_MCAST)) &&
2368                                     !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
2369                                         /*
2370                                          * Update statistics before the possible
2371                                          * blocking 'send_reject'
2372                                          */
2373                                         f->pcnt++;
2374                                         f->bcnt += ip_len;
2375                                         f->timestamp = time_second;
2376
2377                                         send_reject(args, cmd->arg1,
2378                                             offset,ip_len);
2379                                         m = args->m;
2380
2381                                         /*
2382                                          * Return directly here, rule stats
2383                                          * have been updated above.
2384                                          */
2385                                         return IP_FW_DENY;
2386                                 }
2387                                 /* FALLTHROUGH */
2388                         case O_DENY:
2389                                 retval = IP_FW_DENY;
2390                                 goto done;
2391
2392                         case O_FORWARD_IP:
2393                                 if (args->eh)   /* not valid on layer2 pkts */
2394                                         break;
2395                                 if (!dyn_f || dyn_dir == MATCH_FORWARD) {
2396                                         struct sockaddr_in *sin;
2397
2398                                         mtag = m_tag_get(PACKET_TAG_IPFORWARD,
2399                                                sizeof(*sin), MB_DONTWAIT);
2400                                         if (mtag == NULL) {
2401                                                 retval = IP_FW_DENY;
2402                                                 goto done;
2403                                         }
2404                                         sin = m_tag_data(mtag);
2405
2406                                         /* Structure copy */
2407                                         *sin = ((ipfw_insn_sa *)cmd)->sa;
2408
2409                                         m_tag_prepend(m, mtag);
2410                                         m->m_pkthdr.fw_flags |=
2411                                                 IPFORWARD_MBUF_TAGGED;
2412                                 }
2413                                 retval = IP_FW_PASS;
2414                                 goto done;
2415
2416                         default:
2417                                 panic("-- unknown opcode %d\n", cmd->opcode);
2418                         } /* end of switch() on opcodes */
2419
2420                         if (cmd->len & F_NOT)
2421                                 match = !match;
2422
2423                         if (match) {
2424                                 if (cmd->len & F_OR)
2425                                         skip_or = 1;
2426                         } else {
2427                                 if (!(cmd->len & F_OR)) /* not an OR block, */
2428                                         break;          /* try next rule    */
2429                         }
2430
2431                 }       /* end of inner for, scan opcodes */
2432
2433 next_rule:;             /* try next rule                */
2434
2435         }               /* end of outer for, scan rules */
2436         kprintf("+++ ipfw: ouch!, skip past end of rules, denying packet\n");
2437         return IP_FW_DENY;
2438
2439 done:
2440         /* Update statistics */
2441         f->pcnt++;
2442         f->bcnt += ip_len;
2443         f->timestamp = time_second;
2444         return retval;
2445
2446 pullup_failed:
2447         if (fw_verbose)
2448                 kprintf("pullup failed\n");
2449         return IP_FW_DENY;
2450 }
2451
2452 static void
2453 ipfw_dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa)
2454 {
2455         struct m_tag *mtag;
2456         struct dn_pkt *pkt;
2457         ipfw_insn *cmd;
2458         const struct ipfw_flow_id *id;
2459         struct dn_flow_id *fid;
2460
2461         M_ASSERTPKTHDR(m);
2462
2463         mtag = m_tag_get(PACKET_TAG_DUMMYNET, sizeof(*pkt), MB_DONTWAIT);
2464         if (mtag == NULL) {
2465                 m_freem(m);
2466                 return;
2467         }
2468         m_tag_prepend(m, mtag);
2469
2470         pkt = m_tag_data(mtag);
2471         bzero(pkt, sizeof(*pkt));
2472
2473         cmd = fwa->rule->cmd + fwa->rule->act_ofs;
2474         if (cmd->opcode == O_LOG)
2475                 cmd += F_LEN(cmd);
2476         KASSERT(cmd->opcode == O_PIPE || cmd->opcode == O_QUEUE,
2477                 ("Rule is not PIPE or QUEUE, opcode %d\n", cmd->opcode));
2478
2479         pkt->dn_m = m;
2480         pkt->dn_flags = (dir & DN_FLAGS_DIR_MASK);
2481         pkt->ifp = fwa->oif;
2482         pkt->pipe_nr = pipe_nr;
2483
2484         pkt->cpuid = mycpuid;
2485         pkt->msgport = curnetport;
2486
2487         id = &fwa->f_id;
2488         fid = &pkt->id;
2489         fid->fid_dst_ip = id->dst_ip;
2490         fid->fid_src_ip = id->src_ip;
2491         fid->fid_dst_port = id->dst_port;
2492         fid->fid_src_port = id->src_port;
2493         fid->fid_proto = id->proto;
2494         fid->fid_flags = id->flags;
2495
2496         ipfw_ref_rule(fwa->rule);
2497         pkt->dn_priv = fwa->rule;
2498         pkt->dn_unref_priv = ipfw_unref_rule;
2499
2500         if (cmd->opcode == O_PIPE)
2501                 pkt->dn_flags |= DN_FLAGS_IS_PIPE;
2502
2503         m->m_pkthdr.fw_flags |= DUMMYNET_MBUF_TAGGED;
2504 }
2505
2506 /*
2507  * When a rule is added/deleted, clear the next_rule pointers in all rules.
2508  * These will be reconstructed on the fly as packets are matched.
2509  * Must be called at splimp().
