2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2017-2018 Yandex LLC
5 * Copyright (c) 2017-2018 Andrey V. Elsukov <ae@FreeBSD.org>
6 * Copyright (c) 2002 Luigi Rizzo, Universita` di Pisa
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
34 #include "opt_inet6.h"
37 #error IPFIREWALL requires INET.
40 #include <sys/param.h>
41 #include <sys/systm.h>
44 #include <sys/kernel.h>
47 #include <sys/queue.h>
48 #include <sys/rmlock.h>
50 #include <sys/socket.h>
51 #include <sys/sysctl.h>
52 #include <sys/syslog.h>
53 #include <net/ethernet.h>
55 #include <net/if_var.h>
58 #include <netinet/in.h>
59 #include <netinet/ip.h>
60 #include <netinet/ip_var.h>
61 #include <netinet/ip_fw.h>
62 #include <netinet/tcp_var.h>
63 #include <netinet/udp.h>
65 #include <netinet/ip6.h> /* IN6_ARE_ADDR_EQUAL */
67 #include <netinet6/in6_var.h>
68 #include <netinet6/ip6_var.h>
69 #include <netinet6/scope6_var.h>
72 #include <netpfil/ipfw/ip_fw_private.h>
74 #include <machine/in_cksum.h> /* XXX for in_cksum */
77 #include <security/mac/mac_framework.h>
81 * Description of dynamic states.
83 * Dynamic states are stored in lists accessed through a hash tables
84 * whose size is curr_dyn_buckets. This value can be modified through
85 * the sysctl variable dyn_buckets.
87 * Currently there are four tables: dyn_ipv4, dyn_ipv6, dyn_ipv4_parent,
88 * and dyn_ipv6_parent.
90 * When a packet is received, its address fields hashed, then matched
91 * against the entries in the corresponding list by addr_type.
92 * Dynamic states can be used for different purposes:
94 * + enforcing limits on the number of sessions;
95 * + in-kernel NAT (not implemented yet)
97 * The lifetime of dynamic states is regulated by dyn_*_lifetime,
98 * measured in seconds and depending on the flags.
100 * The total number of dynamic states is equal to UMA zone items count.
101 * The max number of dynamic states is dyn_max. When we reach
102 * the maximum number of rules we do not create anymore. This is
103 * done to avoid consuming too much memory, but also too much
104 * time when searching on each packet (ideally, we should try instead
105 * to put a limit on the length of the list on each bucket...).
107 * Each state holds a pointer to the parent ipfw rule so we know what
108 * action to perform. Dynamic rules are removed when the parent rule is
111 * There are some limitations with dynamic rules -- we do not
112 * obey the 'randomized match', and we do not do multiple
113 * passes through the firewall. XXX check the latter!!!
116 /* By default use jenkins hash function */
117 #define IPFIREWALL_JENKINSHASH
119 #define DYN_COUNTER_INC(d, dir, pktlen) do { \
120 (d)->pcnt_ ## dir++; \
121 (d)->bcnt_ ## dir += pktlen; \
124 #define DYN_REFERENCED 0x01
126 * DYN_REFERENCED flag is used to show that state keeps reference to named
127 * object, and this reference should be released when state becomes expired.
131 void *parent; /* pointer to parent rule */
132 uint32_t chain_id; /* cached ruleset id */
133 uint32_t f_pos; /* cached rule index */
135 uint32_t hashval; /* hash value used for hash resize */
136 uint16_t fibnum; /* fib used to send keepalives */
138 uint8_t flags; /* internal flags */
139 uint16_t rulenum; /* parent rule number */
140 uint32_t ruleid; /* parent rule id */
142 uint32_t state; /* TCP session state and flags */
143 uint32_t ack_fwd; /* most recent ACKs in forward */
144 uint32_t ack_rev; /* and reverse direction (used */
145 /* to generate keepalives) */
146 uint32_t sync; /* synchronization time */
147 uint32_t expire; /* expire time */
149 uint64_t pcnt_fwd; /* bytes counter in forward */
150 uint64_t bcnt_fwd; /* packets counter in forward */
151 uint64_t pcnt_rev; /* bytes counter in reverse */
152 uint64_t bcnt_rev; /* packets counter in reverse */
155 #define DPARENT_COUNT_DEC(p) do { \
156 MPASS(p->count > 0); \
157 ck_pr_dec_32(&(p)->count); \
159 #define DPARENT_COUNT_INC(p) ck_pr_inc_32(&(p)->count)
160 #define DPARENT_COUNT(p) ck_pr_load_32(&(p)->count)
162 void *parent; /* pointer to parent rule */
163 uint32_t count; /* number of linked states */
165 uint16_t rulenum; /* parent rule number */
166 uint32_t ruleid; /* parent rule id */
167 uint32_t hashval; /* hash value used for hash resize */
168 uint32_t expire; /* expire time */
171 struct dyn_ipv4_state {
172 uint8_t type; /* State type */
173 uint8_t proto; /* UL Protocol */
174 uint16_t kidx; /* named object index */
175 uint16_t sport, dport; /* ULP source and destination ports */
176 in_addr_t src, dst; /* IPv4 source and destination */
179 struct dyn_data *data;
180 struct dyn_parent *limit;
182 CK_SLIST_ENTRY(dyn_ipv4_state) entry;
183 SLIST_ENTRY(dyn_ipv4_state) expired;
185 CK_SLIST_HEAD(dyn_ipv4ck_slist, dyn_ipv4_state);
186 VNET_DEFINE_STATIC(struct dyn_ipv4ck_slist *, dyn_ipv4);
187 VNET_DEFINE_STATIC(struct dyn_ipv4ck_slist *, dyn_ipv4_parent);
189 SLIST_HEAD(dyn_ipv4_slist, dyn_ipv4_state);
190 VNET_DEFINE_STATIC(struct dyn_ipv4_slist, dyn_expired_ipv4);
191 #define V_dyn_ipv4 VNET(dyn_ipv4)
192 #define V_dyn_ipv4_parent VNET(dyn_ipv4_parent)
193 #define V_dyn_expired_ipv4 VNET(dyn_expired_ipv4)
196 struct dyn_ipv6_state {
197 uint8_t type; /* State type */
198 uint8_t proto; /* UL Protocol */
199 uint16_t kidx; /* named object index */
200 uint16_t sport, dport; /* ULP source and destination ports */
201 struct in6_addr src, dst; /* IPv6 source and destination */
202 uint32_t zoneid; /* IPv6 scope zone id */
204 struct dyn_data *data;
205 struct dyn_parent *limit;
207 CK_SLIST_ENTRY(dyn_ipv6_state) entry;
208 SLIST_ENTRY(dyn_ipv6_state) expired;
210 CK_SLIST_HEAD(dyn_ipv6ck_slist, dyn_ipv6_state);
211 VNET_DEFINE_STATIC(struct dyn_ipv6ck_slist *, dyn_ipv6);
212 VNET_DEFINE_STATIC(struct dyn_ipv6ck_slist *, dyn_ipv6_parent);
214 SLIST_HEAD(dyn_ipv6_slist, dyn_ipv6_state);
215 VNET_DEFINE_STATIC(struct dyn_ipv6_slist, dyn_expired_ipv6);
216 #define V_dyn_ipv6 VNET(dyn_ipv6)
217 #define V_dyn_ipv6_parent VNET(dyn_ipv6_parent)
218 #define V_dyn_expired_ipv6 VNET(dyn_expired_ipv6)
222 * Per-CPU pointer indicates that specified state is currently in use
223 * and must not be reclaimed by expiration callout.
225 static void **dyn_hp_cache;
226 DPCPU_DEFINE_STATIC(void *, dyn_hp);
227 #define DYNSTATE_GET(cpu) ck_pr_load_ptr(DPCPU_ID_PTR((cpu), dyn_hp))
228 #define DYNSTATE_PROTECT(v) ck_pr_store_ptr(DPCPU_PTR(dyn_hp), (v))
229 #define DYNSTATE_RELEASE() DYNSTATE_PROTECT(NULL)
230 #define DYNSTATE_CRITICAL_ENTER() critical_enter()
231 #define DYNSTATE_CRITICAL_EXIT() do { \
232 DYNSTATE_RELEASE(); \
237 * We keep two version numbers, one is updated when new entry added to
238 * the list. Second is updated when an entry deleted from the list.
239 * Versions are updated under bucket lock.
241 * Bucket "add" version number is used to know, that in the time between
242 * state lookup (i.e. ipfw_dyn_lookup_state()) and the followed state
243 * creation (i.e. ipfw_dyn_install_state()) another concurrent thread did
244 * not install some state in this bucket. Using this info we can avoid
245 * additional state lookup, because we are sure that we will not install
248 * Also doing the tracking of bucket "del" version during lookup we can
249 * be sure, that state entry was not unlinked and freed in time between
250 * we read the state pointer and protect it with hazard pointer.
252 * An entry unlinked from CK list keeps unchanged until it is freed.
253 * Unlinked entries are linked into expired lists using "expired" field.
257 * dyn_expire_lock is used to protect access to dyn_expired_xxx lists.
258 * dyn_bucket_lock is used to get write access to lists in specific bucket.
259 * Currently one dyn_bucket_lock is used for all ipv4, ipv4_parent, ipv6,
260 * and ipv6_parent lists.
262 VNET_DEFINE_STATIC(struct mtx, dyn_expire_lock);
263 VNET_DEFINE_STATIC(struct mtx *, dyn_bucket_lock);
264 #define V_dyn_expire_lock VNET(dyn_expire_lock)
265 #define V_dyn_bucket_lock VNET(dyn_bucket_lock)
268 * Bucket's add/delete generation versions.
270 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv4_add);
271 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv4_del);
272 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv4_parent_add);
273 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv4_parent_del);
274 #define V_dyn_ipv4_add VNET(dyn_ipv4_add)
275 #define V_dyn_ipv4_del VNET(dyn_ipv4_del)
276 #define V_dyn_ipv4_parent_add VNET(dyn_ipv4_parent_add)
277 #define V_dyn_ipv4_parent_del VNET(dyn_ipv4_parent_del)
280 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv6_add);
281 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv6_del);
282 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv6_parent_add);
283 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv6_parent_del);
284 #define V_dyn_ipv6_add VNET(dyn_ipv6_add)
285 #define V_dyn_ipv6_del VNET(dyn_ipv6_del)
286 #define V_dyn_ipv6_parent_add VNET(dyn_ipv6_parent_add)
287 #define V_dyn_ipv6_parent_del VNET(dyn_ipv6_parent_del)
290 #define DYN_BUCKET(h, b) ((h) & (b - 1))
291 #define DYN_BUCKET_VERSION(b, v) ck_pr_load_32(&V_dyn_ ## v[(b)])
292 #define DYN_BUCKET_VERSION_BUMP(b, v) ck_pr_inc_32(&V_dyn_ ## v[(b)])
294 #define DYN_BUCKET_LOCK_INIT(lock, b) \
295 mtx_init(&lock[(b)], "IPFW dynamic bucket", NULL, MTX_DEF)
296 #define DYN_BUCKET_LOCK_DESTROY(lock, b) mtx_destroy(&lock[(b)])
297 #define DYN_BUCKET_LOCK(b) mtx_lock(&V_dyn_bucket_lock[(b)])
298 #define DYN_BUCKET_UNLOCK(b) mtx_unlock(&V_dyn_bucket_lock[(b)])
299 #define DYN_BUCKET_ASSERT(b) mtx_assert(&V_dyn_bucket_lock[(b)], MA_OWNED)
301 #define DYN_EXPIRED_LOCK_INIT() \
302 mtx_init(&V_dyn_expire_lock, "IPFW expired states list", NULL, MTX_DEF)
303 #define DYN_EXPIRED_LOCK_DESTROY() mtx_destroy(&V_dyn_expire_lock)
304 #define DYN_EXPIRED_LOCK() mtx_lock(&V_dyn_expire_lock)
305 #define DYN_EXPIRED_UNLOCK() mtx_unlock(&V_dyn_expire_lock)
307 VNET_DEFINE_STATIC(uint32_t, dyn_buckets_max);
308 VNET_DEFINE_STATIC(uint32_t, curr_dyn_buckets);
309 VNET_DEFINE_STATIC(struct callout, dyn_timeout);
310 #define V_dyn_buckets_max VNET(dyn_buckets_max)
311 #define V_curr_dyn_buckets VNET(curr_dyn_buckets)
312 #define V_dyn_timeout VNET(dyn_timeout)
314 /* Maximum length of states chain in a bucket */
315 VNET_DEFINE_STATIC(uint32_t, curr_max_length);
316 #define V_curr_max_length VNET(curr_max_length)
318 VNET_DEFINE_STATIC(uint32_t, dyn_keep_states);
319 #define V_dyn_keep_states VNET(dyn_keep_states)
321 VNET_DEFINE_STATIC(uma_zone_t, dyn_data_zone);
322 VNET_DEFINE_STATIC(uma_zone_t, dyn_parent_zone);
323 VNET_DEFINE_STATIC(uma_zone_t, dyn_ipv4_zone);
325 VNET_DEFINE_STATIC(uma_zone_t, dyn_ipv6_zone);
326 #define V_dyn_ipv6_zone VNET(dyn_ipv6_zone)
328 #define V_dyn_data_zone VNET(dyn_data_zone)
329 #define V_dyn_parent_zone VNET(dyn_parent_zone)
330 #define V_dyn_ipv4_zone VNET(dyn_ipv4_zone)
333 * Timeouts for various events in handing dynamic rules.
335 VNET_DEFINE_STATIC(uint32_t, dyn_ack_lifetime);
336 VNET_DEFINE_STATIC(uint32_t, dyn_syn_lifetime);
337 VNET_DEFINE_STATIC(uint32_t, dyn_fin_lifetime);
338 VNET_DEFINE_STATIC(uint32_t, dyn_rst_lifetime);
339 VNET_DEFINE_STATIC(uint32_t, dyn_udp_lifetime);
340 VNET_DEFINE_STATIC(uint32_t, dyn_short_lifetime);
342 #define V_dyn_ack_lifetime VNET(dyn_ack_lifetime)
343 #define V_dyn_syn_lifetime VNET(dyn_syn_lifetime)
344 #define V_dyn_fin_lifetime VNET(dyn_fin_lifetime)
345 #define V_dyn_rst_lifetime VNET(dyn_rst_lifetime)
346 #define V_dyn_udp_lifetime VNET(dyn_udp_lifetime)
347 #define V_dyn_short_lifetime VNET(dyn_short_lifetime)
350 * Keepalives are sent if dyn_keepalive is set. They are sent every
351 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
352 * seconds of lifetime of a rule.
