1 /*-
2 * SPDX-License-Identifier: (BSD-2-Clause AND ISC)
3 *
4 * Copyright (c) 2002 Michael Shalayeff
5 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
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.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
21 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
22 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
23 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
25 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
26 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
27 * THE POSSIBILITY OF SUCH DAMAGE.
28 */
29
30 /*-
31 * Copyright (c) 2009 David Gwynne <dlg@openbsd.org>
32 *
33 * Permission to use, copy, modify, and distribute this software for any
34 * purpose with or without fee is hereby granted, provided that the above
35 * copyright notice and this permission notice appear in all copies.
36 *
37 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
38 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
39 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
40 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
41 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
42 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
43 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
44 */
45
46 /*
47 * $OpenBSD: if_pfsync.c,v 1.110 2009/02/24 05:39:19 dlg Exp $
48 *
49 * Revisions picked from OpenBSD after revision 1.110 import:
50 * 1.119 - don't m_copydata() beyond the len of mbuf in pfsync_input()
51 * 1.118, 1.124, 1.148, 1.149, 1.151, 1.171 - fixes to bulk updates
52 * 1.120, 1.175 - use monotonic time_uptime
53 * 1.122 - reduce number of updates for non-TCP sessions
54 * 1.125, 1.127 - rewrite merge or stale processing
55 * 1.128 - cleanups
56 * 1.146 - bzero() mbuf before sparsely filling it with data
57 * 1.170 - SIOCSIFMTU checks
58 * 1.126, 1.142 - deferred packets processing
59 * 1.173 - correct expire time processing
60 */
61
62 #include <sys/cdefs.h>
63 #include "opt_inet.h"
64 #include "opt_inet6.h"
65 #include "opt_pf.h"
66
67 #include <sys/param.h>
68 #include <sys/bus.h>
69 #include <sys/endian.h>
70 #include <sys/interrupt.h>
71 #include <sys/kernel.h>
72 #include <sys/lock.h>
73 #include <sys/mbuf.h>
74 #include <sys/module.h>
75 #include <sys/mutex.h>
76 #include <sys/nv.h>
77 #include <sys/priv.h>
78 #include <sys/smp.h>
79 #include <sys/socket.h>
80 #include <sys/sockio.h>
81 #include <sys/sysctl.h>
82 #include <sys/syslog.h>
83
84 #include <net/bpf.h>
85 #include <net/if.h>
86 #include <net/if_var.h>
87 #include <net/if_clone.h>
88 #include <net/if_private.h>
89 #include <net/if_types.h>
90 #include <net/vnet.h>
91 #include <net/pfvar.h>
92 #include <net/route.h>
93 #include <net/if_pfsync.h>
94
95 #include <netinet/if_ether.h>
96 #include <netinet/in.h>
97 #include <netinet/in_var.h>
98 #include <netinet6/in6_var.h>
99 #include <netinet/ip.h>
100 #include <netinet/ip6.h>
101 #include <netinet/ip_carp.h>
102 #include <netinet/ip_var.h>
103 #include <netinet/tcp.h>
104 #include <netinet/tcp_fsm.h>
105 #include <netinet/tcp_seq.h>