2510  */
2511 static void
2512 ipfw_flush_rule_ptrs(struct ipfw_context *ctx)
2513 {
2514         struct ip_fw *rule;
2515
2516         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
2517                 rule->next_rule = NULL;
2518 }
2519
2520 static __inline void
2521 ipfw_inc_static_count(struct ip_fw *rule)
2522 {
2523         /* Static rule's counts are updated only on CPU0 */
2524         KKASSERT(mycpuid == 0);
2525
2526         static_count++;
2527         static_ioc_len += IOC_RULESIZE(rule);
2528 }
2529
2530 static __inline void
2531 ipfw_dec_static_count(struct ip_fw *rule)
2532 {
2533         int l = IOC_RULESIZE(rule);
2534
2535         /* Static rule's counts are updated only on CPU0 */
2536         KKASSERT(mycpuid == 0);
2537
2538         KASSERT(static_count > 0, ("invalid static count %u\n", static_count));
2539         static_count--;
2540
2541         KASSERT(static_ioc_len >= l,
2542                 ("invalid static len %u\n", static_ioc_len));
2543         static_ioc_len -= l;
2544 }
2545
2546 static void
2547 ipfw_link_sibling(struct netmsg_ipfw *fwmsg, struct ip_fw *rule)
2548 {
2549         if (fwmsg->sibling != NULL) {
2550                 KKASSERT(mycpuid > 0 && fwmsg->sibling->cpuid == mycpuid - 1);
2551                 fwmsg->sibling->sibling = rule;
2552         }
2553         fwmsg->sibling = rule;
2554 }
2555
2556 static struct ip_fw *
2557 ipfw_create_rule(const struct ipfw_ioc_rule *ioc_rule, struct ip_fw_stub *stub)
2558 {
2559         struct ip_fw *rule;
2560
2561         rule = kmalloc(RULESIZE(ioc_rule), M_IPFW, M_WAITOK | M_ZERO);
2562
2563         rule->act_ofs = ioc_rule->act_ofs;
2564         rule->cmd_len = ioc_rule->cmd_len;
2565         rule->rulenum = ioc_rule->rulenum;
2566         rule->set = ioc_rule->set;
2567         rule->usr_flags = ioc_rule->usr_flags;
2568
2569         bcopy(ioc_rule->cmd, rule->cmd, rule->cmd_len * 4 /* XXX */);
2570
2571         rule->refcnt = 1;
2572         rule->cpuid = mycpuid;
2573
2574         rule->stub = stub;
2575         if (stub != NULL)
2576                 stub->rule[mycpuid] = rule;
2577
2578         return rule;
2579 }
2580
2581 static void
2582 ipfw_add_rule_dispatch(struct netmsg *nmsg)
2583 {
2584         struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg;
2585         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2586         struct ip_fw *rule;
2587
2588         rule = ipfw_create_rule(fwmsg->ioc_rule, fwmsg->stub);
2589
2590         /*
2591          * Bump generation after ipfw_create_rule(),
2592          * since this function is blocking
2593          */
2594         ctx->ipfw_gen++;
2595
2596         /*
2597          * Insert rule into the pre-determined position
2598          */
2599         if (fwmsg->prev_rule != NULL) {
2600                 struct ip_fw *prev, *next;
2601
2602                 prev = fwmsg->prev_rule;
2603                 KKASSERT(prev->cpuid == mycpuid);
2604
2605                 next = fwmsg->next_rule;
2606                 KKASSERT(next->cpuid == mycpuid);
2607
2608                 rule->next = next;
2609                 prev->next = rule;
2610
2611                 /*
2612                  * Move to the position on the next CPU
2613                  * before the msg is forwarded.
2614                  */
2615                 fwmsg->prev_rule = prev->sibling;
2616                 fwmsg->next_rule = next->sibling;
2617         } else {
2618                 KKASSERT(fwmsg->next_rule == NULL);
2619                 rule->next = ctx->ipfw_layer3_chain;
2620                 ctx->ipfw_layer3_chain = rule;
2621         }
2622
2623         /* Link rule CPU sibling */
2624         ipfw_link_sibling(fwmsg, rule);
2625
2626         ipfw_flush_rule_ptrs(ctx);
2627
2628         if (mycpuid == 0) {
2629                 /* Statistics only need to be updated once */
2630                 ipfw_inc_static_count(rule);
2631
2632                 /* Return the rule on CPU0 */
2633                 nmsg->nm_lmsg.u.ms_resultp = rule;
2634         }
2635
2636         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
2637 }
2638
2639 static void
2640 ipfw_enable_state_dispatch(struct netmsg *nmsg)
2641 {
2642         struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
2643         struct ip_fw *rule = lmsg->u.ms_resultp;
2644         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2645
2646         ctx->ipfw_gen++;
2647
2648         KKASSERT(rule->cpuid == mycpuid);
2649         KKASSERT(rule->stub != NULL && rule->stub->rule[mycpuid] == rule);
2650         KKASSERT(!(rule->rule_flags & IPFW_RULE_F_STATE));
2651         rule->rule_flags |= IPFW_RULE_F_STATE;
2652         lmsg->u.ms_resultp = rule->sibling;
2653
2654         ifnet_forwardmsg(lmsg, mycpuid + 1);
2655 }
2656
2657 /*
2658  * Add a new rule to the list.  Copy the rule into a malloc'ed area,
2659  * then possibly create a rule number and add the rule to the list.
2660  * Update the rule_number in the input struct so the caller knows
2661  * it as well.
2662  */
2663 static void
2664 ipfw_add_rule(struct ipfw_ioc_rule *ioc_rule, uint32_t rule_flags)
2665 {
2666         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2667         struct netmsg_ipfw fwmsg;
2668         struct netmsg *nmsg;
2669         struct ip_fw *f, *prev, *rule;
2670         struct ip_fw_stub *stub;
2671
2672         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2673
2674         /*
2675          * If rulenum is 0, find highest numbered rule before the
2676          * default rule, and add rule number incremental step.
2677          */
2678         if (ioc_rule->rulenum == 0) {
2679                 int step = autoinc_step;
2680
2681                 KKASSERT(step >= IPFW_AUTOINC_STEP_MIN &&
2682                          step <= IPFW_AUTOINC_STEP_MAX);
2683
2684                 /*
2685                  * Locate the highest numbered rule before default
2686                  */
2687                 for (f = ctx->ipfw_layer3_chain; f; f = f->next) {
2688                         if (f->rulenum == IPFW_DEFAULT_RULE)
2689                                 break;
2690                         ioc_rule->rulenum = f->rulenum;
2691                 }
2692                 if (ioc_rule->rulenum < IPFW_DEFAULT_RULE - step)
2693                         ioc_rule->rulenum += step;
2694         }
2695         KASSERT(ioc_rule->rulenum != IPFW_DEFAULT_RULE &&
2696                 ioc_rule->rulenum != 0,
2697                 ("invalid rule num %d\n", ioc_rule->rulenum));
2698
2699         /*
2700          * Now find the right place for the new rule in the sorted list.