353 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
354 * than dyn_keepalive_period.
356 VNET_DEFINE_STATIC(uint32_t, dyn_keepalive_interval);
357 VNET_DEFINE_STATIC(uint32_t, dyn_keepalive_period);
358 VNET_DEFINE_STATIC(uint32_t, dyn_keepalive);
359 VNET_DEFINE_STATIC(time_t, dyn_keepalive_last);
361 #define V_dyn_keepalive_interval VNET(dyn_keepalive_interval)
362 #define V_dyn_keepalive_period VNET(dyn_keepalive_period)
363 #define V_dyn_keepalive VNET(dyn_keepalive)
364 #define V_dyn_keepalive_last VNET(dyn_keepalive_last)
366 VNET_DEFINE_STATIC(uint32_t, dyn_max); /* max # of dynamic states */
367 VNET_DEFINE_STATIC(uint32_t, dyn_count); /* number of states */
368 VNET_DEFINE_STATIC(uint32_t, dyn_parent_max); /* max # of parent states */
369 VNET_DEFINE_STATIC(uint32_t, dyn_parent_count); /* number of parent states */
371 #define V_dyn_max VNET(dyn_max)
372 #define V_dyn_count VNET(dyn_count)
373 #define V_dyn_parent_max VNET(dyn_parent_max)
374 #define V_dyn_parent_count VNET(dyn_parent_count)
376 #define DYN_COUNT_DEC(name) do { \
377 MPASS((V_ ## name) > 0); \
378 ck_pr_dec_32(&(V_ ## name)); \
380 #define DYN_COUNT_INC(name) ck_pr_inc_32(&(V_ ## name))
381 #define DYN_COUNT(name) ck_pr_load_32(&(V_ ## name))
383 static time_t last_log; /* Log ratelimiting */
386 * Get/set maximum number of dynamic states in given VNET instance.
389 sysctl_dyn_max(SYSCTL_HANDLER_ARGS)
395 error = sysctl_handle_32(oidp, &nstates, 0, req);
396 /* Read operation or some error */
397 if ((error != 0) || (req->newptr == NULL))
401 uma_zone_set_max(V_dyn_data_zone, V_dyn_max);
406 sysctl_dyn_parent_max(SYSCTL_HANDLER_ARGS)
411 nstates = V_dyn_parent_max;
412 error = sysctl_handle_32(oidp, &nstates, 0, req);
413 /* Read operation or some error */
414 if ((error != 0) || (req->newptr == NULL))
417 V_dyn_parent_max = nstates;
418 uma_zone_set_max(V_dyn_parent_zone, V_dyn_parent_max);
423 sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS)
428 nbuckets = V_dyn_buckets_max;
429 error = sysctl_handle_32(oidp, &nbuckets, 0, req);
430 /* Read operation or some error */
431 if ((error != 0) || (req->newptr == NULL))
435 V_dyn_buckets_max = 1 << fls(nbuckets - 1);
441 SYSCTL_DECL(_net_inet_ip_fw);
443 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_count,
444 CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(dyn_count), 0,
445 "Current number of dynamic states.");
446 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_parent_count,
447 CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(dyn_parent_count), 0,
448 "Current number of parent states. ");
449 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets,
450 CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(curr_dyn_buckets), 0,
451 "Current number of buckets for states hash table.");
452 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, curr_max_length,
453 CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(curr_max_length), 0,
454 "Current maximum length of states chains in hash buckets.");
455 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_buckets,
456 CTLFLAG_VNET | CTLTYPE_U32 | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
457 0, 0, sysctl_dyn_buckets, "IU",
458 "Max number of buckets for dynamic states hash table.");
459 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_max,
460 CTLFLAG_VNET | CTLTYPE_U32 | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
461 0, 0, sysctl_dyn_max, "IU",
462 "Max number of dynamic states.");
463 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_parent_max,
464 CTLFLAG_VNET | CTLTYPE_U32 | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
465 0, 0, sysctl_dyn_parent_max, "IU",
466 "Max number of parent dynamic states.");
467 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime,
468 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_ack_lifetime), 0,
469 "Lifetime of dynamic states for TCP ACK.");
470 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime,
471 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_syn_lifetime), 0,
472 "Lifetime of dynamic states for TCP SYN.");
473 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime,
474 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_fin_lifetime), 0,
475 "Lifetime of dynamic states for TCP FIN.");
476 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime,
477 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_rst_lifetime), 0,
478 "Lifetime of dynamic states for TCP RST.");
479 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime,
480 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_udp_lifetime), 0,
481 "Lifetime of dynamic states for UDP.");
482 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime,
483 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_short_lifetime), 0,
484 "Lifetime of dynamic states for other situations.");
485 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_keepalive,
486 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_keepalive), 0,
487 "Enable keepalives for dynamic states.");
488 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_keep_states,
489 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_keep_states), 0,
490 "Do not flush dynamic states on rule deletion");
492 #ifdef IPFIREWALL_DYNDEBUG
493 #define DYN_DEBUG(fmt, ...) do { \
494 printf("%s: " fmt "\n", __func__, __VA_ARGS__); \
497 #define DYN_DEBUG(fmt, ...)
498 #endif /* !IPFIREWALL_DYNDEBUG */
501 /* Functions to work with IPv6 states */
502 static struct dyn_ipv6_state *dyn_lookup_ipv6_state(
503 const struct ipfw_flow_id *, uint32_t, const void *,
504 struct ipfw_dyn_info *, int);
505 static int dyn_lookup_ipv6_state_locked(const struct ipfw_flow_id *,
506 uint32_t, const void *, int, uint32_t, uint16_t);
507 static struct dyn_ipv6_state *dyn_alloc_ipv6_state(
508 const struct ipfw_flow_id *, uint32_t, uint16_t, uint8_t);
509 static int dyn_add_ipv6_state(void *, uint32_t, uint16_t,
510 const struct ipfw_flow_id *, uint32_t, const void *, int, uint32_t,
511 struct ipfw_dyn_info *, uint16_t, uint16_t, uint8_t);
512 static void dyn_export_ipv6_state(const struct dyn_ipv6_state *,
515 static uint32_t dyn_getscopeid(const struct ip_fw_args *);
516 static void dyn_make_keepalive_ipv6(struct mbuf *, const struct in6_addr *,
517 const struct in6_addr *, uint32_t, uint32_t, uint32_t, uint16_t,
519 static void dyn_enqueue_keepalive_ipv6(struct mbufq *,
520 const struct dyn_ipv6_state *);
521 static void dyn_send_keepalive_ipv6(struct ip_fw_chain *);
523 static struct dyn_ipv6_state *dyn_lookup_ipv6_parent(
524 const struct ipfw_flow_id *, uint32_t, const void *, uint32_t, uint16_t,
526 static struct dyn_ipv6_state *dyn_lookup_ipv6_parent_locked(
527 const struct ipfw_flow_id *, uint32_t, const void *, uint32_t, uint16_t,
529 static struct dyn_ipv6_state *dyn_add_ipv6_parent(void *, uint32_t, uint16_t,
530 const struct ipfw_flow_id *, uint32_t, uint32_t, uint32_t, uint16_t);
533 /* Functions to work with limit states */
534 static void *dyn_get_parent_state(const struct ipfw_flow_id *, uint32_t,
535 struct ip_fw *, uint32_t, uint32_t, uint16_t);
536 static struct dyn_ipv4_state *dyn_lookup_ipv4_parent(
537 const struct ipfw_flow_id *, const void *, uint32_t, uint16_t, uint32_t);
538 static struct dyn_ipv4_state *dyn_lookup_ipv4_parent_locked(
539 const struct ipfw_flow_id *, const void *, uint32_t, uint16_t, uint32_t);
540 static struct dyn_parent *dyn_alloc_parent(void *, uint32_t, uint16_t,
542 static struct dyn_ipv4_state *dyn_add_ipv4_parent(void *, uint32_t, uint16_t,
543 const struct ipfw_flow_id *, uint32_t, uint32_t, uint16_t);
545 static void dyn_tick(void *);
546 static void dyn_expire_states(struct ip_fw_chain *, ipfw_range_tlv *);
547 static void dyn_free_states(struct ip_fw_chain *);
548 static void dyn_export_parent(const struct dyn_parent *, uint16_t, uint8_t,
550 static void dyn_export_data(const struct dyn_data *, uint16_t, uint8_t,
551 uint8_t, ipfw_dyn_rule *);
552 static uint32_t dyn_update_tcp_state(struct dyn_data *,
553 const struct ipfw_flow_id *, const struct tcphdr *, int);
554 static void dyn_update_proto_state(struct dyn_data *,
555 const struct ipfw_flow_id *, const void *, int, int);
557 /* Functions to work with IPv4 states */
558 struct dyn_ipv4_state *dyn_lookup_ipv4_state(const struct ipfw_flow_id *,
559 const void *, struct ipfw_dyn_info *, int);
560 static int dyn_lookup_ipv4_state_locked(const struct ipfw_flow_id *,
561 const void *, int, uint32_t, uint16_t);
562 static struct dyn_ipv4_state *dyn_alloc_ipv4_state(
563 const struct ipfw_flow_id *, uint16_t, uint8_t);
564 static int dyn_add_ipv4_state(void *, uint32_t, uint16_t,
565 const struct ipfw_flow_id *, const void *, int, uint32_t,
566 struct ipfw_dyn_info *, uint16_t, uint16_t, uint8_t);
567 static void dyn_export_ipv4_state(const struct dyn_ipv4_state *,
571 * Named states support.
573 static char *default_state_name = "default";
574 struct dyn_state_obj {
575 struct named_object no;
579 #define DYN_STATE_OBJ(ch, cmd) \
580 ((struct dyn_state_obj *)SRV_OBJECT(ch, (cmd)->arg1))
582 * Classifier callback.
583 * Return 0 if opcode contains object that should be referenced
587 dyn_classify(ipfw_insn *cmd, uint16_t *puidx, uint8_t *ptype)
590 DYN_DEBUG("opcode %d, arg1 %d", cmd->opcode, cmd->arg1);
591 /* Don't rewrite "check-state any" */
592 if (cmd->arg1 == 0 &&
593 cmd->opcode == O_CHECK_STATE)
602 dyn_update(ipfw_insn *cmd, uint16_t idx)
606 DYN_DEBUG("opcode %d, arg1 %d", cmd->opcode, cmd->arg1);
610 dyn_findbyname(struct ip_fw_chain *ch, struct tid_info *ti,
611 struct named_object **pno)
616 DYN_DEBUG("uidx %d", ti->uidx);
618 if (ti->tlvs == NULL)
620 /* Search ntlv in the buffer provided by user */
621 ntlv = ipfw_find_name_tlv_type(ti->tlvs, ti->tlen, ti->uidx,
622 IPFW_TLV_STATE_NAME);
627 name = default_state_name;
629 * Search named object with corresponding name.
630 * Since states objects are global - ignore the set value
631 * and use zero instead.
633 *pno = ipfw_objhash_lookup_name_type(CHAIN_TO_SRV(ch), 0,
634 IPFW_TLV_STATE_NAME, name);
636 * We always return success here.
637 * The caller will check *pno and mark object as unresolved,
638 * then it will automatically create "default" object.
643 static struct named_object *
644 dyn_findbykidx(struct ip_fw_chain *ch, uint16_t idx)
647 DYN_DEBUG("kidx %d", idx);
648 return (ipfw_objhash_lookup_kidx(CHAIN_TO_SRV(ch), idx));
652 dyn_create(struct ip_fw_chain *ch, struct tid_info *ti,
655 struct namedobj_instance *ni;
656 struct dyn_state_obj *obj;
657 struct named_object *no;
661 DYN_DEBUG("uidx %d", ti->uidx);
663 if (ti->tlvs == NULL)
665 ntlv = ipfw_find_name_tlv_type(ti->tlvs, ti->tlen, ti->uidx,
666 IPFW_TLV_STATE_NAME);
671 name = default_state_name;
673 ni = CHAIN_TO_SRV(ch);
674 obj = malloc(sizeof(*obj), M_IPFW, M_WAITOK | M_ZERO);
675 obj->no.name = obj->name;
676 obj->no.etlv = IPFW_TLV_STATE_NAME;
677 strlcpy(obj->name, name, sizeof(obj->name));
680 no = ipfw_objhash_lookup_name_type(ni, 0,
681 IPFW_TLV_STATE_NAME, name);
684 * Object is already created.
685 * Just return its kidx and bump refcount.
691 DYN_DEBUG("\tfound kidx %d", *pkidx);
694 if (ipfw_objhash_alloc_idx(ni, &obj->no.kidx) != 0) {
695 DYN_DEBUG("\talloc_idx failed for %s", name);
700 ipfw_objhash_add(ni, &obj->no);
701 SRV_OBJECT(ch, obj->no.kidx) = obj;
703 *pkidx = obj->no.kidx;
705 DYN_DEBUG("\tcreated kidx %d", *pkidx);
710 dyn_destroy(struct ip_fw_chain *ch, struct named_object *no)
712 struct dyn_state_obj *obj;
714 IPFW_UH_WLOCK_ASSERT(ch);
716 KASSERT(no->etlv == IPFW_TLV_STATE_NAME,
717 ("%s: wrong object type %u", __func__, no->etlv));
718 KASSERT(no->refcnt == 1,
719 ("Destroying object '%s' (type %u, idx %u) with refcnt %u",
720 no->name, no->etlv, no->kidx, no->refcnt));
721 DYN_DEBUG("kidx %d", no->kidx);
722 obj = SRV_OBJECT(ch, no->kidx);
723 SRV_OBJECT(ch, no->kidx) = NULL;
724 ipfw_objhash_del(CHAIN_TO_SRV(ch), no);
725 ipfw_objhash_free_idx(CHAIN_TO_SRV(ch), no->kidx);
730 static struct opcode_obj_rewrite dyn_opcodes[] = {
732 O_KEEP_STATE, IPFW_TLV_STATE_NAME,
733 dyn_classify, dyn_update,
734 dyn_findbyname, dyn_findbykidx,
735 dyn_create, dyn_destroy
738 O_CHECK_STATE, IPFW_TLV_STATE_NAME,
739 dyn_classify, dyn_update,
740 dyn_findbyname, dyn_findbykidx,
741 dyn_create, dyn_destroy
744 O_PROBE_STATE, IPFW_TLV_STATE_NAME,
745 dyn_classify, dyn_update,
746 dyn_findbyname, dyn_findbykidx,
747 dyn_create, dyn_destroy
750 O_LIMIT, IPFW_TLV_STATE_NAME,
751 dyn_classify, dyn_update,
752 dyn_findbyname, dyn_findbykidx,
753 dyn_create, dyn_destroy
758 * IMPORTANT: the hash function for dynamic rules must be commutative
759 * in source and destination (ip,port), because rules are bidirectional
760 * and we want to find both in the same bucket.