106
107 #include <netinet/ip6.h>
108 #include <netinet6/ip6_var.h>
109 #include <netinet6/scope6_var.h>
110
111 #include <netpfil/pf/pfsync_nv.h>
112
113 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
114
115 struct pfsync_bucket;
116 struct pfsync_softc;
117
118 union inet_template {
119 struct ip ipv4;
120 struct ip6_hdr ipv6;
121 };
122
123 #define PFSYNC_MINPKT ( \
124 sizeof(union inet_template) + \
125 sizeof(struct pfsync_header) + \
126 sizeof(struct pfsync_subheader) )
127
128 static int pfsync_upd_tcp(struct pf_kstate *, struct pfsync_state_peer *,
129 struct pfsync_state_peer *);
130 static int pfsync_in_clr(struct mbuf *, int, int, int, int);
131 static int pfsync_in_ins(struct mbuf *, int, int, int, int);
132 static int pfsync_in_iack(struct mbuf *, int, int, int, int);
133 static int pfsync_in_upd(struct mbuf *, int, int, int, int);
134 static int pfsync_in_upd_c(struct mbuf *, int, int, int, int);
135 static int pfsync_in_ureq(struct mbuf *, int, int, int, int);
136 static int pfsync_in_del_c(struct mbuf *, int, int, int, int);
137 static int pfsync_in_bus(struct mbuf *, int, int, int, int);
138 static int pfsync_in_tdb(struct mbuf *, int, int, int, int);
139 static int pfsync_in_eof(struct mbuf *, int, int, int, int);
140 static int pfsync_in_error(struct mbuf *, int, int, int, int);
141
142 static int (*pfsync_acts[])(struct mbuf *, int, int, int, int) = {
143 pfsync_in_clr, /* PFSYNC_ACT_CLR */
144 pfsync_in_ins, /* PFSYNC_ACT_INS_1301 */
145 pfsync_in_iack, /* PFSYNC_ACT_INS_ACK */
146 pfsync_in_upd, /* PFSYNC_ACT_UPD_1301 */
147 pfsync_in_upd_c, /* PFSYNC_ACT_UPD_C */
148 pfsync_in_ureq, /* PFSYNC_ACT_UPD_REQ */
149 pfsync_in_error, /* PFSYNC_ACT_DEL */
150 pfsync_in_del_c, /* PFSYNC_ACT_DEL_C */
151 pfsync_in_error, /* PFSYNC_ACT_INS_F */
152 pfsync_in_error, /* PFSYNC_ACT_DEL_F */
153 pfsync_in_bus, /* PFSYNC_ACT_BUS */
154 pfsync_in_tdb, /* PFSYNC_ACT_TDB */
155 pfsync_in_eof, /* PFSYNC_ACT_EOF */
156 pfsync_in_ins, /* PFSYNC_ACT_INS_1400 */
157 pfsync_in_upd, /* PFSYNC_ACT_UPD_1400 */
158 };
159
160 struct pfsync_q {
161 void (*write)(struct pf_kstate *, void *);
162 size_t len;
163 u_int8_t action;
164 };
165
166 /* We have the following sync queues */
167 enum pfsync_q_id {
168 PFSYNC_Q_INS_1301,
169 PFSYNC_Q_INS_1400,
170 PFSYNC_Q_IACK,
171 PFSYNC_Q_UPD_1301,
172 PFSYNC_Q_UPD_1400,
173 PFSYNC_Q_UPD_C,
174 PFSYNC_Q_DEL_C,
175 PFSYNC_Q_COUNT,
176 };
177
178 /* Functions for building messages for given queue */
179 static void pfsync_out_state_1301(struct pf_kstate *, void *);
180 static void pfsync_out_state_1400(struct pf_kstate *, void *);
181 static void pfsync_out_iack(struct pf_kstate *, void *);
182 static void pfsync_out_upd_c(struct pf_kstate *, void *);
183 static void pfsync_out_del_c(struct pf_kstate *, void *);
184
185 /* Attach those functions to queue */
186 static struct pfsync_q pfsync_qs[] = {
187 { pfsync_out_state_1301, sizeof(struct pfsync_state_1301), PFSYNC_ACT_INS_1301 },
188 { pfsync_out_state_1400, sizeof(struct pfsync_state_1400), PFSYNC_ACT_INS_1400 },