2701          */
2702         for (prev = NULL, f = ctx->ipfw_layer3_chain; f;
2703              prev = f, f = f->next) {
2704                 if (f->rulenum > ioc_rule->rulenum) {
2705                         /* Found the location */
2706                         break;
2707                 }
2708         }
2709         KASSERT(f != NULL, ("no default rule?!\n"));
2710
2711         if (rule_flags & IPFW_RULE_F_STATE) {
2712                 int size;
2713
2714                 /*
2715                  * If the new rule will create states, then allocate
2716                  * a rule stub, which will be referenced by states
2717                  * (dyn rules)
2718                  */
2719                 size = sizeof(*stub) + ((ncpus - 1) * sizeof(struct ip_fw *));
2720                 stub = kmalloc(size, M_IPFW, M_WAITOK | M_ZERO);
2721         } else {
2722                 stub = NULL;
2723         }
2724
2725         /*
2726          * Duplicate the rule onto each CPU.
2727          * The rule duplicated on CPU0 will be returned.
2728          */
2729         bzero(&fwmsg, sizeof(fwmsg));
2730         nmsg = &fwmsg.nmsg;
2731         netmsg_init(nmsg, &curthread->td_msgport, 0, ipfw_add_rule_dispatch);
2732         fwmsg.ioc_rule = ioc_rule;
2733         fwmsg.prev_rule = prev;
2734         fwmsg.next_rule = prev == NULL ? NULL : f;
2735         fwmsg.stub = stub;
2736
2737         ifnet_domsg(&nmsg->nm_lmsg, 0);
2738         KKASSERT(fwmsg.prev_rule == NULL && fwmsg.next_rule == NULL);
2739
2740         rule = nmsg->nm_lmsg.u.ms_resultp;
2741         KKASSERT(rule != NULL && rule->cpuid == mycpuid);
2742
2743         if (rule_flags & IPFW_RULE_F_STATE) {
2744                 /*
2745                  * Turn on state flag, _after_ everything on all
2746                  * CPUs have been setup.
2747                  */
2748                 bzero(nmsg, sizeof(*nmsg));
2749                 netmsg_init(nmsg, &curthread->td_msgport, 0,
2750                             ipfw_enable_state_dispatch);
2751                 nmsg->nm_lmsg.u.ms_resultp = rule;
2752
2753                 ifnet_domsg(&nmsg->nm_lmsg, 0);
2754                 KKASSERT(nmsg->nm_lmsg.u.ms_resultp == NULL);
2755         }
2756
2757         DPRINTF("++ installed rule %d, static count now %d\n",
2758                 rule->rulenum, static_count);
2759 }
2760
2761 /**
2762  * Free storage associated with a static rule (including derived
2763  * dynamic rules).
2764  * The caller is in charge of clearing rule pointers to avoid
2765  * dangling pointers.
2766  * @return a pointer to the next entry.
2767  * Arguments are not checked, so they better be correct.
2768  * Must be called at splimp().
2769  */
2770 static struct ip_fw *
2771 ipfw_delete_rule(struct ipfw_context *ctx,
2772                  struct ip_fw *prev, struct ip_fw *rule)
2773 {
2774         struct ip_fw *n;
2775         struct ip_fw_stub *stub;
2776
2777         ctx->ipfw_gen++;
2778
2779         /* STATE flag should have been cleared before we reach here */
2780         KKASSERT((rule->rule_flags & IPFW_RULE_F_STATE) == 0);
2781
2782         stub = rule->stub;
2783         n = rule->next;
2784         if (prev == NULL)
2785                 ctx->ipfw_layer3_chain = n;
2786         else
2787                 prev->next = n;
2788
2789         /* Mark the rule as invalid */
2790         rule->rule_flags |= IPFW_RULE_F_INVALID;
2791         rule->next_rule = NULL;
2792         rule->sibling = NULL;
2793         rule->stub = NULL;
2794 #ifdef foo
2795         /* Don't reset cpuid here; keep various assertion working */
2796         rule->cpuid = -1;
2797 #endif
2798
2799         /* Statistics only need to be updated once */
2800         if (mycpuid == 0)
2801                 ipfw_dec_static_count(rule);
2802
2803         /* Free 'stub' on the last CPU */
2804         if (stub != NULL && mycpuid == ncpus - 1)
2805                 kfree(stub, M_IPFW);
2806
2807         /* Try to free this rule */
2808         ipfw_free_rule(rule);
2809
2810         /* Return the next rule */
2811         return n;
2812 }
2813
2814 static void
2815 ipfw_flush_dispatch(struct netmsg *nmsg)
2816 {
2817         struct lwkt_msg *lmsg = &nmsg->nm_lmsg;
2818         int kill_default = lmsg->u.ms_result;
2819         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2820         struct ip_fw *rule;
2821
2822         ipfw_flush_rule_ptrs(ctx); /* more efficient to do outside the loop */
2823
2824         while ((rule = ctx->ipfw_layer3_chain) != NULL &&
2825                (kill_default || rule->rulenum != IPFW_DEFAULT_RULE))
2826                 ipfw_delete_rule(ctx, NULL, rule);
2827
2828         ifnet_forwardmsg(lmsg, mycpuid + 1);
2829 }
2830
2831 static void
2832 ipfw_disable_rule_state_dispatch(struct netmsg *nmsg)
2833 {
2834         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
2835         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2836         struct ip_fw *rule;
2837
2838         ctx->ipfw_gen++;
2839
2840         rule = dmsg->start_rule;
2841         if (rule != NULL) {
2842                 KKASSERT(rule->cpuid == mycpuid);
2843
2844                 /*
2845                  * Move to the position on the next CPU
2846                  * before the msg is forwarded.
2847                  */
2848                 dmsg->start_rule = rule->sibling;
2849         } else {
2850                 KKASSERT(dmsg->rulenum == 0);
2851                 rule = ctx->ipfw_layer3_chain;
2852         }
2853
2854         while (rule != NULL) {
2855                 if (dmsg->rulenum && rule->rulenum != dmsg->rulenum)
2856                         break;
2857                 rule->rule_flags &= ~IPFW_RULE_F_STATE;
2858                 rule = rule->next;
2859         }
2860
2861         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
2862 }
2863
2864 /*
2865  * Deletes all rules from a chain (including the default rule
2866  * if the second argument is set).
2867  * Must be called at splimp().
2868  */
2869 static void
2870 ipfw_flush(int kill_default)
2871 {
2872         struct netmsg_del dmsg;
2873         struct netmsg nmsg;
2874         struct lwkt_msg *lmsg;
2875         struct ip_fw *rule;
2876         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2877
2878         IPFW_ASSERT_CFGPORT(&curthread->td_msgport);
2879
2880         /*
2881          * If 'kill_default' then caller has done the necessary
2882          * msgport syncing; unnecessary to do it again.