762 #ifndef IPFIREWALL_JENKINSHASH
763 static __inline uint32_t
764 hash_packet(const struct ipfw_flow_id *id)
769 if (IS_IP6_FLOW_ID(id))
770 i = ntohl((id->dst_ip6.__u6_addr.__u6_addr32[2]) ^
771 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^
772 (id->src_ip6.__u6_addr.__u6_addr32[2]) ^
773 (id->src_ip6.__u6_addr.__u6_addr32[3]));
776 i = (id->dst_ip) ^ (id->src_ip);
777 i ^= (id->dst_port) ^ (id->src_port);
781 static __inline uint32_t
782 hash_parent(const struct ipfw_flow_id *id, const void *rule)
785 return (hash_packet(id) ^ ((uintptr_t)rule));
788 #else /* IPFIREWALL_JENKINSHASH */
790 VNET_DEFINE_STATIC(uint32_t, dyn_hashseed);
791 #define V_dyn_hashseed VNET(dyn_hashseed)
794 addrcmp4(const struct ipfw_flow_id *id)
797 if (id->src_ip < id->dst_ip)
799 if (id->src_ip > id->dst_ip)
801 if (id->src_port <= id->dst_port)
808 addrcmp6(const struct ipfw_flow_id *id)
812 ret = memcmp(&id->src_ip6, &id->dst_ip6, sizeof(struct in6_addr));
817 if (id->src_port <= id->dst_port)
822 static __inline uint32_t
823 hash_packet6(const struct ipfw_flow_id *id)
826 struct in6_addr addr[2];
830 if (addrcmp6(id) == 0) {
831 t6.addr[0] = id->src_ip6;
832 t6.addr[1] = id->dst_ip6;
833 t6.port[0] = id->src_port;
834 t6.port[1] = id->dst_port;
836 t6.addr[0] = id->dst_ip6;
837 t6.addr[1] = id->src_ip6;
838 t6.port[0] = id->dst_port;
839 t6.port[1] = id->src_port;
841 return (jenkins_hash32((const uint32_t *)&t6,
842 sizeof(t6) / sizeof(uint32_t), V_dyn_hashseed));
846 static __inline uint32_t
847 hash_packet(const struct ipfw_flow_id *id)
854 if (IS_IP4_FLOW_ID(id)) {
855 /* All fields are in host byte order */
856 if (addrcmp4(id) == 0) {
857 t4.addr[0] = id->src_ip;
858 t4.addr[1] = id->dst_ip;
859 t4.port[0] = id->src_port;
860 t4.port[1] = id->dst_port;
862 t4.addr[0] = id->dst_ip;
863 t4.addr[1] = id->src_ip;
864 t4.port[0] = id->dst_port;
865 t4.port[1] = id->src_port;
867 return (jenkins_hash32((const uint32_t *)&t4,
868 sizeof(t4) / sizeof(uint32_t), V_dyn_hashseed));
871 if (IS_IP6_FLOW_ID(id))
872 return (hash_packet6(id));
877 static __inline uint32_t
878 hash_parent(const struct ipfw_flow_id *id, const void *rule)
881 return (jenkins_hash32((const uint32_t *)&rule,
882 sizeof(rule) / sizeof(uint32_t), hash_packet(id)));
884 #endif /* IPFIREWALL_JENKINSHASH */
887 * Print customizable flow id description via log(9) facility.
890 print_dyn_rule_flags(const struct ipfw_flow_id *id, int dyn_type,
891 int log_flags, char *prefix, char *postfix)
895 char src[INET6_ADDRSTRLEN], dst[INET6_ADDRSTRLEN];
897 char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN];
901 if (IS_IP6_FLOW_ID(id)) {
902 ip6_sprintf(src, &id->src_ip6);
903 ip6_sprintf(dst, &id->dst_ip6);
907 da.s_addr = htonl(id->src_ip);
908 inet_ntop(AF_INET, &da, src, sizeof(src));
909 da.s_addr = htonl(id->dst_ip);
910 inet_ntop(AF_INET, &da, dst, sizeof(dst));
912 log(log_flags, "ipfw: %s type %d %s %d -> %s %d, %d %s\n",
913 prefix, dyn_type, src, id->src_port, dst,
914 id->dst_port, V_dyn_count, postfix);
917 #define print_dyn_rule(id, dtype, prefix, postfix) \
918 print_dyn_rule_flags(id, dtype, LOG_DEBUG, prefix, postfix)
920 #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
921 #define TIME_LE(a,b) ((int)((a)-(b)) < 0)
922 #define _SEQ_GE(a,b) ((int)((a)-(b)) >= 0)
923 #define BOTH_SYN (TH_SYN | (TH_SYN << 8))
924 #define BOTH_FIN (TH_FIN | (TH_FIN << 8))
925 #define TCP_FLAGS (TH_FLAGS | (TH_FLAGS << 8))
926 #define ACK_FWD 0x00010000 /* fwd ack seen */
927 #define ACK_REV 0x00020000 /* rev ack seen */
928 #define ACK_BOTH (ACK_FWD | ACK_REV)
931 dyn_update_tcp_state(struct dyn_data *data, const struct ipfw_flow_id *pkt,
932 const struct tcphdr *tcp, int dir)
934 uint32_t ack, expire;
938 expire = data->expire;
939 old = state = data->state;
940 th_flags = pkt->_flags & (TH_FIN | TH_SYN | TH_RST);
941 state |= (dir == MATCH_FORWARD) ? th_flags: (th_flags << 8);
942 switch (state & TCP_FLAGS) {
943 case TH_SYN: /* opening */
944 expire = time_uptime + V_dyn_syn_lifetime;
947 case BOTH_SYN: /* move to established */
948 case BOTH_SYN | TH_FIN: /* one side tries to close */
949 case BOTH_SYN | (TH_FIN << 8):
952 ack = ntohl(tcp->th_ack);
953 if (dir == MATCH_FORWARD) {
954 if (data->ack_fwd == 0 ||
955 _SEQ_GE(ack, data->ack_fwd)) {
957 if (data->ack_fwd != ack)
958 ck_pr_store_32(&data->ack_fwd, ack);
961 if (data->ack_rev == 0 ||
962 _SEQ_GE(ack, data->ack_rev)) {
964 if (data->ack_rev != ack)
965 ck_pr_store_32(&data->ack_rev, ack);
968 if ((state & ACK_BOTH) == ACK_BOTH) {
970 * Set expire time to V_dyn_ack_lifetime only if
971 * we got ACKs for both directions.
972 * We use XOR here to avoid possible state
973 * overwriting in concurrent thread.
975 expire = time_uptime + V_dyn_ack_lifetime;
976 ck_pr_xor_32(&data->state, ACK_BOTH);
977 } else if ((data->state & ACK_BOTH) != (state & ACK_BOTH))
978 ck_pr_or_32(&data->state, state & ACK_BOTH);
981 case BOTH_SYN | BOTH_FIN: /* both sides closed */
982 if (V_dyn_fin_lifetime >= V_dyn_keepalive_period)
983 V_dyn_fin_lifetime = V_dyn_keepalive_period - 1;
984 expire = time_uptime + V_dyn_fin_lifetime;
988 if (V_dyn_keepalive != 0 &&
989 V_dyn_rst_lifetime >= V_dyn_keepalive_period)
990 V_dyn_rst_lifetime = V_dyn_keepalive_period - 1;
991 expire = time_uptime + V_dyn_rst_lifetime;
993 /* Save TCP state if it was changed */
994 if ((state & TCP_FLAGS) != (old & TCP_FLAGS))
995 ck_pr_or_32(&data->state, state & TCP_FLAGS);
1000 * Update ULP specific state.
1001 * For TCP we keep sequence numbers and flags. For other protocols
1002 * currently we update only expire time. Packets and bytes counters
1003 * are also updated here.
1006 dyn_update_proto_state(struct dyn_data *data, const struct ipfw_flow_id *pkt,
1007 const void *ulp, int pktlen, int dir)
1011 /* NOTE: we are in critical section here. */
1012 switch (pkt->proto) {
1014 case IPPROTO_UDPLITE:
1015 expire = time_uptime + V_dyn_udp_lifetime;
1018 expire = dyn_update_tcp_state(data, pkt, ulp, dir);
1021 expire = time_uptime + V_dyn_short_lifetime;
1024 * Expiration timer has the per-second granularity, no need to update
1025 * it every time when state is matched.
1027 if (data->expire != expire)
1028 ck_pr_store_32(&data->expire, expire);
1030 if (dir == MATCH_FORWARD)
1031 DYN_COUNTER_INC(data, fwd, pktlen);
1033 DYN_COUNTER_INC(data, rev, pktlen);
1037 * Lookup IPv4 state.
1038 * Must be called in critical section.
1040 struct dyn_ipv4_state *
1041 dyn_lookup_ipv4_state(const struct ipfw_flow_id *pkt, const void *ulp,
1042 struct ipfw_dyn_info *info, int pktlen)
1044 struct dyn_ipv4_state *s;
1045 uint32_t version, bucket;
1047 bucket = DYN_BUCKET(info->hashval, V_curr_dyn_buckets);
1048 info->version = DYN_BUCKET_VERSION(bucket, ipv4_add);
1050 version = DYN_BUCKET_VERSION(bucket, ipv4_del);
1051 CK_SLIST_FOREACH(s, &V_dyn_ipv4[bucket], entry) {
1052 DYNSTATE_PROTECT(s);
1053 if (version != DYN_BUCKET_VERSION(bucket, ipv4_del))
1055 if (s->proto != pkt->proto)
1057 if (info->kidx != 0 && s->kidx != info->kidx)
1059 if (s->sport == pkt->src_port && s->dport == pkt->dst_port &&
1060 s->src == pkt->src_ip && s->dst == pkt->dst_ip) {
1061 info->direction = MATCH_FORWARD;
1064 if (s->sport == pkt->dst_port && s->dport == pkt->src_port &&
1065 s->src == pkt->dst_ip && s->dst == pkt->src_ip) {
1066 info->direction = MATCH_REVERSE;
1072 dyn_update_proto_state(s->data, pkt, ulp, pktlen,
1078 * Lookup IPv4 state.
1079 * Simplifed version is used to check that matching state doesn't exist.
1082 dyn_lookup_ipv4_state_locked(const struct ipfw_flow_id *pkt,
1083 const void *ulp, int pktlen, uint32_t bucket, uint16_t kidx)
1085 struct dyn_ipv4_state *s;
1089 DYN_BUCKET_ASSERT(bucket);
1090 CK_SLIST_FOREACH(s, &V_dyn_ipv4[bucket], entry) {
1091 if (s->proto != pkt->proto ||
1094 if (s->sport == pkt->src_port &&
1095 s->dport == pkt->dst_port &&
1096 s->src == pkt->src_ip && s->dst == pkt->dst_ip) {
1097 dir = MATCH_FORWARD;
1100 if (s->sport == pkt->dst_port && s->dport == pkt->src_port &&
1101 s->src == pkt->dst_ip && s->dst == pkt->src_ip) {
1102 dir = MATCH_REVERSE;
1107 dyn_update_proto_state(s->data, pkt, ulp, pktlen, dir);
1111 struct dyn_ipv4_state *
1112 dyn_lookup_ipv4_parent(const struct ipfw_flow_id *pkt, const void *rule,
1113 uint32_t ruleid, uint16_t rulenum, uint32_t hashval)
1115 struct dyn_ipv4_state *s;
1116 uint32_t version, bucket;
1118 bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets);
1120 version = DYN_BUCKET_VERSION(bucket, ipv4_parent_del);
1121 CK_SLIST_FOREACH(s, &V_dyn_ipv4_parent[bucket], entry) {
1122 DYNSTATE_PROTECT(s);
1123 if (version != DYN_BUCKET_VERSION(bucket, ipv4_parent_del))
1126 * NOTE: we do not need to check kidx, because parent rule
1127 * can not create states with different kidx.
1128 * And parent rule always created for forward direction.
1130 if (s->limit->parent == rule &&
1131 s->limit->ruleid == ruleid &&
1132 s->limit->rulenum == rulenum &&
1133 s->proto == pkt->proto &&
1134 s->sport == pkt->src_port &&
1135 s->dport == pkt->dst_port &&
1136 s->src == pkt->src_ip && s->dst == pkt->dst_ip) {
1137 if (s->limit->expire != time_uptime +
1138 V_dyn_short_lifetime)
1139 ck_pr_store_32(&s->limit->expire,
1140 time_uptime + V_dyn_short_lifetime);
1147 static struct dyn_ipv4_state *
1148 dyn_lookup_ipv4_parent_locked(const struct ipfw_flow_id *pkt,
1149 const void *rule, uint32_t ruleid, uint16_t rulenum, uint32_t bucket)
1151 struct dyn_ipv4_state *s;
1153 DYN_BUCKET_ASSERT(bucket);
1154 CK_SLIST_FOREACH(s, &V_dyn_ipv4_parent[bucket], entry) {
1155 if (s->limit->parent == rule &&
1156 s->limit->ruleid == ruleid &&
1157 s->limit->rulenum == rulenum &&
1158 s->proto == pkt->proto &&
1159 s->sport == pkt->src_port &&
1160 s->dport == pkt->dst_port &&
1161 s->src == pkt->src_ip && s->dst == pkt->dst_ip)
1169 dyn_getscopeid(const struct ip_fw_args *args)
1173 * If source or destination address is an scopeid address, we need
1174 * determine the scope zone id to resolve address scope ambiguity.