189 { pfsync_out_iack, sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK },
190 { pfsync_out_state_1301, sizeof(struct pfsync_state_1301), PFSYNC_ACT_UPD_1301 },
191 { pfsync_out_state_1400, sizeof(struct pfsync_state_1400), PFSYNC_ACT_UPD_1400 },
192 { pfsync_out_upd_c, sizeof(struct pfsync_upd_c), PFSYNC_ACT_UPD_C },
193 { pfsync_out_del_c, sizeof(struct pfsync_del_c), PFSYNC_ACT_DEL_C }
194 };
195
196 /* Map queue to pf_kstate->sync_state */
197 static u_int8_t pfsync_qid_sstate[] = {
198 PFSYNC_S_INS, /* PFSYNC_Q_INS_1301 */
199 PFSYNC_S_INS, /* PFSYNC_Q_INS_1400 */
200 PFSYNC_S_IACK, /* PFSYNC_Q_IACK */
201 PFSYNC_S_UPD, /* PFSYNC_Q_UPD_1301 */
202 PFSYNC_S_UPD, /* PFSYNC_Q_UPD_1400 */
203 PFSYNC_S_UPD_C, /* PFSYNC_Q_UPD_C */
204 PFSYNC_S_DEL_C, /* PFSYNC_Q_DEL_C */
205 };
206
207 /* Map pf_kstate->sync_state to queue */
208 static enum pfsync_q_id pfsync_sstate_to_qid(u_int8_t);
209
210 static void pfsync_q_ins(struct pf_kstate *, int sync_state, bool);
211 static void pfsync_q_del(struct pf_kstate *, bool, struct pfsync_bucket *);
212
213 static void pfsync_update_state(struct pf_kstate *);
214 static void pfsync_tx(struct pfsync_softc *, struct mbuf *);
215
216 struct pfsync_upd_req_item {
217 TAILQ_ENTRY(pfsync_upd_req_item) ur_entry;
218 struct pfsync_upd_req ur_msg;
219 };
220
221 struct pfsync_deferral {
222 struct pfsync_softc *pd_sc;
223 TAILQ_ENTRY(pfsync_deferral) pd_entry;
224 struct callout pd_tmo;
225
226 struct pf_kstate *pd_st;
227 struct mbuf *pd_m;
228 };
229
230 struct pfsync_bucket
231 {
232 int b_id;
233 struct pfsync_softc *b_sc;
234 struct mtx b_mtx;
235 struct callout b_tmo;
236 int b_flags;
237 #define PFSYNCF_BUCKET_PUSH 0x00000001
238
239 size_t b_len;
240 TAILQ_HEAD(, pf_kstate) b_qs[PFSYNC_Q_COUNT];
241 TAILQ_HEAD(, pfsync_upd_req_item) b_upd_req_list;
242 TAILQ_HEAD(, pfsync_deferral) b_deferrals;
243 u_int b_deferred;
244 uint8_t *b_plus;
245 size_t b_pluslen;
246
247 struct ifaltq b_snd;
248 };
249
250 struct pfsync_softc {
251 /* Configuration */
252 struct ifnet *sc_ifp;
253 struct ifnet *sc_sync_if;
254 struct ip_moptions sc_imo;
255 struct ip6_moptions sc_im6o;
256 struct sockaddr_storage sc_sync_peer;
257 uint32_t sc_flags;
258 uint8_t sc_maxupdates;
259 union inet_template sc_template;
260 struct mtx sc_mtx;
261 uint32_t sc_version;
262
263 /* Queued data */
264 struct pfsync_bucket *sc_buckets;
265
266 /* Bulk update info */
267 struct mtx sc_bulk_mtx;
268 uint32_t sc_ureq_sent;
269 int sc_bulk_tries;
270 uint32_t sc_ureq_received;
271 int sc_bulk_hashid;
272 uint64_t sc_bulk_stateid;
273 uint32_t sc_bulk_creatorid;
274 struct callout sc_bulk_tmo;
275 struct callout sc_bulkfail_tmo;
276 };
277
278 #define PFSYNC_LOCK(sc) mtx_lock(&(sc)->sc_mtx)
279 #define PFSYNC_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx)
280 #define PFSYNC_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED)
281
282 #define PFSYNC_BUCKET_LOCK(b) mtx_lock(&(b)->b_mtx)
283 #define PFSYNC_BUCKET_UNLOCK(b) mtx_unlock(&(b)->b_mtx)
284 #define PFSYNC_BUCKET_LOCK_ASSERT(b) mtx_assert(&(b)->b_mtx, MA_OWNED)
285