2883          */
2884         if (!kill_default) {
2885                 /*
2886                  * Let ipfw_chk() know the rules are going to
2887                  * be flushed, so it could jump directly to
2888                  * the default rule.
2889                  */
2890                 ipfw_flushing = 1;
2891                 netmsg_service_sync();
2892         }
2893
2894         /*
2895          * Clear STATE flag on rules, so no more states (dyn rules)
2896          * will be created.
2897          */
2898         bzero(&dmsg, sizeof(dmsg));
2899         netmsg_init(&dmsg.nmsg, &curthread->td_msgport, 0,
2900                     ipfw_disable_rule_state_dispatch);
2901         ifnet_domsg(&dmsg.nmsg.nm_lmsg, 0);
2902
2903         /*
2904          * This actually nukes all states (dyn rules)
2905          */
2906         lockmgr(&dyn_lock, LK_EXCLUSIVE);
2907         for (rule = ctx->ipfw_layer3_chain; rule != NULL; rule = rule->next) {
2908                 /*
2909                  * Can't check IPFW_RULE_F_STATE here,
2910                  * since it has been cleared previously.
2911                  * Check 'stub' instead.
2912                  */
2913                 if (rule->stub != NULL) {
2914                         /* Force removal */
2915                         remove_dyn_rule_locked(rule, NULL);
2916                 }
2917         }
2918         lockmgr(&dyn_lock, LK_RELEASE);
2919
2920         /*
2921          * Press the 'flush' button
2922          */
2923         bzero(&nmsg, sizeof(nmsg));
2924         netmsg_init(&nmsg, &curthread->td_msgport, 0, ipfw_flush_dispatch);
2925         lmsg = &nmsg.nm_lmsg;
2926         lmsg->u.ms_result = kill_default;
2927         ifnet_domsg(lmsg, 0);
2928
2929         KASSERT(dyn_count == 0, ("%u dyn rule remains\n", 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 remains\n", static_count));
2942                 KASSERT(static_ioc_len == 0,
2943                         ("%u bytes of static rules remains\n", static_ioc_len));
2944         } else {
2945                 KASSERT(static_count == 1,
2946                         ("%u static rules remains\n", static_count));
2947                 KASSERT(static_ioc_len == IOC_RULESIZE(ctx->ipfw_default_rule),
2948                         ("%u bytes of static rules remains, should be %lu\n",
2949                          static_ioc_len, IOC_RULESIZE(ctx->ipfw_default_rule)));
2950         }
2951
2952         /* Flush is done */
2953         ipfw_flushing = 0;
2954 }
2955
2956 static void
2957 ipfw_alt_delete_rule_dispatch(struct netmsg *nmsg)
2958 {
2959         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
2960         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2961         struct ip_fw *rule, *prev;
2962
2963         rule = dmsg->start_rule;
2964         KKASSERT(rule->cpuid == mycpuid);
2965         dmsg->start_rule = rule->sibling;
2966
2967         prev = dmsg->prev_rule;
2968         if (prev != NULL) {
2969                 KKASSERT(prev->cpuid == mycpuid);
2970
2971                 /*
2972                  * Move to the position on the next CPU
2973                  * before the msg is forwarded.
2974                  */
2975                 dmsg->prev_rule = prev->sibling;
2976         }
2977
2978         /*
2979          * flush pointers outside the loop, then delete all matching
2980          * rules.  'prev' remains the same throughout the cycle.
2981          */
2982         ipfw_flush_rule_ptrs(ctx);
2983         while (rule && rule->rulenum == dmsg->rulenum)
2984                 rule = ipfw_delete_rule(ctx, prev, rule);
2985
2986         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
2987 }
2988
2989 static int
2990 ipfw_alt_delete_rule(uint16_t rulenum)
2991 {
2992         struct ip_fw *prev, *rule, *f;
2993         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
2994         struct netmsg_del dmsg;
2995         struct netmsg *nmsg;
2996         int state;
2997
2998         /*
2999          * Locate first rule to delete
3000          */
3001         for (prev = NULL, rule = ctx->ipfw_layer3_chain;
3002              rule && rule->rulenum < rulenum;
3003              prev = rule, rule = rule->next)
3004                 ; /* EMPTY */
3005         if (rule->rulenum != rulenum)
3006                 return EINVAL;
3007
3008         /*
3009          * Check whether any rules with the given number will
3010          * create states.
3011          */
3012         state = 0;
3013         for (f = rule; f && f->rulenum == rulenum; f = f->next) {
3014                 if (f->rule_flags & IPFW_RULE_F_STATE) {
3015                         state = 1;
3016                         break;
3017                 }
3018         }
3019
3020         if (state) {
3021                 /*
3022                  * Clear the STATE flag, so no more states will be
3023                  * created based the rules numbered 'rulenum'.
3024                  */
3025                 bzero(&dmsg, sizeof(dmsg));
3026                 nmsg = &dmsg.nmsg;
3027                 netmsg_init(nmsg, &curthread->td_msgport, 0,
3028                             ipfw_disable_rule_state_dispatch);
3029                 dmsg.start_rule = rule;
3030                 dmsg.rulenum = rulenum;
3031
3032                 ifnet_domsg(&nmsg->nm_lmsg, 0);
3033                 KKASSERT(dmsg.start_rule == NULL);
3034
3035                 /*
3036                  * Nuke all related states
3037                  */
3038                 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3039                 for (f = rule; f && f->rulenum == rulenum; f = f->next) {
3040                         /*
3041                          * Can't check IPFW_RULE_F_STATE here,
3042                          * since it has been cleared previously.
3043                          * Check 'stub' instead.