1176 if (IN6_IS_ADDR_LINKLOCAL(&args->f_id.src_ip6) ||
1177 IN6_IS_ADDR_LINKLOCAL(&args->f_id.dst_ip6))
1178 return (in6_getscopezone(args->ifp, IPV6_ADDR_SCOPE_LINKLOCAL));
1184 * Lookup IPv6 state.
1185 * Must be called in critical section.
1187 static struct dyn_ipv6_state *
1188 dyn_lookup_ipv6_state(const struct ipfw_flow_id *pkt, uint32_t zoneid,
1189 const void *ulp, struct ipfw_dyn_info *info, int pktlen)
1191 struct dyn_ipv6_state *s;
1192 uint32_t version, bucket;
1194 bucket = DYN_BUCKET(info->hashval, V_curr_dyn_buckets);
1195 info->version = DYN_BUCKET_VERSION(bucket, ipv6_add);
1197 version = DYN_BUCKET_VERSION(bucket, ipv6_del);
1198 CK_SLIST_FOREACH(s, &V_dyn_ipv6[bucket], entry) {
1199 DYNSTATE_PROTECT(s);
1200 if (version != DYN_BUCKET_VERSION(bucket, ipv6_del))
1202 if (s->proto != pkt->proto || s->zoneid != zoneid)
1204 if (info->kidx != 0 && s->kidx != info->kidx)
1206 if (s->sport == pkt->src_port && s->dport == pkt->dst_port &&
1207 IN6_ARE_ADDR_EQUAL(&s->src, &pkt->src_ip6) &&
1208 IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->dst_ip6)) {
1209 info->direction = MATCH_FORWARD;
1212 if (s->sport == pkt->dst_port && s->dport == pkt->src_port &&
1213 IN6_ARE_ADDR_EQUAL(&s->src, &pkt->dst_ip6) &&
1214 IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->src_ip6)) {
1215 info->direction = MATCH_REVERSE;
1220 dyn_update_proto_state(s->data, pkt, ulp, pktlen,
1226 * Lookup IPv6 state.
1227 * Simplifed version is used to check that matching state doesn't exist.
1230 dyn_lookup_ipv6_state_locked(const struct ipfw_flow_id *pkt, uint32_t zoneid,
1231 const void *ulp, int pktlen, uint32_t bucket, uint16_t kidx)
1233 struct dyn_ipv6_state *s;
1237 DYN_BUCKET_ASSERT(bucket);
1238 CK_SLIST_FOREACH(s, &V_dyn_ipv6[bucket], entry) {
1239 if (s->proto != pkt->proto || s->kidx != kidx ||
1240 s->zoneid != zoneid)
1242 if (s->sport == pkt->src_port && s->dport == pkt->dst_port &&
1243 IN6_ARE_ADDR_EQUAL(&s->src, &pkt->src_ip6) &&
1244 IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->dst_ip6)) {
1245 dir = MATCH_FORWARD;
1248 if (s->sport == pkt->dst_port && s->dport == pkt->src_port &&
1249 IN6_ARE_ADDR_EQUAL(&s->src, &pkt->dst_ip6) &&
1250 IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->src_ip6)) {
1251 dir = MATCH_REVERSE;
1256 dyn_update_proto_state(s->data, pkt, ulp, pktlen, dir);
1260 static struct dyn_ipv6_state *
1261 dyn_lookup_ipv6_parent(const struct ipfw_flow_id *pkt, uint32_t zoneid,
1262 const void *rule, uint32_t ruleid, uint16_t rulenum, uint32_t hashval)
1264 struct dyn_ipv6_state *s;
1265 uint32_t version, bucket;
1267 bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets);
1269 version = DYN_BUCKET_VERSION(bucket, ipv6_parent_del);
1270 CK_SLIST_FOREACH(s, &V_dyn_ipv6_parent[bucket], entry) {
1271 DYNSTATE_PROTECT(s);
1272 if (version != DYN_BUCKET_VERSION(bucket, ipv6_parent_del))
1275 * NOTE: we do not need to check kidx, because parent rule
1276 * can not create states with different kidx.
1277 * Also parent rule always created for forward direction.
1279 if (s->limit->parent == rule &&
1280 s->limit->ruleid == ruleid &&
1281 s->limit->rulenum == rulenum &&
1282 s->proto == pkt->proto &&
1283 s->sport == pkt->src_port &&
1284 s->dport == pkt->dst_port && s->zoneid == zoneid &&
1285 IN6_ARE_ADDR_EQUAL(&s->src, &pkt->src_ip6) &&
1286 IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->dst_ip6)) {
1287 if (s->limit->expire != time_uptime +
1288 V_dyn_short_lifetime)
1289 ck_pr_store_32(&s->limit->expire,
1290 time_uptime + V_dyn_short_lifetime);
1297 static struct dyn_ipv6_state *
1298 dyn_lookup_ipv6_parent_locked(const struct ipfw_flow_id *pkt, uint32_t zoneid,
1299 const void *rule, uint32_t ruleid, uint16_t rulenum, uint32_t bucket)
1301 struct dyn_ipv6_state *s;
1303 DYN_BUCKET_ASSERT(bucket);
1304 CK_SLIST_FOREACH(s, &V_dyn_ipv6_parent[bucket], entry) {
1305 if (s->limit->parent == rule &&
1306 s->limit->ruleid == ruleid &&
1307 s->limit->rulenum == rulenum &&
1308 s->proto == pkt->proto &&
1309 s->sport == pkt->src_port &&
1310 s->dport == pkt->dst_port && s->zoneid == zoneid &&
1311 IN6_ARE_ADDR_EQUAL(&s->src, &pkt->src_ip6) &&
1312 IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->dst_ip6))
1321 * Lookup dynamic state.
1322 * pkt - filled by ipfw_chk() ipfw_flow_id;
1323 * ulp - determined by ipfw_chk() upper level protocol header;
1324 * dyn_info - info about matched state to return back;
1325 * Returns pointer to state's parent rule and dyn_info. If there is
1326 * no state, NULL is returned.
1327 * On match ipfw_dyn_lookup() updates state's counters.
1330 ipfw_dyn_lookup_state(const struct ip_fw_args *args, const void *ulp,
1331 int pktlen, const ipfw_insn *cmd, struct ipfw_dyn_info *info)
1333 struct dyn_data *data;
1336 IPFW_RLOCK_ASSERT(&V_layer3_chain);
1340 info->kidx = cmd->arg1;
1341 info->direction = MATCH_NONE;
1342 info->hashval = hash_packet(&args->f_id);
1344 DYNSTATE_CRITICAL_ENTER();
1345 if (IS_IP4_FLOW_ID(&args->f_id)) {
1346 struct dyn_ipv4_state *s;
1348 s = dyn_lookup_ipv4_state(&args->f_id, ulp, info, pktlen);
1351 * Dynamic states are created using the same 5-tuple,
1352 * so it is assumed, that parent rule for O_LIMIT
1353 * state has the same address family.
1356 if (s->type == O_LIMIT) {
1358 rule = s->limit->parent;
1360 rule = data->parent;
1364 else if (IS_IP6_FLOW_ID(&args->f_id)) {
1365 struct dyn_ipv6_state *s;
1367 s = dyn_lookup_ipv6_state(&args->f_id, dyn_getscopeid(args),
1371 if (s->type == O_LIMIT) {
1373 rule = s->limit->parent;
1375 rule = data->parent;
1381 * If cached chain id is the same, we can avoid rule index
1382 * lookup. Otherwise do lookup and update chain_id and f_pos.
1383 * It is safe even if there is concurrent thread that want
1384 * update the same state, because chain->id can be changed
1385 * only under IPFW_WLOCK().
1387 if (data->chain_id != V_layer3_chain.id) {
1388 data->f_pos = ipfw_find_rule(&V_layer3_chain,
1389 data->rulenum, data->ruleid);
1391 * Check that found state has not orphaned.
1392 * When chain->id being changed the parent
1393 * rule can be deleted. If found rule doesn't
1394 * match the parent pointer, consider this
1395 * result as MATCH_NONE and return NULL.
1397 * This will lead to creation of new similar state
1398 * that will be added into head of this bucket.
1399 * And the state that we currently have matched
1400 * should be deleted by dyn_expire_states().
1402 * In case when dyn_keep_states is enabled, return
1403 * pointer to deleted rule and f_pos value
1404 * corresponding to penultimate rule.
1405 * When we have enabled V_dyn_keep_states, states
1406 * that become orphaned will get the DYN_REFERENCED
1407 * flag and rule will keep around. So we can return
1408 * it. But since it is not in the rules map, we need
1409 * return such f_pos value, so after the state
1410 * handling if the search will continue, the next rule
1411 * will be the last one - the default rule.
1413 if (V_layer3_chain.map[data->f_pos] == rule) {
1414 data->chain_id = V_layer3_chain.id;
1415 info->f_pos = data->f_pos;
1416 } else if (V_dyn_keep_states != 0) {
1418 * The original rule pointer is still usable.
1419 * So, we return it, but f_pos need to be
1420 * changed to point to the penultimate rule.
1422 MPASS(V_layer3_chain.n_rules > 1);
1423 data->chain_id = V_layer3_chain.id;
1424 data->f_pos = V_layer3_chain.n_rules - 2;
1425 info->f_pos = data->f_pos;
1428 info->direction = MATCH_NONE;
1429 DYN_DEBUG("rule %p [%u, %u] is considered "
1430 "invalid in data %p", rule, data->ruleid,
1431 data->rulenum, data);
1432 /* info->f_pos doesn't matter here. */
1435 info->f_pos = data->f_pos;
1437 DYNSTATE_CRITICAL_EXIT();
1440 * Return MATCH_NONE if parent rule is in disabled set.
1441 * This will lead to creation of new similar state that
1442 * will be added into head of this bucket.
1444 * XXXAE: we need to be able update state's set when parent
1445 * rule set is changed.
1447 if (rule != NULL && (V_set_disable & (1 << rule->set))) {
1449 info->direction = MATCH_NONE;
1455 static struct dyn_parent *
1456 dyn_alloc_parent(void *parent, uint32_t ruleid, uint16_t rulenum,
1459 struct dyn_parent *limit;
1461 limit = uma_zalloc(V_dyn_parent_zone, M_NOWAIT | M_ZERO);
1462 if (limit == NULL) {
1463 if (last_log != time_uptime) {
1464 last_log = time_uptime;
1466 "ipfw: Cannot allocate parent dynamic state, "
1467 "consider increasing "
1468 "net.inet.ip.fw.dyn_parent_max\n");
1473 limit->parent = parent;
1474 limit->ruleid = ruleid;
1475 limit->rulenum = rulenum;
1476 limit->hashval = hashval;
1477 limit->expire = time_uptime + V_dyn_short_lifetime;
1481 static struct dyn_data *
1482 dyn_alloc_dyndata(void *parent, uint32_t ruleid, uint16_t rulenum,
1483 const struct ipfw_flow_id *pkt, const void *ulp, int pktlen,
1484 uint32_t hashval, uint16_t fibnum)
1486 struct dyn_data *data;
1488 data = uma_zalloc(V_dyn_data_zone, M_NOWAIT | M_ZERO);
1490 if (last_log != time_uptime) {
1491 last_log = time_uptime;
1493 "ipfw: Cannot allocate dynamic state, "
1494 "consider increasing net.inet.ip.fw.dyn_max\n");
1499 data->parent = parent;
1500 data->ruleid = ruleid;
1501 data->rulenum = rulenum;
1502 data->fibnum = fibnum;
1503 data->hashval = hashval;
1504 data->expire = time_uptime + V_dyn_syn_lifetime;
1505 dyn_update_proto_state(data, pkt, ulp, pktlen, MATCH_FORWARD);
1509 static struct dyn_ipv4_state *
1510 dyn_alloc_ipv4_state(const struct ipfw_flow_id *pkt, uint16_t kidx,
1513 struct dyn_ipv4_state *s;
1515 s = uma_zalloc(V_dyn_ipv4_zone, M_NOWAIT | M_ZERO);
1521 s->proto = pkt->proto;
1522 s->sport = pkt->src_port;
1523 s->dport = pkt->dst_port;
1524 s->src = pkt->src_ip;
1525 s->dst = pkt->dst_ip;
1530 * Add IPv4 parent state.
1531 * Returns pointer to parent state. When it is not NULL we are in
1532 * critical section and pointer protected by hazard pointer.
1533 * When some error occurs, it returns NULL and exit from critical section
1536 static struct dyn_ipv4_state *
1537 dyn_add_ipv4_parent(void *rule, uint32_t ruleid, uint16_t rulenum,
1538 const struct ipfw_flow_id *pkt, uint32_t hashval, uint32_t version,
1541 struct dyn_ipv4_state *s;
1542 struct dyn_parent *limit;
1545 bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets);
1546 DYN_BUCKET_LOCK(bucket);
1547 if (version != DYN_BUCKET_VERSION(bucket, ipv4_parent_add)) {
1549 * Bucket version has been changed since last lookup,
1550 * do lookup again to be sure that state does not exist.
1552 s = dyn_lookup_ipv4_parent_locked(pkt, rule, ruleid,
1556 * Simultaneous thread has already created this
1557 * state. Just return it.
1559 DYNSTATE_CRITICAL_ENTER();
1560 DYNSTATE_PROTECT(s);
1561 DYN_BUCKET_UNLOCK(bucket);
1566 limit = dyn_alloc_parent(rule, ruleid, rulenum, hashval);
1567 if (limit == NULL) {
1568 DYN_BUCKET_UNLOCK(bucket);
1572 s = dyn_alloc_ipv4_state(pkt, kidx, O_LIMIT_PARENT);
1574 DYN_BUCKET_UNLOCK(bucket);
1575 uma_zfree(V_dyn_parent_zone, limit);
1580 CK_SLIST_INSERT_HEAD(&V_dyn_ipv4_parent[bucket], s, entry);
1581 DYN_COUNT_INC(dyn_parent_count);
1582 DYN_BUCKET_VERSION_BUMP(bucket, ipv4_parent_add);
1583 DYNSTATE_CRITICAL_ENTER();
1584 DYNSTATE_PROTECT(s);
1585 DYN_BUCKET_UNLOCK(bucket);
1590 dyn_add_ipv4_state(void *parent, uint32_t ruleid, uint16_t rulenum,
1591 const struct ipfw_flow_id *pkt, const void *ulp, int pktlen,
1592 uint32_t hashval, struct ipfw_dyn_info *info, uint16_t fibnum,
1593 uint16_t kidx, uint8_t type)
1595 struct dyn_ipv4_state *s;
1599 bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets);
1600 DYN_BUCKET_LOCK(bucket);
1601 if (info->direction == MATCH_UNKNOWN ||
1602 info->kidx != kidx ||
1603 info->hashval != hashval ||
1604 info->version != DYN_BUCKET_VERSION(bucket, ipv4_add)) {
1606 * Bucket version has been changed since last lookup,
1607 * do lookup again to be sure that state does not exist.