3044                          */
3045                         if (f->stub != NULL) {
3046                                 /* Force removal */
3047                                 remove_dyn_rule_locked(f, NULL);
3048                         }
3049                 }
3050                 lockmgr(&dyn_lock, LK_RELEASE);
3051         }
3052
3053         /*
3054          * Get rid of the rule duplications on all CPUs
3055          */
3056         bzero(&dmsg, sizeof(dmsg));
3057         nmsg = &dmsg.nmsg;
3058         netmsg_init(nmsg, &curthread->td_msgport, 0,
3059                     ipfw_alt_delete_rule_dispatch);
3060         dmsg.prev_rule = prev;
3061         dmsg.start_rule = rule;
3062         dmsg.rulenum = rulenum;
3063
3064         ifnet_domsg(&nmsg->nm_lmsg, 0);
3065         KKASSERT(dmsg.prev_rule == NULL && dmsg.start_rule == NULL);
3066         return 0;
3067 }
3068
3069 static void
3070 ipfw_alt_delete_ruleset_dispatch(struct netmsg *nmsg)
3071 {
3072         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3073         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3074         struct ip_fw *prev, *rule;
3075 #ifdef INVARIANTS
3076         int del = 0;
3077 #endif
3078
3079         ipfw_flush_rule_ptrs(ctx);
3080
3081         prev = NULL;
3082         rule = ctx->ipfw_layer3_chain;
3083         while (rule != NULL) {
3084                 if (rule->set == dmsg->from_set) {
3085                         rule = ipfw_delete_rule(ctx, prev, rule);
3086 #ifdef INVARIANTS
3087                         del = 1;
3088 #endif
3089                 } else {
3090                         prev = rule;
3091                         rule = rule->next;
3092                 }
3093         }
3094         KASSERT(del, ("no match set?!\n"));
3095
3096         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3097 }
3098
3099 static void
3100 ipfw_disable_ruleset_state_dispatch(struct netmsg *nmsg)
3101 {
3102         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3103         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3104         struct ip_fw *rule;
3105 #ifdef INVARIANTS
3106         int cleared = 0;
3107 #endif
3108
3109         ctx->ipfw_gen++;
3110
3111         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3112                 if (rule->set == dmsg->from_set) {
3113 #ifdef INVARIANTS
3114                         cleared = 1;
3115 #endif
3116                         rule->rule_flags &= ~IPFW_RULE_F_STATE;
3117                 }
3118         }
3119         KASSERT(cleared, ("no match set?!\n"));
3120
3121         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3122 }
3123
3124 static int
3125 ipfw_alt_delete_ruleset(uint8_t set)
3126 {
3127         struct netmsg_del dmsg;
3128         struct netmsg *nmsg;
3129         int state, del;
3130         struct ip_fw *rule;
3131         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3132
3133         /*
3134          * Check whether the 'set' exists.  If it exists,
3135          * then check whether any rules within the set will
3136          * try to create states.
3137          */
3138         state = 0;
3139         del = 0;
3140         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3141                 if (rule->set == set) {
3142                         del = 1;
3143                         if (rule->rule_flags & IPFW_RULE_F_STATE) {
3144                                 state = 1;
3145                                 break;
3146                         }
3147                 }
3148         }
3149         if (!del)
3150                 return 0; /* XXX EINVAL? */
3151
3152         if (state) {
3153                 /*
3154                  * Clear the STATE flag, so no more states will be
3155                  * created based the rules in this set.
3156                  */
3157                 bzero(&dmsg, sizeof(dmsg));
3158                 nmsg = &dmsg.nmsg;
3159                 netmsg_init(nmsg, &curthread->td_msgport, 0,
3160                             ipfw_disable_ruleset_state_dispatch);
3161                 dmsg.from_set = set;
3162
3163                 ifnet_domsg(&nmsg->nm_lmsg, 0);
3164
3165                 /*
3166                  * Nuke all related states
3167                  */
3168                 lockmgr(&dyn_lock, LK_EXCLUSIVE);
3169                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3170                         if (rule->set != set)
3171                                 continue;
3172
3173                         /*
3174                          * Can't check IPFW_RULE_F_STATE here,
3175                          * since it has been cleared previously.
3176                          * Check 'stub' instead.
3177                          */
3178                         if (rule->stub != NULL) {
3179                                 /* Force removal */
3180                                 remove_dyn_rule_locked(rule, NULL);
3181                         }
3182                 }
3183                 lockmgr(&dyn_lock, LK_RELEASE);
3184         }
3185
3186         /*
3187          * Delete this set
3188          */
3189         bzero(&dmsg, sizeof(dmsg));
3190         nmsg = &dmsg.nmsg;
3191         netmsg_init(nmsg, &curthread->td_msgport, 0,
3192                     ipfw_alt_delete_ruleset_dispatch);
3193         dmsg.from_set = set;
3194
3195         ifnet_domsg(&nmsg->nm_lmsg, 0);
3196         return 0;
3197 }
3198
3199 static void
3200 ipfw_alt_move_rule_dispatch(struct netmsg *nmsg)
3201 {
3202         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3203         struct ip_fw *rule;
3204
3205         rule = dmsg->start_rule;
3206         KKASSERT(rule->cpuid == mycpuid);
3207
3208         /*
3209          * Move to the position on the next CPU
3210          * before the msg is forwarded.