1609 if (dyn_lookup_ipv4_state_locked(pkt, ulp, pktlen,
1610 bucket, kidx) != 0) {
1611 DYN_BUCKET_UNLOCK(bucket);
1616 data = dyn_alloc_dyndata(parent, ruleid, rulenum, pkt, ulp,
1617 pktlen, hashval, fibnum);
1619 DYN_BUCKET_UNLOCK(bucket);
1623 s = dyn_alloc_ipv4_state(pkt, kidx, type);
1625 DYN_BUCKET_UNLOCK(bucket);
1626 uma_zfree(V_dyn_data_zone, data);
1631 CK_SLIST_INSERT_HEAD(&V_dyn_ipv4[bucket], s, entry);
1632 DYN_COUNT_INC(dyn_count);
1633 DYN_BUCKET_VERSION_BUMP(bucket, ipv4_add);
1634 DYN_BUCKET_UNLOCK(bucket);
1639 static struct dyn_ipv6_state *
1640 dyn_alloc_ipv6_state(const struct ipfw_flow_id *pkt, uint32_t zoneid,
1641 uint16_t kidx, uint8_t type)
1643 struct dyn_ipv6_state *s;
1645 s = uma_zalloc(V_dyn_ipv6_zone, M_NOWAIT | M_ZERO);
1652 s->proto = pkt->proto;
1653 s->sport = pkt->src_port;
1654 s->dport = pkt->dst_port;
1655 s->src = pkt->src_ip6;
1656 s->dst = pkt->dst_ip6;
1661 * Add IPv6 parent state.
1662 * Returns pointer to parent state. When it is not NULL we are in
1663 * critical section and pointer protected by hazard pointer.
1664 * When some error occurs, it return NULL and exit from critical section
1667 static struct dyn_ipv6_state *
1668 dyn_add_ipv6_parent(void *rule, uint32_t ruleid, uint16_t rulenum,
1669 const struct ipfw_flow_id *pkt, uint32_t zoneid, uint32_t hashval,
1670 uint32_t version, uint16_t kidx)
1672 struct dyn_ipv6_state *s;
1673 struct dyn_parent *limit;
1676 bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets);
1677 DYN_BUCKET_LOCK(bucket);
1678 if (version != DYN_BUCKET_VERSION(bucket, ipv6_parent_add)) {
1680 * Bucket version has been changed since last lookup,
1681 * do lookup again to be sure that state does not exist.
1683 s = dyn_lookup_ipv6_parent_locked(pkt, zoneid, rule, ruleid,
1687 * Simultaneous thread has already created this
1688 * state. Just return it.
1690 DYNSTATE_CRITICAL_ENTER();
1691 DYNSTATE_PROTECT(s);
1692 DYN_BUCKET_UNLOCK(bucket);
1697 limit = dyn_alloc_parent(rule, ruleid, rulenum, hashval);
1698 if (limit == NULL) {
1699 DYN_BUCKET_UNLOCK(bucket);
1703 s = dyn_alloc_ipv6_state(pkt, zoneid, kidx, O_LIMIT_PARENT);
1705 DYN_BUCKET_UNLOCK(bucket);
1706 uma_zfree(V_dyn_parent_zone, limit);
1711 CK_SLIST_INSERT_HEAD(&V_dyn_ipv6_parent[bucket], s, entry);
1712 DYN_COUNT_INC(dyn_parent_count);
1713 DYN_BUCKET_VERSION_BUMP(bucket, ipv6_parent_add);
1714 DYNSTATE_CRITICAL_ENTER();
1715 DYNSTATE_PROTECT(s);
1716 DYN_BUCKET_UNLOCK(bucket);
1721 dyn_add_ipv6_state(void *parent, uint32_t ruleid, uint16_t rulenum,
1722 const struct ipfw_flow_id *pkt, uint32_t zoneid, const void *ulp,
1723 int pktlen, uint32_t hashval, struct ipfw_dyn_info *info,
1724 uint16_t fibnum, uint16_t kidx, uint8_t type)
1726 struct dyn_ipv6_state *s;
1727 struct dyn_data *data;
1730 bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets);
1731 DYN_BUCKET_LOCK(bucket);
1732 if (info->direction == MATCH_UNKNOWN ||
1733 info->kidx != kidx ||
1734 info->hashval != hashval ||
1735 info->version != DYN_BUCKET_VERSION(bucket, ipv6_add)) {
1737 * Bucket version has been changed since last lookup,
1738 * do lookup again to be sure that state does not exist.
1740 if (dyn_lookup_ipv6_state_locked(pkt, zoneid, ulp, pktlen,
1741 bucket, kidx) != 0) {
1742 DYN_BUCKET_UNLOCK(bucket);
1747 data = dyn_alloc_dyndata(parent, ruleid, rulenum, pkt, ulp,
1748 pktlen, hashval, fibnum);
1750 DYN_BUCKET_UNLOCK(bucket);
1754 s = dyn_alloc_ipv6_state(pkt, zoneid, kidx, type);
1756 DYN_BUCKET_UNLOCK(bucket);
1757 uma_zfree(V_dyn_data_zone, data);
1762 CK_SLIST_INSERT_HEAD(&V_dyn_ipv6[bucket], s, entry);
1763 DYN_COUNT_INC(dyn_count);
1764 DYN_BUCKET_VERSION_BUMP(bucket, ipv6_add);
1765 DYN_BUCKET_UNLOCK(bucket);
1771 dyn_get_parent_state(const struct ipfw_flow_id *pkt, uint32_t zoneid,
1772 struct ip_fw *rule, uint32_t hashval, uint32_t limit, uint16_t kidx)
1775 struct dyn_parent *p;
1777 uint32_t bucket, version;
1781 bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets);
1782 DYNSTATE_CRITICAL_ENTER();
1783 if (IS_IP4_FLOW_ID(pkt)) {
1784 struct dyn_ipv4_state *s;
1786 version = DYN_BUCKET_VERSION(bucket, ipv4_parent_add);
1787 s = dyn_lookup_ipv4_parent(pkt, rule, rule->id,
1788 rule->rulenum, bucket);
1791 * Exit from critical section because dyn_add_parent()
1792 * will acquire bucket lock.
1794 DYNSTATE_CRITICAL_EXIT();
1796 s = dyn_add_ipv4_parent(rule, rule->id,
1797 rule->rulenum, pkt, hashval, version, kidx);
1800 /* Now we are in critical section again. */
1806 else if (IS_IP6_FLOW_ID(pkt)) {
1807 struct dyn_ipv6_state *s;
1809 version = DYN_BUCKET_VERSION(bucket, ipv6_parent_add);
1810 s = dyn_lookup_ipv6_parent(pkt, zoneid, rule, rule->id,
1811 rule->rulenum, bucket);
1814 * Exit from critical section because dyn_add_parent()
1815 * can acquire bucket mutex.
1817 DYNSTATE_CRITICAL_EXIT();
1819 s = dyn_add_ipv6_parent(rule, rule->id,
1820 rule->rulenum, pkt, zoneid, hashval, version,
1824 /* Now we are in critical section again. */
1831 DYNSTATE_CRITICAL_EXIT();
1835 /* Check the limit */
1836 if (DPARENT_COUNT(p) >= limit) {
1837 DYNSTATE_CRITICAL_EXIT();
1838 if (V_fw_verbose && last_log != time_uptime) {
1839 last_log = time_uptime;
1840 snprintf(sbuf, sizeof(sbuf), "%u drop session",
1842 print_dyn_rule_flags(pkt, O_LIMIT,
1843 LOG_SECURITY | LOG_DEBUG, sbuf,
1844 "too many entries");
1849 /* Take new session into account. */
1850 DPARENT_COUNT_INC(p);
1852 * We must exit from critical section because the following code
1853 * can acquire bucket mutex.
1854 * We rely on the the 'count' field. The state will not expire
1855 * until it has some child states, i.e. 'count' field is not zero.
1856 * Return state pointer, it will be used by child states as parent.
1858 DYNSTATE_CRITICAL_EXIT();
1863 dyn_install_state(const struct ipfw_flow_id *pkt, uint32_t zoneid,
1864 uint16_t fibnum, const void *ulp, int pktlen, struct ip_fw *rule,
1865 struct ipfw_dyn_info *info, uint32_t limit, uint16_t limit_mask,
1866 uint16_t kidx, uint8_t type)
1868 struct ipfw_flow_id id;
1869 uint32_t hashval, parent_hashval, ruleid, rulenum;
1872 MPASS(type == O_LIMIT || type == O_KEEP_STATE);
1875 rulenum = rule->rulenum;
1876 if (type == O_LIMIT) {
1877 /* Create masked flow id and calculate bucket */
1878 id.addr_type = pkt->addr_type;
1879 id.proto = pkt->proto;
1880 id.fib = fibnum; /* unused */
1881 id.src_port = (limit_mask & DYN_SRC_PORT) ?
1883 id.dst_port = (limit_mask & DYN_DST_PORT) ?
1885 if (IS_IP4_FLOW_ID(pkt)) {
1886 id.src_ip = (limit_mask & DYN_SRC_ADDR) ?
1888 id.dst_ip = (limit_mask & DYN_DST_ADDR) ?
1892 else if (IS_IP6_FLOW_ID(pkt)) {
1893 if (limit_mask & DYN_SRC_ADDR)
1894 id.src_ip6 = pkt->src_ip6;
1896 memset(&id.src_ip6, 0, sizeof(id.src_ip6));
1897 if (limit_mask & DYN_DST_ADDR)
1898 id.dst_ip6 = pkt->dst_ip6;
1900 memset(&id.dst_ip6, 0, sizeof(id.dst_ip6));
1904 return (EAFNOSUPPORT);
1906 parent_hashval = hash_parent(&id, rule);
1907 rule = dyn_get_parent_state(&id, zoneid, rule, parent_hashval,
1911 if (V_fw_verbose && last_log != time_uptime) {
1912 last_log = time_uptime;
1913 snprintf(sbuf, sizeof(sbuf),
1914 "%u drop session", rule->rulenum);
1915 print_dyn_rule_flags(pkt, O_LIMIT,
1916 LOG_SECURITY | LOG_DEBUG, sbuf,
1917 "too many entries");
1923 * Limit is not reached, create new state.
1924 * Now rule points to parent state.
1928 hashval = hash_packet(pkt);
1929 if (IS_IP4_FLOW_ID(pkt))
1930 ret = dyn_add_ipv4_state(rule, ruleid, rulenum, pkt,
1931 ulp, pktlen, hashval, info, fibnum, kidx, type);
1933 else if (IS_IP6_FLOW_ID(pkt))
1934 ret = dyn_add_ipv6_state(rule, ruleid, rulenum, pkt,
1935 zoneid, ulp, pktlen, hashval, info, fibnum, kidx, type);
1940 if (type == O_LIMIT) {
1943 * We failed to create child state for O_LIMIT
1944 * opcode. Since we already counted it in the parent,
1945 * we must revert counter back. The 'rule' points to
1946 * parent state, use it to get dyn_parent.
1948 * XXXAE: it should be safe to use 'rule' pointer
1949 * without extra lookup, parent state is referenced
1950 * and should not be freed.
1952 if (IS_IP4_FLOW_ID(&id))
1954 ((struct dyn_ipv4_state *)rule)->limit);
1956 else if (IS_IP6_FLOW_ID(&id))
1958 ((struct dyn_ipv6_state *)rule)->limit);
1963 * EEXIST means that simultaneous thread has created this
1964 * state. Consider this as success.
1966 * XXXAE: should we invalidate 'info' content here?
1974 * Install dynamic state.
1975 * chain - ipfw's instance;
1976 * rule - the parent rule that installs the state;
1977 * cmd - opcode that installs the state;
1978 * args - ipfw arguments;
1979 * ulp - upper level protocol header;
1980 * pktlen - packet length;
1981 * info - dynamic state lookup info;
1982 * tablearg - tablearg id.
1984 * Returns non-zero value (failure) if state is not installed because
1985 * of errors or because session limitations are enforced.
1988 ipfw_dyn_install_state(struct ip_fw_chain *chain, struct ip_fw *rule,
1989 const ipfw_insn_limit *cmd, const struct ip_fw_args *args,
1990 const void *ulp, int pktlen, struct ipfw_dyn_info *info,
1994 uint16_t limit_mask;
1996 if (cmd->o.opcode == O_LIMIT) {
1997 limit = IP_FW_ARG_TABLEARG(chain, cmd->conn_limit, limit);
1998 limit_mask = cmd->limit_mask;
2003 return (dyn_install_state(&args->f_id,
2005 IS_IP6_FLOW_ID(&args->f_id) ? dyn_getscopeid(args):
2007 0, M_GETFIB(args->m), ulp, pktlen, rule, info, limit,
2008 limit_mask, cmd->o.arg1, cmd->o.opcode));
2012 * Free safe to remove state entries from expired lists.
2015 dyn_free_states(struct ip_fw_chain *chain)
2017 struct dyn_ipv4_state *s4, *s4n;
2019 struct dyn_ipv6_state *s6, *s6n;
2021 int cached_count, i;
2024 * We keep pointers to objects that are in use on each CPU
2025 * in the per-cpu dyn_hp pointer. When object is going to be
2026 * removed, first of it is unlinked from the corresponding
2027 * list. This leads to changing of dyn_bucket_xxx_delver version.
2028 * Unlinked objects is placed into corresponding dyn_expired_xxx
2029 * list. Reader that is going to dereference object pointer checks
2030 * dyn_bucket_xxx_delver version before and after storing pointer
2031 * into dyn_hp. If version is the same, the object is protected
2032 * from freeing and it is safe to dereference. Othervise reader
2033 * tries to iterate list again from the beginning, but this object
2034 * now unlinked and thus will not be accessible.