3211          */
3212         dmsg->start_rule = rule->sibling;
3213
3214         while (rule && rule->rulenum <= dmsg->rulenum) {
3215                 if (rule->rulenum == dmsg->rulenum)
3216                         rule->set = dmsg->to_set;
3217                 rule = rule->next;
3218         }
3219         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3220 }
3221
3222 static int
3223 ipfw_alt_move_rule(uint16_t rulenum, uint8_t set)
3224 {
3225         struct netmsg_del dmsg;
3226         struct netmsg *nmsg;
3227         struct ip_fw *rule;
3228         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3229
3230         /*
3231          * Locate first rule to move
3232          */
3233         for (rule = ctx->ipfw_layer3_chain; rule && rule->rulenum <= rulenum;
3234              rule = rule->next) {
3235                 if (rule->rulenum == rulenum && rule->set != set)
3236                         break;
3237         }
3238         if (rule == NULL || rule->rulenum > rulenum)
3239                 return 0; /* XXX error? */
3240
3241         bzero(&dmsg, sizeof(dmsg));
3242         nmsg = &dmsg.nmsg;
3243         netmsg_init(nmsg, &curthread->td_msgport, 0,
3244                     ipfw_alt_move_rule_dispatch);
3245         dmsg.start_rule = rule;
3246         dmsg.rulenum = rulenum;
3247         dmsg.to_set = set;
3248
3249         ifnet_domsg(&nmsg->nm_lmsg, 0);
3250         KKASSERT(dmsg.start_rule == NULL);
3251         return 0;
3252 }
3253
3254 static void
3255 ipfw_alt_move_ruleset_dispatch(struct netmsg *nmsg)
3256 {
3257         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3258         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3259         struct ip_fw *rule;
3260
3261         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3262                 if (rule->set == dmsg->from_set)
3263                         rule->set = dmsg->to_set;
3264         }
3265         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3266 }
3267
3268 static int
3269 ipfw_alt_move_ruleset(uint8_t from_set, uint8_t to_set)
3270 {
3271         struct netmsg_del dmsg;
3272         struct netmsg *nmsg;
3273
3274         bzero(&dmsg, sizeof(dmsg));
3275         nmsg = &dmsg.nmsg;
3276         netmsg_init(nmsg, &curthread->td_msgport, 0,
3277                     ipfw_alt_move_ruleset_dispatch);
3278         dmsg.from_set = from_set;
3279         dmsg.to_set = to_set;
3280
3281         ifnet_domsg(&nmsg->nm_lmsg, 0);
3282         return 0;
3283 }
3284
3285 static void
3286 ipfw_alt_swap_ruleset_dispatch(struct netmsg *nmsg)
3287 {
3288         struct netmsg_del *dmsg = (struct netmsg_del *)nmsg;
3289         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3290         struct ip_fw *rule;
3291
3292         for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3293                 if (rule->set == dmsg->from_set)
3294                         rule->set = dmsg->to_set;
3295                 else if (rule->set == dmsg->to_set)
3296                         rule->set = dmsg->from_set;
3297         }
3298         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3299 }
3300
3301 static int
3302 ipfw_alt_swap_ruleset(uint8_t set1, uint8_t set2)
3303 {
3304         struct netmsg_del dmsg;
3305         struct netmsg *nmsg;
3306
3307         bzero(&dmsg, sizeof(dmsg));
3308         nmsg = &dmsg.nmsg;
3309         netmsg_init(nmsg, &curthread->td_msgport, 0,
3310                     ipfw_alt_swap_ruleset_dispatch);
3311         dmsg.from_set = set1;
3312         dmsg.to_set = set2;
3313
3314         ifnet_domsg(&nmsg->nm_lmsg, 0);
3315         return 0;
3316 }
3317
3318 /**
3319  * Remove all rules with given number, and also do set manipulation.
3320  *
3321  * The argument is an uint32_t. The low 16 bit are the rule or set number,
3322  * the next 8 bits are the new set, the top 8 bits are the command:
3323  *
3324  *      0       delete rules with given number
3325  *      1       delete rules with given set number
3326  *      2       move rules with given number to new set
3327  *      3       move rules with given set number to new set
3328  *      4       swap sets with given numbers
3329  */
3330 static int
3331 ipfw_ctl_alter(uint32_t arg)
3332 {
3333         uint16_t rulenum;
3334         uint8_t cmd, new_set;
3335         int error = 0;
3336
3337         rulenum = arg & 0xffff;
3338         cmd = (arg >> 24) & 0xff;
3339         new_set = (arg >> 16) & 0xff;
3340
3341         if (cmd > 4)
3342                 return EINVAL;
3343         if (new_set >= IPFW_DEFAULT_SET)
3344                 return EINVAL;
3345         if (cmd == 0 || cmd == 2) {
3346                 if (rulenum == IPFW_DEFAULT_RULE)
3347                         return EINVAL;
3348         } else {
3349                 if (rulenum >= IPFW_DEFAULT_SET)
3350                         return EINVAL;
3351         }
3352
3353         switch (cmd) {
3354         case 0: /* delete rules with given number */
3355                 error = ipfw_alt_delete_rule(rulenum);
3356                 break;
3357
3358         case 1: /* delete all rules with given set number */
3359                 error = ipfw_alt_delete_ruleset(rulenum);
3360                 break;
3361
3362         case 2: /* move rules with given number to new set */
3363                 error = ipfw_alt_move_rule(rulenum, new_set);
3364                 break;
3365
3366         case 3: /* move rules with given set number to new set */
3367                 error = ipfw_alt_move_ruleset(rulenum, new_set);
3368                 break;
3369
3370         case 4: /* swap two sets */
3371                 error = ipfw_alt_swap_ruleset(rulenum, new_set);
3372                 break;
3373         }
3374         return error;
3375 }
3376
3377 /*
3378  * Clear counters for a specific rule.
3379  */
3380 static void
3381 clear_counters(struct ip_fw *rule, int log_only)
3382 {
3383         ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3384
3385         if (log_only == 0) {
3386                 rule->bcnt = rule->pcnt = 0;
3387                 rule->timestamp = 0;
3388         }
3389         if (l->o.opcode == O_LOG)
3390                 l->log_left = l->max_log;
3391 }
3392
3393 static void
3394 ipfw_zero_entry_dispatch(struct netmsg *nmsg)
3395 {
3396         struct netmsg_zent *zmsg = (struct netmsg_zent *)nmsg;
3397         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3398         struct ip_fw *rule;
3399
3400         if (zmsg->rulenum == 0) {
3401                 KKASSERT(zmsg->start_rule == NULL);
3402
3403                 ctx->ipfw_norule_counter = 0;
3404                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next)
3405                         clear_counters(rule, zmsg->log_only);
3406         } else {
3407                 struct ip_fw *start = zmsg->start_rule;
3408
3409                 KKASSERT(start->cpuid == mycpuid);
3410                 KKASSERT(start->rulenum == zmsg->rulenum);
3411
3412                 /*
3413                  * We can have multiple rules with the same number, so we
3414                  * need to clear them all.
3415                  */
3416                 for (rule = start; rule && rule->rulenum == zmsg->rulenum;
3417                      rule = rule->next)
3418                         clear_counters(rule, zmsg->log_only);
3419
3420                 /*
3421                  * Move to the position on the next CPU
3422                  * before the msg is forwarded.
3423                  */
3424                 zmsg->start_rule = start->sibling;
3425         }
3426         ifnet_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
3427 }
3428
3429 /**
3430  * Reset some or all counters on firewall rules.
3431  * @arg frwl is null to clear all entries, or contains a specific
3432  * rule number.
3433  * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3434  */
3435 static int
3436 ipfw_ctl_zero_entry(int rulenum, int log_only)
3437 {
3438         struct netmsg_zent zmsg;
3439         struct netmsg *nmsg;
3440         const char *msg;
3441         struct ipfw_context *ctx = ipfw_ctx[mycpuid];
3442
3443         bzero(&zmsg, sizeof(zmsg));
3444         nmsg = &zmsg.nmsg;
3445         netmsg_init(nmsg, &curthread->td_msgport, 0, ipfw_zero_entry_dispatch);
3446         zmsg.log_only = log_only;
3447
3448         if (rulenum == 0) {
3449                 msg = log_only ? "ipfw: All logging counts reset.\n"
3450                                : "ipfw: Accounting cleared.\n";
3451         } else {
3452                 struct ip_fw *rule;
3453
3454                 /*
3455                  * Locate the first rule with 'rulenum'
3456                  */
3457                 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) {
3458                         if (rule->rulenum == rulenum)
3459                                 break;
3460                 }
3461                 if (rule == NULL) /* we did not find any matching rules */
3462                         return (EINVAL);
3463                 zmsg.start_rule = rule;
3464                 zmsg.rulenum = rulenum;
3465
3466                 msg = log_only ? "ipfw: Entry %d logging count reset.\n"
3467                                : "ipfw: Entry %d cleared.\n";
3468         }
3469         ifnet_domsg(&nmsg->nm_lmsg, 0);
3470         KKASSERT(zmsg.start_rule == NULL);
3471
3472         if (fw_verbose)
3473                 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3474         return (0);
3475 }
3476
3477 /*
3478  * Check validity of the structure before insert.
3479  * Fortunately rules are simple, so this mostly need to check rule sizes.
3480  */
3481 static int
3482 ipfw_check_ioc_rule(struct ipfw_ioc_rule *rule, int size, uint32_t *rule_flags)
3483 {
3484         int l, cmdlen = 0;
3485         int have_action = 0;
3486         ipfw_insn *cmd;
3487
3488         *rule_flags = 0;
3489
3490         /* Check for valid size */
3491         if (size < sizeof(*rule)) {
3492                 kprintf("ipfw: rule too short\n");
3493                 return EINVAL;
3494         }
3495         l = IOC_RULESIZE(rule);
3496         if (l != size) {
3497                 kprintf("ipfw: size mismatch (have %d want %d)\n", size, l);
3498                 return EINVAL;
3499         }
3500
3501         /* Check rule number */
3502         if (rule->rulenum == IPFW_DEFAULT_RULE) {
3503                 kprintf("ipfw: invalid rule number\n");
3504                 return EINVAL;
3505         }
3506
3507         /*
3508          * Now go for the individual checks. Very simple ones, basically only
3509          * instruction sizes.
3510          */
3511         for (l = rule->cmd_len, cmd = rule->cmd; l > 0;
3512              l -= cmdlen, cmd += cmdlen) {
3513                 cmdlen = F_LEN(cmd);
3514                 if (cmdlen > l) {
3515                         kprintf("ipfw: opcode %d size truncated\n",
3516                                 cmd->opcode);
3517                         return EINVAL;
3518                 }
3519
3520                 DPRINTF("ipfw: opcode %d\n", cmd->opcode);
3521
3522                 if (cmd->opcode == O_KEEP_STATE || cmd->opcode == O_LIMIT) {
3523                         /* This rule will create states */
3524                         *rule_flags |= IPFW_RULE_F_STATE;
3525                 }
3526
3527                 switch (cmd->opcode) {
3528                 case O_NOP:
3529                 case O_PROBE_STATE:
3530                 case O_KEEP_STATE:
3531                 case O_PROTO:
3532                 case O_IP_SRC_ME:
3533                 case O_IP_DST_ME:
3534                 case O_LAYER2:
3535                 case O_IN:
3536                 case O_FRAG:
3537                 case O_IPOPT:
3538                 case O_IPLEN:
3539                 case O_IPID:
3540                 case O_IPTOS:
3541                 case O_IPPRECEDENCE:
3542                 case O_IPTTL:
3543                 case O_IPVER:
3544                 case O_TCPWIN:
3545                 case O_TCPFLAGS:
3546                 case O_TCPOPTS:
3547                 case O_ESTAB:
3548                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
3549                                 goto bad_size;
3550                         break;
3551
3552                 case O_UID:
3553                 case O_GID:
3554                 case O_IP_SRC:
3555                 case O_IP_DST:
3556                 case O_TCPSEQ:
3557                 case O_TCPACK:
3558                 case O_PROB:
3559                 case O_ICMPTYPE:
3560                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3561                                 goto bad_size;
3562                         break;
3563
3564                 case O_LIMIT:
3565                         if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3566                                 goto bad_size;
3567                         break;
3568
3569                 case O_LOG:
3570                         if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3571                                 goto bad_size;
3572
3573                         ((ipfw_insn_log *)cmd)->log_left =
3574                             ((ipfw_insn_log *)cmd)->max_log;
3575
3576                         break;
3577
3578                 case O_IP_SRC_MASK:
3579                 case O_IP_DST_MASK:
3580                         if (cmdlen != F_INSN_SIZE(ipfw_insn_ip))
3581                                 goto bad_size;
3582                         if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) {
3583                                 kprintf("ipfw: opcode %d, useless rule\n",
3584                                         cmd->opcode);
3585                                 return EINVAL;
3586                         }
3587                         break;
3588
3589                 case O_IP_SRC_SET:
3590                 case O_IP_DST_SET:
3591                         if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3592                                 kprintf("ipfw: invalid set size %d\n",
3593                                         cmd->arg1);
3594                                 return EINVAL;
3595                         }
3596                         if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3597                             (cmd->arg1+31)/32 )
3598                                 goto bad_size;
3599                         break;
3600
3601                 case O_MACADDR2:
3602                         if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3603                                 goto bad_size;
3604                         break;
3605
3606                 case O_MAC_TYPE:
3607                 case O_IP_SRCPORT:
3608                 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3609                         if (cmdlen < 2 || cmdlen > 31)
3610                                 goto bad_size;