2036 * Copy dyn_hp pointers for each CPU into dyn_hp_cache array.
2037 * It does not matter that some pointer can be changed in
2038 * time while we are copying. We need to check, that objects
2039 * removed in the previous pass are not in use. And if dyn_hp
2040 * pointer does not contain it in the time when we are copying,
2041 * it will not appear there, because it is already unlinked.
2042 * And for new pointers we will not free objects that will be
2043 * unlinked in this pass.
2047 dyn_hp_cache[cached_count] = DYNSTATE_GET(i);
2048 if (dyn_hp_cache[cached_count] != NULL)
2053 * Free expired states that are safe to free.
2054 * Check each entry from previous pass in the dyn_expired_xxx
2055 * list, if pointer to the object is in the dyn_hp_cache array,
2056 * keep it until next pass. Otherwise it is safe to free the
2059 * XXXAE: optimize this to use SLIST_REMOVE_AFTER.
2061 #define DYN_FREE_STATES(s, next, name) do { \
2062 s = SLIST_FIRST(&V_dyn_expired_ ## name); \
2063 while (s != NULL) { \
2064 next = SLIST_NEXT(s, expired); \
2065 for (i = 0; i < cached_count; i++) \
2066 if (dyn_hp_cache[i] == s) \
2068 if (i == cached_count) { \
2069 if (s->type == O_LIMIT_PARENT && \
2070 s->limit->count != 0) { \
2074 SLIST_REMOVE(&V_dyn_expired_ ## name, \
2075 s, dyn_ ## name ## _state, expired); \
2076 if (s->type == O_LIMIT_PARENT) \
2077 uma_zfree(V_dyn_parent_zone, s->limit); \
2079 uma_zfree(V_dyn_data_zone, s->data); \
2080 uma_zfree(V_dyn_ ## name ## _zone, s); \
2087 * Protect access to expired lists with DYN_EXPIRED_LOCK.
2088 * Userland can invoke ipfw_expire_dyn_states() to delete
2089 * specific states, this will lead to modification of expired
2092 * XXXAE: do we need DYN_EXPIRED_LOCK? We can just use
2093 * IPFW_UH_WLOCK to protect access to these lists.
2096 DYN_FREE_STATES(s4, s4n, ipv4);
2098 DYN_FREE_STATES(s6, s6n, ipv6);
2100 DYN_EXPIRED_UNLOCK();
2101 #undef DYN_FREE_STATES
2106 * 0 when state is not matched by specified range;
2107 * 1 when state is matched by specified range;
2108 * 2 when state is matched by specified range and requested deletion of
2112 dyn_match_range(uint16_t rulenum, uint8_t set, const ipfw_range_tlv *rt)
2116 /* flush all states */
2117 if (rt->flags & IPFW_RCFLAG_ALL) {
2118 if (rt->flags & IPFW_RCFLAG_DYNAMIC)
2119 return (2); /* forced */
2122 if ((rt->flags & IPFW_RCFLAG_SET) != 0 && set != rt->set)
2124 if ((rt->flags & IPFW_RCFLAG_RANGE) != 0 &&
2125 (rulenum < rt->start_rule || rulenum > rt->end_rule))
2127 if (rt->flags & IPFW_RCFLAG_DYNAMIC)
2133 dyn_acquire_rule(struct ip_fw_chain *ch, struct dyn_data *data,
2134 struct ip_fw *rule, uint16_t kidx)
2136 struct dyn_state_obj *obj;
2139 * Do not acquire reference twice.
2140 * This can happen when rule deletion executed for
2141 * the same range, but different ruleset id.
2143 if (data->flags & DYN_REFERENCED)
2146 IPFW_UH_WLOCK_ASSERT(ch);
2149 data->flags |= DYN_REFERENCED;
2150 /* Reference the named object */
2151 obj = SRV_OBJECT(ch, kidx);
2153 MPASS(obj->no.etlv == IPFW_TLV_STATE_NAME);
2155 /* Reference the parent rule */
2160 dyn_release_rule(struct ip_fw_chain *ch, struct dyn_data *data,
2161 struct ip_fw *rule, uint16_t kidx)
2163 struct dyn_state_obj *obj;
2165 IPFW_UH_WLOCK_ASSERT(ch);
2168 obj = SRV_OBJECT(ch, kidx);
2169 if (obj->no.refcnt == 1)
2170 dyn_destroy(ch, &obj->no);
2174 if (--rule->refcnt == 1)
2175 ipfw_free_rule(rule);
2179 * We do not keep O_LIMIT_PARENT states when V_dyn_keep_states is enabled.
2180 * O_LIMIT state is created when new connection is going to be established
2181 * and there is no matching state. So, since the old parent rule was deleted
2182 * we can't create new states with old parent, and thus we can not account
2183 * new connections with already established connections, and can not do
2187 dyn_match_ipv4_state(struct ip_fw_chain *ch, struct dyn_ipv4_state *s,
2188 const ipfw_range_tlv *rt)
2193 if (s->type == O_LIMIT_PARENT) {
2194 rule = s->limit->parent;
2195 return (dyn_match_range(s->limit->rulenum, rule->set, rt));
2198 rule = s->data->parent;
2199 if (s->type == O_LIMIT)
2200 rule = ((struct dyn_ipv4_state *)rule)->limit->parent;
2202 ret = dyn_match_range(s->data->rulenum, rule->set, rt);
2203 if (ret == 0 || V_dyn_keep_states == 0 || ret > 1)
2206 dyn_acquire_rule(ch, s->data, rule, s->kidx);
2212 dyn_match_ipv6_state(struct ip_fw_chain *ch, struct dyn_ipv6_state *s,
2213 const ipfw_range_tlv *rt)
2218 if (s->type == O_LIMIT_PARENT) {
2219 rule = s->limit->parent;
2220 return (dyn_match_range(s->limit->rulenum, rule->set, rt));
2223 rule = s->data->parent;
2224 if (s->type == O_LIMIT)
2225 rule = ((struct dyn_ipv6_state *)rule)->limit->parent;
2227 ret = dyn_match_range(s->data->rulenum, rule->set, rt);
2228 if (ret == 0 || V_dyn_keep_states == 0 || ret > 1)
2231 dyn_acquire_rule(ch, s->data, rule, s->kidx);
2237 * Unlink expired entries from states lists.
2238 * @rt can be used to specify the range of states for deletion.
2241 dyn_expire_states(struct ip_fw_chain *ch, ipfw_range_tlv *rt)
2243 struct dyn_ipv4_slist expired_ipv4;
2245 struct dyn_ipv6_slist expired_ipv6;
2246 struct dyn_ipv6_state *s6, *s6n, *s6p;
2248 struct dyn_ipv4_state *s4, *s4n, *s4p;
2250 int bucket, removed, length, max_length;
2252 IPFW_UH_WLOCK_ASSERT(ch);
2255 * Unlink expired states from each bucket.
2256 * With acquired bucket lock iterate entries of each lists:
2257 * ipv4, ipv4_parent, ipv6, and ipv6_parent. Check expired time
2258 * and unlink entry from the list, link entry into temporary
2259 * expired_xxx lists then bump "del" bucket version.
2261 * When an entry is removed, corresponding states counter is
2262 * decremented. If entry has O_LIMIT type, parent's reference
2263 * counter is decremented.
2265 * NOTE: this function can be called from userspace context
2266 * when user deletes rules. In this case all matched states
2267 * will be forcedly unlinked. O_LIMIT_PARENT states will be kept
2268 * in the expired lists until reference counter become zero.
2270 #define DYN_UNLINK_STATES(s, prev, next, exp, af, name, extra) do { \
2274 s = CK_SLIST_FIRST(&V_dyn_ ## name [bucket]); \
2275 while (s != NULL) { \
2276 next = CK_SLIST_NEXT(s, entry); \
2277 if ((TIME_LEQ((s)->exp, time_uptime) && extra) || \
2279 dyn_match_ ## af ## _state(ch, s, rt))) { \
2281 CK_SLIST_REMOVE_AFTER(prev, entry); \
2283 CK_SLIST_REMOVE_HEAD( \
2284 &V_dyn_ ## name [bucket], entry); \
2286 SLIST_INSERT_HEAD(&expired_ ## af, s, expired); \
2287 if (s->type == O_LIMIT_PARENT) \
2288 DYN_COUNT_DEC(dyn_parent_count); \
2290 DYN_COUNT_DEC(dyn_count); \
2291 if (s->data->flags & DYN_REFERENCED) { \
2292 rule = s->data->parent; \
2293 if (s->type == O_LIMIT) \
2294 rule = ((__typeof(s)) \
2295 rule)->limit->parent;\
2296 dyn_release_rule(ch, s->data, \
2299 if (s->type == O_LIMIT) { \
2300 s = s->data->parent; \
2301 DPARENT_COUNT_DEC(s->limit); \
2311 DYN_BUCKET_VERSION_BUMP(bucket, name ## _del); \
2312 if (length > max_length) \
2313 max_length = length; \
2316 SLIST_INIT(&expired_ipv4);
2318 SLIST_INIT(&expired_ipv6);
2321 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) {
2322 DYN_BUCKET_LOCK(bucket);
2323 DYN_UNLINK_STATES(s4, s4p, s4n, data->expire, ipv4, ipv4, 1);
2324 DYN_UNLINK_STATES(s4, s4p, s4n, limit->expire, ipv4,
2325 ipv4_parent, (s4->limit->count == 0));
2327 DYN_UNLINK_STATES(s6, s6p, s6n, data->expire, ipv6, ipv6, 1);
2328 DYN_UNLINK_STATES(s6, s6p, s6n, limit->expire, ipv6,
2329 ipv6_parent, (s6->limit->count == 0));
2331 DYN_BUCKET_UNLOCK(bucket);
2333 /* Update curr_max_length for statistics. */
2334 V_curr_max_length = max_length;
2336 * Concatenate temporary lists with global expired lists.
2339 SLIST_CONCAT(&V_dyn_expired_ipv4, &expired_ipv4,
2340 dyn_ipv4_state, expired);
2342 SLIST_CONCAT(&V_dyn_expired_ipv6, &expired_ipv6,
2343 dyn_ipv6_state, expired);
2345 DYN_EXPIRED_UNLOCK();
2346 #undef DYN_UNLINK_STATES
2347 #undef DYN_UNREF_STATES
2350 static struct mbuf *
2351 dyn_mgethdr(int len, uint16_t fibnum)
2355 m = m_gethdr(M_NOWAIT, MT_DATA);
2359 mac_netinet_firewall_send(m);
2361 M_SETFIB(m, fibnum);
2362 m->m_data += max_linkhdr;
2363 m->m_flags |= M_SKIP_FIREWALL;
2364 m->m_len = m->m_pkthdr.len = len;
2365 bzero(m->m_data, len);
2370 dyn_make_keepalive_ipv4(struct mbuf *m, in_addr_t src, in_addr_t dst,
2371 uint32_t seq, uint32_t ack, uint16_t sport, uint16_t dport)
2376 ip = mtod(m, struct ip *);
2378 ip->ip_hl = sizeof(*ip) >> 2;
2379 ip->ip_tos = IPTOS_LOWDELAY;
2380 ip->ip_len = htons(m->m_len);
2381 ip->ip_off |= htons(IP_DF);
2382 ip->ip_ttl = V_ip_defttl;
2383 ip->ip_p = IPPROTO_TCP;
2384 ip->ip_src.s_addr = htonl(src);
2385 ip->ip_dst.s_addr = htonl(dst);
2387 tcp = mtodo(m, sizeof(struct ip));
2388 tcp->th_sport = htons(sport);
2389 tcp->th_dport = htons(dport);
2390 tcp->th_off = sizeof(struct tcphdr) >> 2;
2391 tcp->th_seq = htonl(seq);
2392 tcp->th_ack = htonl(ack);
2393 tcp->th_flags = TH_ACK;
2394 tcp->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2395 htons(sizeof(struct tcphdr) + IPPROTO_TCP));
2397 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2398 m->m_pkthdr.csum_flags = CSUM_TCP;
2402 dyn_enqueue_keepalive_ipv4(struct mbufq *q, const struct dyn_ipv4_state *s)
2406 if ((s->data->state & ACK_FWD) == 0 && s->data->ack_fwd > 0) {
2407 m = dyn_mgethdr(sizeof(struct ip) + sizeof(struct tcphdr),
2410 dyn_make_keepalive_ipv4(m, s->dst, s->src,
2411 s->data->ack_fwd - 1, s->data->ack_rev,
2412 s->dport, s->sport);
2413 if (mbufq_enqueue(q, m)) {
2415 log(LOG_DEBUG, "ipfw: limit for IPv4 "
2416 "keepalive queue is reached.\n");
2422 if ((s->data->state & ACK_REV) == 0 && s->data->ack_rev > 0) {
2423 m = dyn_mgethdr(sizeof(struct ip) + sizeof(struct tcphdr),
2426 dyn_make_keepalive_ipv4(m, s->src, s->dst,
2427 s->data->ack_rev - 1, s->data->ack_fwd,
2428 s->sport, s->dport);
2429 if (mbufq_enqueue(q, m)) {
2431 log(LOG_DEBUG, "ipfw: limit for IPv4 "
2432 "keepalive queue is reached.\n");
2440 * Prepare and send keep-alive packets.
2443 dyn_send_keepalive_ipv4(struct ip_fw_chain *chain)
2447 struct dyn_ipv4_state *s;
2450 mbufq_init(&q, INT_MAX);
2451 IPFW_UH_RLOCK(chain);
2453 * It is safe to not use hazard pointer and just do lockless
2454 * access to the lists, because states entries can not be deleted
2455 * while we hold IPFW_UH_RLOCK.
2457 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) {
2458 CK_SLIST_FOREACH(s, &V_dyn_ipv4[bucket], entry) {
2460 * Only established TCP connections that will
2461 * become expired withing dyn_keepalive_interval.