3611                         break;
3612
3613                 case O_RECV:
3614                 case O_XMIT:
3615                 case O_VIA:
3616                         if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3617                                 goto bad_size;
3618                         break;
3619
3620                 case O_PIPE:
3621                 case O_QUEUE:
3622                         if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe))
3623                                 goto bad_size;
3624                         goto check_action;
3625
3626                 case O_FORWARD_IP:
3627                         if (cmdlen != F_INSN_SIZE(ipfw_insn_sa)) {
3628                                 goto bad_size;
3629                         } else {
3630                                 in_addr_t fwd_addr;
3631
3632                                 fwd_addr = ((ipfw_insn_sa *)cmd)->
3633                                            sa.sin_addr.s_addr;
3634                                 if (IN_MULTICAST(ntohl(fwd_addr))) {
3635                                         kprintf("ipfw: try forwarding to "
3636                                                 "multicast address\n");
3637                                         return EINVAL;
3638                                 }
3639                         }
3640                         goto check_action;
3641
3642                 case O_FORWARD_MAC: /* XXX not implemented yet */
3643                 case O_CHECK_STATE:
3644                 case O_COUNT:
3645                 case O_ACCEPT:
3646                 case O_DENY:
3647                 case O_REJECT:
3648                 case O_SKIPTO:
3649                 case O_DIVERT:
3650                 case O_TEE:
3651                         if (cmdlen != F_INSN_SIZE(ipfw_insn))
3652                                 goto bad_size;
3653 check_action:
3654                         if (have_action) {
3655                                 kprintf("ipfw: opcode %d, multiple actions"
3656                                         " not allowed\n",
3657                                         cmd->opcode);
3658                                 return EINVAL;
3659                         }
3660                         have_action = 1;
3661                         if (l != cmdlen) {
3662                                 kprintf("ipfw: opcode %d, action must be"
3663                                         " last opcode\n",
3664                                         cmd->opcode);
3665                                 return EINVAL;
3666                         }
3667                         break;
3668                 default:
3669                         kprintf("ipfw: opcode %d, unknown opcode\n",
3670                                 cmd->opcode);
3671                         return EINVAL;
3672                 }
3673         }
3674         if (have_action == 0) {
3675                 kprintf("ipfw: missing action\n");
3676                 return EINVAL;
3677         }
3678         return 0;
3679
3680 bad_size:
3681         kprintf("ipfw: opcode %d size %d wrong\n",
3682                 cmd->opcode, cmdlen);
3683         return EINVAL;
3684 }
3685
3686 static int
3687 ipfw_ctl_add_rule(struct sockopt *sopt)
3688 {
3689         struct ipfw_ioc_rule *ioc_rule;
3690         size_t size;
3691         uint32_t rule_flags;
3692         int error;
3693         
3694         size = sopt->sopt_valsize;
3695         if (size > (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX) ||
3696             size < sizeof(*ioc_rule)) {
3697                 return EINVAL;
3698         }
3699         if (size != (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX)) {
3700                 sopt->sopt_val = krealloc(sopt->sopt_val, sizeof(uint32_t) *
3701                                           IPFW_RULE_SIZE_MAX, M_TEMP, M_WAITOK);
3702         }
3703         ioc_rule = sopt->sopt_val;
3704
3705         error = ipfw_check_ioc_rule(ioc_rule, size, &rule_flags);
3706         if (error)
3707                 return error;
3708
3709         ipfw_add_rule(ioc_rule, rule_flags);
3710
3711         if (sopt->sopt_dir == SOPT_GET)
3712                 sopt->sopt_valsize = IOC_RULESIZE(ioc_rule);
3713         return 0;
3714 }
3715
3716 static void *
3717 ipfw_copy_rule(const struct ip_fw *rule, struct ipfw_ioc_rule *ioc_rule)
3718 {
3719         const struct ip_fw *sibling;
3720 #ifdef INVARIANTS
3721         int i;
3722 #endif
3723
3724         KKASSERT(rule->cpuid == IPFW_CFGCPUID);
3725
3726         ioc_rule->act_ofs = rule->act_ofs;
3727         ioc_rule->cmd_len = rule->cmd_len;
3728         ioc_rule->rulenum = rule->rulenum;
3729         ioc_rule->set = rule->set;
3730         ioc_rule->usr_flags = rule->usr_flags;
3731
3732         ioc_rule->set_disable = ipfw_ctx[mycpuid]->ipfw_set_disable;
3733         ioc_rule->static_count = static_count;
3734         ioc_rule->static_len = static_ioc_len;
3735
3736         /*
3737          * Visit (read-only) all of the rule's duplications to get
3738          * the necessary statistics
3739          */
3740 #ifdef INVARIANTS
3741         i = 0;
3742 #endif
3743         ioc_rule->pcnt = 0;
3744         ioc_rule->bcnt = 0;
3745         ioc_rule->timestamp = 0;
3746         for (sibling = rule; sibling != NULL; sibling = sibling->sibling) {
3747                 ioc_rule->pcnt += sibling->pcnt;
3748                 ioc_rule->bcnt += sibling->bcnt;
3749                 if (sibling->timestamp > ioc_rule->timestamp)
3750                         ioc_rule->timestamp = sibling->timestamp;
3751 #ifdef INVARIANTS
3752                 ++i;
3753 #endif
3754         }
3755         KASSERT(i == ncpus, ("static rule is not duplicated on every cpu\n"));
3756
3757         bcopy(rule->cmd, ioc_rule->cmd, ioc_rule->cmd_len * 4 /* XXX */);
3758
3759         return ((uint8_t *)ioc_rule + IOC_RULESIZE(ioc_rule));
3760 }
3761
3762 static void
3763 ipfw_copy_state(const ipfw_dyn_rule *dyn_rule,
3764                 struct ipfw_ioc_state *ioc_state)
3765 {
3766         const struct ipfw_flow_id *id;
3767         struct ipfw_ioc_flowid *ioc_id;
3768
3769         ioc_state->expire = TIME_LEQ(dyn_rule->expire, time_second) ?
3770                             0 : dyn_rule->expire - time_second;
3771         ioc_state->pcnt = dyn_rule->pcnt;
3772         ioc_state->bcnt = dyn_rule->bcnt;
3773
3774         ioc_state->dyn_type = dyn_rule->dyn_type;
3775         ioc_state->count = dyn_rule->count;
3776
3777         ioc_state->rulenum = dyn_rule->stub->rule[mycpuid]->rulenum;
3778
3779         id = &dyn_rule->id;
3780         ioc_id = &ioc_state->id;
3781
3782         ioc_id->type = ETHERTYPE_IP;
3783         ioc_id->u.ip.dst_ip = id->dst_ip;
3784         ioc_id->u.ip.src_ip = id->src_ip;
3785         ioc_id->u.ip.dst_port = id->dst_port;
3786         ioc_id->u.ip.src_port = id->src_port;
3787         ioc_id->u.ip.proto = id->proto;
3788 }
3789
3790 static int
3791 ipfw_ctl_get_rules(struct sockopt *sopt)
3792 {
3793         struct ipfw_context *ctx = ipfw_ctx[mycpuid];