2463 if (s->proto != IPPROTO_TCP ||
2464 (s->data->state & BOTH_SYN) != BOTH_SYN ||
2465 TIME_LEQ(time_uptime + V_dyn_keepalive_interval,
2468 dyn_enqueue_keepalive_ipv4(&q, s);
2471 IPFW_UH_RUNLOCK(chain);
2472 while ((m = mbufq_dequeue(&q)) != NULL)
2473 ip_output(m, NULL, NULL, 0, NULL, NULL);
2478 dyn_make_keepalive_ipv6(struct mbuf *m, const struct in6_addr *src,
2479 const struct in6_addr *dst, uint32_t zoneid, uint32_t seq, uint32_t ack,
2480 uint16_t sport, uint16_t dport)
2483 struct ip6_hdr *ip6;
2485 ip6 = mtod(m, struct ip6_hdr *);
2486 ip6->ip6_vfc |= IPV6_VERSION;
2487 ip6->ip6_plen = htons(sizeof(struct tcphdr));
2488 ip6->ip6_nxt = IPPROTO_TCP;
2489 ip6->ip6_hlim = IPV6_DEFHLIM;
2490 ip6->ip6_src = *src;
2491 if (IN6_IS_ADDR_LINKLOCAL(src))
2492 ip6->ip6_src.s6_addr16[1] = htons(zoneid & 0xffff);
2493 ip6->ip6_dst = *dst;
2494 if (IN6_IS_ADDR_LINKLOCAL(dst))
2495 ip6->ip6_dst.s6_addr16[1] = htons(zoneid & 0xffff);
2497 tcp = mtodo(m, sizeof(struct ip6_hdr));
2498 tcp->th_sport = htons(sport);
2499 tcp->th_dport = htons(dport);
2500 tcp->th_off = sizeof(struct tcphdr) >> 2;
2501 tcp->th_seq = htonl(seq);
2502 tcp->th_ack = htonl(ack);
2503 tcp->th_flags = TH_ACK;
2504 tcp->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr),
2507 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2508 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2512 dyn_enqueue_keepalive_ipv6(struct mbufq *q, const struct dyn_ipv6_state *s)
2516 if ((s->data->state & ACK_FWD) == 0 && s->data->ack_fwd > 0) {
2517 m = dyn_mgethdr(sizeof(struct ip6_hdr) +
2518 sizeof(struct tcphdr), s->data->fibnum);
2520 dyn_make_keepalive_ipv6(m, &s->dst, &s->src,
2521 s->zoneid, s->data->ack_fwd - 1, s->data->ack_rev,
2522 s->dport, s->sport);
2523 if (mbufq_enqueue(q, m)) {
2525 log(LOG_DEBUG, "ipfw: limit for IPv6 "
2526 "keepalive queue is reached.\n");
2532 if ((s->data->state & ACK_REV) == 0 && s->data->ack_rev > 0) {
2533 m = dyn_mgethdr(sizeof(struct ip6_hdr) +
2534 sizeof(struct tcphdr), s->data->fibnum);
2536 dyn_make_keepalive_ipv6(m, &s->src, &s->dst,
2537 s->zoneid, s->data->ack_rev - 1, s->data->ack_fwd,
2538 s->sport, s->dport);
2539 if (mbufq_enqueue(q, m)) {
2541 log(LOG_DEBUG, "ipfw: limit for IPv6 "
2542 "keepalive queue is reached.\n");
2550 dyn_send_keepalive_ipv6(struct ip_fw_chain *chain)
2554 struct dyn_ipv6_state *s;
2557 mbufq_init(&q, INT_MAX);
2558 IPFW_UH_RLOCK(chain);
2560 * It is safe to not use hazard pointer and just do lockless
2561 * access to the lists, because states entries can not be deleted
2562 * while we hold IPFW_UH_RLOCK.
2564 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) {
2565 CK_SLIST_FOREACH(s, &V_dyn_ipv6[bucket], entry) {
2567 * Only established TCP connections that will
2568 * become expired withing dyn_keepalive_interval.
2570 if (s->proto != IPPROTO_TCP ||
2571 (s->data->state & BOTH_SYN) != BOTH_SYN ||
2572 TIME_LEQ(time_uptime + V_dyn_keepalive_interval,
2575 dyn_enqueue_keepalive_ipv6(&q, s);
2578 IPFW_UH_RUNLOCK(chain);
2579 while ((m = mbufq_dequeue(&q)) != NULL)
2580 ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
2585 dyn_grow_hashtable(struct ip_fw_chain *chain, uint32_t new)
2588 struct dyn_ipv6ck_slist *ipv6, *ipv6_parent;
2589 uint32_t *ipv6_add, *ipv6_del, *ipv6_parent_add, *ipv6_parent_del;
2590 struct dyn_ipv6_state *s6;
2592 struct dyn_ipv4ck_slist *ipv4, *ipv4_parent;
2593 uint32_t *ipv4_add, *ipv4_del, *ipv4_parent_add, *ipv4_parent_del;
2594 struct dyn_ipv4_state *s4;
2595 struct mtx *bucket_lock;
2599 MPASS(powerof2(new));
2600 DYN_DEBUG("grow hash size %u -> %u", V_curr_dyn_buckets, new);
2602 * Allocate and initialize new lists.
2603 * XXXAE: on memory pressure this can disable callout timer.
2605 bucket_lock = malloc(new * sizeof(struct mtx), M_IPFW,
2607 ipv4 = malloc(new * sizeof(struct dyn_ipv4ck_slist), M_IPFW,
2609 ipv4_parent = malloc(new * sizeof(struct dyn_ipv4ck_slist), M_IPFW,
2611 ipv4_add = malloc(new * sizeof(uint32_t), M_IPFW, M_WAITOK | M_ZERO);
2612 ipv4_del = malloc(new * sizeof(uint32_t), M_IPFW, M_WAITOK | M_ZERO);
2613 ipv4_parent_add = malloc(new * sizeof(uint32_t), M_IPFW,
2615 ipv4_parent_del = malloc(new * sizeof(uint32_t), M_IPFW,
2618 ipv6 = malloc(new * sizeof(struct dyn_ipv6ck_slist), M_IPFW,
2620 ipv6_parent = malloc(new * sizeof(struct dyn_ipv6ck_slist), M_IPFW,
2622 ipv6_add = malloc(new * sizeof(uint32_t), M_IPFW, M_WAITOK | M_ZERO);
2623 ipv6_del = malloc(new * sizeof(uint32_t), M_IPFW, M_WAITOK | M_ZERO);
2624 ipv6_parent_add = malloc(new * sizeof(uint32_t), M_IPFW,
2626 ipv6_parent_del = malloc(new * sizeof(uint32_t), M_IPFW,
2629 for (bucket = 0; bucket < new; bucket++) {
2630 DYN_BUCKET_LOCK_INIT(bucket_lock, bucket);
2631 CK_SLIST_INIT(&ipv4[bucket]);
2632 CK_SLIST_INIT(&ipv4_parent[bucket]);
2634 CK_SLIST_INIT(&ipv6[bucket]);
2635 CK_SLIST_INIT(&ipv6_parent[bucket]);
2639 #define DYN_RELINK_STATES(s, hval, i, head, ohead) do { \
2640 while ((s = CK_SLIST_FIRST(&V_dyn_ ## ohead[i])) != NULL) { \
2641 CK_SLIST_REMOVE_HEAD(&V_dyn_ ## ohead[i], entry); \
2642 CK_SLIST_INSERT_HEAD(&head[DYN_BUCKET(s->hval, new)], \
2647 * Prevent rules changing from userland.
2649 IPFW_UH_WLOCK(chain);
2651 * Hold traffic processing until we finish resize to
2652 * prevent access to states lists.
2655 /* Re-link all dynamic states */
2656 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) {
2657 DYN_RELINK_STATES(s4, data->hashval, bucket, ipv4, ipv4);
2658 DYN_RELINK_STATES(s4, limit->hashval, bucket, ipv4_parent,
2661 DYN_RELINK_STATES(s6, data->hashval, bucket, ipv6, ipv6);
2662 DYN_RELINK_STATES(s6, limit->hashval, bucket, ipv6_parent,
2667 #define DYN_SWAP_PTR(old, new, tmp) do { \
2673 DYN_SWAP_PTR(V_dyn_bucket_lock, bucket_lock, tmp);
2674 DYN_SWAP_PTR(V_dyn_ipv4, ipv4, tmp);
2675 DYN_SWAP_PTR(V_dyn_ipv4_parent, ipv4_parent, tmp);
2676 DYN_SWAP_PTR(V_dyn_ipv4_add, ipv4_add, tmp);
2677 DYN_SWAP_PTR(V_dyn_ipv4_parent_add, ipv4_parent_add, tmp);
2678 DYN_SWAP_PTR(V_dyn_ipv4_del, ipv4_del, tmp);
2679 DYN_SWAP_PTR(V_dyn_ipv4_parent_del, ipv4_parent_del, tmp);
2682 DYN_SWAP_PTR(V_dyn_ipv6, ipv6, tmp);
2683 DYN_SWAP_PTR(V_dyn_ipv6_parent, ipv6_parent, tmp);
2684 DYN_SWAP_PTR(V_dyn_ipv6_add, ipv6_add, tmp);
2685 DYN_SWAP_PTR(V_dyn_ipv6_parent_add, ipv6_parent_add, tmp);
2686 DYN_SWAP_PTR(V_dyn_ipv6_del, ipv6_del, tmp);
2687 DYN_SWAP_PTR(V_dyn_ipv6_parent_del, ipv6_parent_del, tmp);
2689 bucket = V_curr_dyn_buckets;
2690 V_curr_dyn_buckets = new;
2692 IPFW_WUNLOCK(chain);
2693 IPFW_UH_WUNLOCK(chain);
2695 /* Release old resources */
2696 while (bucket-- != 0)
2697 DYN_BUCKET_LOCK_DESTROY(bucket_lock, bucket);
2698 free(bucket_lock, M_IPFW);
2700 free(ipv4_parent, M_IPFW);
2701 free(ipv4_add, M_IPFW);
2702 free(ipv4_parent_add, M_IPFW);
2703 free(ipv4_del, M_IPFW);
2704 free(ipv4_parent_del, M_IPFW);
2707 free(ipv6_parent, M_IPFW);
2708 free(ipv6_add, M_IPFW);
2709 free(ipv6_parent_add, M_IPFW);
2710 free(ipv6_del, M_IPFW);
2711 free(ipv6_parent_del, M_IPFW);
2716 * This function is used to perform various maintenance
2717 * on dynamic hash lists. Currently it is called every second.
2720 dyn_tick(void *vnetx)
2722 struct epoch_tracker et;
2725 CURVNET_SET((struct vnet *)vnetx);
2727 * First free states unlinked in previous passes.
2729 dyn_free_states(&V_layer3_chain);
2731 * Now unlink others expired states.
2732 * We use IPFW_UH_WLOCK to avoid concurrent call of
2733 * dyn_expire_states(). It is the only function that does
2734 * deletion of state entries from states lists.
2736 IPFW_UH_WLOCK(&V_layer3_chain);
2737 dyn_expire_states(&V_layer3_chain, NULL);
2738 IPFW_UH_WUNLOCK(&V_layer3_chain);
2740 * Send keepalives if they are enabled and the time has come.
2742 if (V_dyn_keepalive != 0 &&
2743 V_dyn_keepalive_last + V_dyn_keepalive_period <= time_uptime) {
2744 V_dyn_keepalive_last = time_uptime;
2745 NET_EPOCH_ENTER(et);
2746 dyn_send_keepalive_ipv4(&V_layer3_chain);
2748 dyn_send_keepalive_ipv6(&V_layer3_chain);
2753 * Check if we need to resize the hash:
2754 * if current number of states exceeds number of buckets in hash,
2755 * and dyn_buckets_max permits to grow the number of buckets, then
2756 * do it. Grow hash size to the minimum power of 2 which is bigger
2757 * than current states count.
2759 if (V_curr_dyn_buckets < V_dyn_buckets_max &&
2760 (V_curr_dyn_buckets < V_dyn_count / 2 || (
2761 V_curr_dyn_buckets < V_dyn_count && V_curr_max_length > 8))) {
2762 buckets = 1 << fls(V_dyn_count);
2763 if (buckets > V_dyn_buckets_max)
2764 buckets = V_dyn_buckets_max;
2765 dyn_grow_hashtable(&V_layer3_chain, buckets);
2768 callout_reset_on(&V_dyn_timeout, hz, dyn_tick, vnetx, 0);
2773 ipfw_expire_dyn_states(struct ip_fw_chain *chain, ipfw_range_tlv *rt)
2776 * Do not perform any checks if we currently have no dynamic states
2778 if (V_dyn_count == 0)
2781 IPFW_UH_WLOCK_ASSERT(chain);
2782 dyn_expire_states(chain, rt);
2786 * Pass through all states and reset eaction for orphaned rules.
2789 ipfw_dyn_reset_eaction(struct ip_fw_chain *ch, uint16_t eaction_id,
2790 uint16_t default_id, uint16_t instance_id)
2793 struct dyn_ipv6_state *s6;
2795 struct dyn_ipv4_state *s4;
2799 #define DYN_RESET_EACTION(s, h, b) \
2800 CK_SLIST_FOREACH(s, &V_dyn_ ## h[b], entry) { \
2801 if ((s->data->flags & DYN_REFERENCED) == 0) \
2803 rule = s->data->parent; \
2804 if (s->type == O_LIMIT) \
2805 rule = ((__typeof(s))rule)->limit->parent; \
2806 ipfw_reset_eaction(ch, rule, eaction_id, \
2807 default_id, instance_id); \
2810 IPFW_UH_WLOCK_ASSERT(ch);
2811 if (V_dyn_count == 0)
2813 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) {
2814 DYN_RESET_EACTION(s4, ipv4, bucket);
2816 DYN_RESET_EACTION(s6, ipv6, bucket);
2822 * Returns size of dynamic states in legacy format
2828 return ((V_dyn_count + V_dyn_parent_count) * sizeof(ipfw_dyn_rule));
2832 * Returns number of dynamic states.
2833 * Marks every named object index used by dynamic states with bit in @bmask.
2834 * Returns number of named objects accounted in bmask via @nocnt.
2835 * Used by dump format v1 (current).
2838 ipfw_dyn_get_count(uint32_t *bmask, int *nocnt)
2841 struct dyn_ipv6_state *s6;
2843 struct dyn_ipv4_state *s4;
2846 #define DYN_COUNT_OBJECTS(s, h, b) \
2847 CK_SLIST_FOREACH(s, &V_dyn_ ## h[b], entry) { \
2848 MPASS(s->kidx != 0); \
2849 if (ipfw_mark_object_kidx(bmask, IPFW_TLV_STATE_NAME, \
2854 IPFW_UH_RLOCK_ASSERT(&V_layer3_chain);
2856 /* No need to pass through all the buckets. */
2858 if (V_dyn_count + V_dyn_parent_count == 0)
2861 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) {
2862 DYN_COUNT_OBJECTS(s4, ipv4, bucket);
2864 DYN_COUNT_OBJECTS(s6, ipv6, bucket);
2868 return (V_dyn_count + V_dyn_parent_count);
2872 * Check if rule contains at least one dynamic opcode.
2874 * Returns 1 if such opcode is found, 0 otherwise.
2877 ipfw_is_dyn_rule(struct ip_fw *rule)
2885 for ( ; l > 0 ; l -= cmdlen, cmd += cmdlen) {
2886 cmdlen = F_LEN(cmd);
2888 switch (cmd->opcode) {
2901 dyn_export_parent(const struct dyn_parent *p, uint16_t kidx, uint8_t set,
2905 dst->dyn_type = O_LIMIT_PARENT;
2907 dst->count = (uint16_t)DPARENT_COUNT(p);
2908 dst->expire = TIME_LEQ(p->expire, time_uptime) ? 0:
2909 p->expire - time_uptime;
2911 /* 'rule' is used to pass up the rule number and set */
2912 memcpy(&dst->rule, &p->rulenum, sizeof(p->rulenum));
2914 /* store set number into high word of dst->rule pointer. */
2915 memcpy((char *)&dst->rule + sizeof(p->rulenum), &set, sizeof(set));
2924 dst->bucket = p->hashval;
2926 * The legacy userland code will interpret a NULL here as a marker
2927 * for the last dynamic rule.
2929 dst->next = (ipfw_dyn_rule *)1;
2933 dyn_export_data(const struct dyn_data *data, uint16_t kidx, uint8_t type,
2934 uint8_t set, ipfw_dyn_rule *dst)
2937 dst->dyn_type = type;
2939 dst->pcnt = data->pcnt_fwd + data->pcnt_rev;
2940 dst->bcnt = data->bcnt_fwd + data->bcnt_rev;
2941 dst->expire = TIME_LEQ(data->expire, time_uptime) ? 0:
2942 data->expire - time_uptime;
2944 /* 'rule' is used to pass up the rule number and set */
2945 memcpy(&dst->rule, &data->rulenum, sizeof(data->rulenum));
2947 /* store set number into high word of dst->rule pointer. */
2948 memcpy((char *)&dst->rule + sizeof(data->rulenum), &set, sizeof(set));
2950 dst->state = data->state;
2951 if (data->flags & DYN_REFERENCED)
2952 dst->state |= IPFW_DYN_ORPHANED;
2956 dst->ack_fwd = data->ack_fwd;
2957 dst->ack_rev = data->ack_rev;
2959 dst->bucket = data->hashval;
2961 * The legacy userland code will interpret a NULL here as a marker
2962 * for the last dynamic rule.
2964 dst->next = (ipfw_dyn_rule *)1;
2968 dyn_export_ipv4_state(const struct dyn_ipv4_state *s, ipfw_dyn_rule *dst)
2973 case O_LIMIT_PARENT:
2974 rule = s->limit->parent;
2975 dyn_export_parent(s->limit, s->kidx, rule->set, dst);
2978 rule = s->data->parent;
2979 if (s->type == O_LIMIT)
2980 rule = ((struct dyn_ipv4_state *)rule)->limit->parent;
2981 dyn_export_data(s->data, s->kidx, s->type, rule->set, dst);
2984 dst->id.dst_ip = s->dst;
2985 dst->id.src_ip = s->src;
2986 dst->id.dst_port = s->dport;
2987 dst->id.src_port = s->sport;
2988 dst->id.fib = s->data->fibnum;
2989 dst->id.proto = s->proto;
2991 dst->id.addr_type = 4;
2993 memset(&dst->id.dst_ip6, 0, sizeof(dst->id.dst_ip6));
2994 memset(&dst->id.src_ip6, 0, sizeof(dst->id.src_ip6));
2995 dst->id.flow_id6 = dst->id.extra = 0;
3000 dyn_export_ipv6_state(const struct dyn_ipv6_state *s, ipfw_dyn_rule *dst)
3005 case O_LIMIT_PARENT:
3006 rule = s->limit->parent;
3007 dyn_export_parent(s->limit, s->kidx, rule->set, dst);
3010 rule = s->data->parent;
3011 if (s->type == O_LIMIT)
3012 rule = ((struct dyn_ipv6_state *)rule)->limit->parent;
3013 dyn_export_data(s->data, s->kidx, s->type, rule->set, dst);
3016 dst->id.src_ip6 = s->src;
3017 dst->id.dst_ip6 = s->dst;
3018 dst->id.dst_port = s->dport;
3019 dst->id.src_port = s->sport;
3020 dst->id.fib = s->data->fibnum;
3021 dst->id.proto = s->proto;
3023 dst->id.addr_type = 6;
3025 dst->id.dst_ip = dst->id.src_ip = 0;
3026 dst->id.flow_id6 = dst->id.extra = 0;
3031 * Fills the buffer given by @sd with dynamic states.
3032 * Used by dump format v1 (current).
3034 * Returns 0 on success.
3037 ipfw_dump_states(struct ip_fw_chain *chain, struct sockopt_data *sd)
3040 struct dyn_ipv6_state *s6;
3042 struct dyn_ipv4_state *s4;
3043 ipfw_obj_dyntlv *dst, *last;
3044 ipfw_obj_ctlv *ctlv;
3047 if (V_dyn_count == 0)
3051 * IPFW_UH_RLOCK garantees that another userland request
3052 * and callout thread will not delete entries from states
3055 IPFW_UH_RLOCK_ASSERT(chain);
3057 ctlv = (ipfw_obj_ctlv *)ipfw_get_sopt_space(sd, sizeof(*ctlv));
3060 ctlv->head.type = IPFW_TLV_DYNSTATE_LIST;
3061 ctlv->objsize = sizeof(ipfw_obj_dyntlv);
3064 #define DYN_EXPORT_STATES(s, af, h, b) \
3065 CK_SLIST_FOREACH(s, &V_dyn_ ## h[b], entry) { \
3066 dst = (ipfw_obj_dyntlv *)ipfw_get_sopt_space(sd, \
3067 sizeof(ipfw_obj_dyntlv)); \
3070 dyn_export_ ## af ## _state(s, &dst->state); \
3071 dst->head.length = sizeof(ipfw_obj_dyntlv); \
3072 dst->head.type = IPFW_TLV_DYN_ENT; \
3076 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) {
3077 DYN_EXPORT_STATES(s4, ipv4, ipv4_parent, bucket);
3078 DYN_EXPORT_STATES(s4, ipv4, ipv4, bucket);
3080 DYN_EXPORT_STATES(s6, ipv6, ipv6_parent, bucket);
3081 DYN_EXPORT_STATES(s6, ipv6, ipv6, bucket);
3085 /* mark last dynamic rule */
3087 last->head.flags = IPFW_DF_LAST; /* XXX: unused */
3089 #undef DYN_EXPORT_STATES
3093 * Fill given buffer with dynamic states (legacy format).
3094 * IPFW_UH_RLOCK has to be held while calling.
3097 ipfw_get_dynamic(struct ip_fw_chain *chain, char **pbp, const char *ep)
3100 struct dyn_ipv6_state *s6;
3102 struct dyn_ipv4_state *s4;
3103 ipfw_dyn_rule *p, *last = NULL;
3107 if (V_dyn_count == 0)
3111 IPFW_UH_RLOCK_ASSERT(chain);
3113 #define DYN_EXPORT_STATES(s, af, head, b) \
3114 CK_SLIST_FOREACH(s, &V_dyn_ ## head[b], entry) { \
3115 if (bp + sizeof(*p) > ep) \
3117 p = (ipfw_dyn_rule *)bp; \
3118 dyn_export_ ## af ## _state(s, p); \
3123 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) {
3124 DYN_EXPORT_STATES(s4, ipv4, ipv4_parent, bucket);
3125 DYN_EXPORT_STATES(s4, ipv4, ipv4, bucket);
3127 DYN_EXPORT_STATES(s6, ipv6, ipv6_parent, bucket);
3128 DYN_EXPORT_STATES(s6, ipv6, ipv6, bucket);
3132 if (last != NULL) /* mark last dynamic rule */
3135 #undef DYN_EXPORT_STATES
3139 ipfw_dyn_init(struct ip_fw_chain *chain)
3142 #ifdef IPFIREWALL_JENKINSHASH
3143 V_dyn_hashseed = arc4random();
3145 V_dyn_max = 16384; /* max # of states */
3146 V_dyn_parent_max = 4096; /* max # of parent states */
3147 V_dyn_buckets_max = 8192; /* must be power of 2 */
3149 V_dyn_ack_lifetime = 300;
3150 V_dyn_syn_lifetime = 20;
3151 V_dyn_fin_lifetime = 1;
3152 V_dyn_rst_lifetime = 1;
3153 V_dyn_udp_lifetime = 10;
3154 V_dyn_short_lifetime = 5;
3156 V_dyn_keepalive_interval = 20;
3157 V_dyn_keepalive_period = 5;
3158 V_dyn_keepalive = 1; /* send keepalives */
3159 V_dyn_keepalive_last = time_uptime;
3161 V_dyn_data_zone = uma_zcreate("IPFW dynamic states data",
3162 sizeof(struct dyn_data), NULL, NULL, NULL, NULL,
3164 uma_zone_set_max(V_dyn_data_zone, V_dyn_max);
3166 V_dyn_parent_zone = uma_zcreate("IPFW parent dynamic states",
3167 sizeof(struct dyn_parent), NULL, NULL, NULL, NULL,
3169 uma_zone_set_max(V_dyn_parent_zone, V_dyn_parent_max);
3171 SLIST_INIT(&V_dyn_expired_ipv4);
3173 V_dyn_ipv4_parent = NULL;
3174 V_dyn_ipv4_zone = uma_zcreate("IPFW IPv4 dynamic states",
3175 sizeof(struct dyn_ipv4_state), NULL, NULL, NULL, NULL,
3179 SLIST_INIT(&V_dyn_expired_ipv6);
3181 V_dyn_ipv6_parent = NULL;
3182 V_dyn_ipv6_zone = uma_zcreate("IPFW IPv6 dynamic states",
3183 sizeof(struct dyn_ipv6_state), NULL, NULL, NULL, NULL,
3187 /* Initialize buckets. */
3188 V_curr_dyn_buckets = 0;
3189 V_dyn_bucket_lock = NULL;
3190 dyn_grow_hashtable(chain, 256);
3192 if (IS_DEFAULT_VNET(curvnet))
3193 dyn_hp_cache = malloc(mp_ncpus * sizeof(void *), M_IPFW,
3196 DYN_EXPIRED_LOCK_INIT();
3197 callout_init(&V_dyn_timeout, 1);
3198 callout_reset(&V_dyn_timeout, hz, dyn_tick, curvnet);
3199 IPFW_ADD_OBJ_REWRITER(IS_DEFAULT_VNET(curvnet), dyn_opcodes);
3203 ipfw_dyn_uninit(int pass)
3206 struct dyn_ipv6_state *s6;
3208 struct dyn_ipv4_state *s4;
3212 callout_drain(&V_dyn_timeout);
3215 IPFW_DEL_OBJ_REWRITER(IS_DEFAULT_VNET(curvnet), dyn_opcodes);
3216 DYN_EXPIRED_LOCK_DESTROY();
3218 #define DYN_FREE_STATES_FORCED(CK, s, af, name, en) do { \
3219 while ((s = CK ## SLIST_FIRST(&V_dyn_ ## name)) != NULL) { \
3220 CK ## SLIST_REMOVE_HEAD(&V_dyn_ ## name, en); \
3221 if (s->type == O_LIMIT_PARENT) \
3222 uma_zfree(V_dyn_parent_zone, s->limit); \
3224 uma_zfree(V_dyn_data_zone, s->data); \
3225 uma_zfree(V_dyn_ ## af ## _zone, s); \
3228 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) {
3229 DYN_BUCKET_LOCK_DESTROY(V_dyn_bucket_lock, bucket);
3231 DYN_FREE_STATES_FORCED(CK_, s4, ipv4, ipv4[bucket], entry);
3232 DYN_FREE_STATES_FORCED(CK_, s4, ipv4, ipv4_parent[bucket],
3235 DYN_FREE_STATES_FORCED(CK_, s6, ipv6, ipv6[bucket], entry);
3236 DYN_FREE_STATES_FORCED(CK_, s6, ipv6, ipv6_parent[bucket],
3240 DYN_FREE_STATES_FORCED(, s4, ipv4, expired_ipv4, expired);
3242 DYN_FREE_STATES_FORCED(, s6, ipv6, expired_ipv6, expired);
3244 #undef DYN_FREE_STATES_FORCED
3246 uma_zdestroy(V_dyn_ipv4_zone);
3247 uma_zdestroy(V_dyn_data_zone);
3248 uma_zdestroy(V_dyn_parent_zone);
3250 uma_zdestroy(V_dyn_ipv6_zone);
3251 free(V_dyn_ipv6, M_IPFW);
3252 free(V_dyn_ipv6_parent, M_IPFW);
3253 free(V_dyn_ipv6_add, M_IPFW);
3254 free(V_dyn_ipv6_parent_add, M_IPFW);
3255 free(V_dyn_ipv6_del, M_IPFW);
3256 free(V_dyn_ipv6_parent_del, M_IPFW);
3258 free(V_dyn_bucket_lock, M_IPFW);
3259 free(V_dyn_ipv4, M_IPFW);
3260 free(V_dyn_ipv4_parent, M_IPFW);
3261 free(V_dyn_ipv4_add, M_IPFW);
3262 free(V_dyn_ipv4_parent_add, M_IPFW);
3263 free(V_dyn_ipv4_del, M_IPFW);
3264 free(V_dyn_ipv4_parent_del, M_IPFW);
3265 if (IS_DEFAULT_VNET(curvnet))
3266 free(dyn_hp_cache, M_IPFW);