286 #define PFSYNC_BLOCK(sc) mtx_lock(&(sc)->sc_bulk_mtx)
287 #define PFSYNC_BUNLOCK(sc) mtx_unlock(&(sc)->sc_bulk_mtx)
288 #define PFSYNC_BLOCK_ASSERT(sc) mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED)
289
290 #define PFSYNC_DEFER_TIMEOUT 20
291
292 static const char pfsyncname[] = "pfsync";
293 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data");
294 VNET_DEFINE_STATIC(struct pfsync_softc *, pfsyncif) = NULL;
295 #define V_pfsyncif VNET(pfsyncif)
296 VNET_DEFINE_STATIC(void *, pfsync_swi_cookie) = NULL;
297 #define V_pfsync_swi_cookie VNET(pfsync_swi_cookie)
298 VNET_DEFINE_STATIC(struct intr_event *, pfsync_swi_ie);
299 #define V_pfsync_swi_ie VNET(pfsync_swi_ie)
300 VNET_DEFINE_STATIC(struct pfsyncstats, pfsyncstats);
301 #define V_pfsyncstats VNET(pfsyncstats)
302 VNET_DEFINE_STATIC(int, pfsync_carp_adj) = CARP_MAXSKEW;
303 #define V_pfsync_carp_adj VNET(pfsync_carp_adj)
304 VNET_DEFINE_STATIC(unsigned int, pfsync_defer_timeout) = PFSYNC_DEFER_TIMEOUT;
305 #define V_pfsync_defer_timeout VNET(pfsync_defer_timeout)
306
307 static void pfsync_timeout(void *);
308 static void pfsync_push(struct pfsync_bucket *);
309 static void pfsync_push_all(struct pfsync_softc *);
310 static void pfsyncintr(void *);
311 static int pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *,
312 struct in_mfilter *, struct in6_mfilter *);
313 static void pfsync_multicast_cleanup(struct pfsync_softc *);
314 static void pfsync_pointers_init(void);
315 static void pfsync_pointers_uninit(void);
316 static int pfsync_init(void);
317 static void pfsync_uninit(void);
318
319 static unsigned long pfsync_buckets;
320
321 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
322 "PFSYNC");
323 SYSCTL_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_VNET | CTLFLAG_RW,
324 &VNET_NAME(pfsyncstats), pfsyncstats,
325 "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)");
326 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_VNET | CTLFLAG_RW,
327 &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment");
328 SYSCTL_ULONG(_net_pfsync, OID_AUTO, pfsync_buckets, CTLFLAG_RDTUN,
329 &pfsync_buckets, 0, "Number of pfsync hash buckets");
330 SYSCTL_UINT(_net_pfsync, OID_AUTO, defer_delay, CTLFLAG_VNET | CTLFLAG_RW,
331 &VNET_NAME(pfsync_defer_timeout), 0, "Deferred packet timeout (in ms)");
332
333 static int pfsync_clone_create(struct if_clone *, int, caddr_t);
334 static void pfsync_clone_destroy(struct ifnet *);
335 static int pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
336 struct pf_state_peer *);
337 static int pfsyncoutput(struct ifnet *, struct mbuf *,
338 const struct sockaddr *, struct route *);
339 static int pfsyncioctl(struct ifnet *, u_long, caddr_t);
340
341 static int pfsync_defer(struct pf_kstate *, struct mbuf *);
342 static void pfsync_undefer(struct pfsync_deferral *, int);
343 static void pfsync_undefer_state_locked(struct pf_kstate *, int);
344 static void pfsync_undefer_state(struct pf_kstate *, int);
345 static void pfsync_defer_tmo(void *);
346
347 static void pfsync_request_update(u_int32_t, u_int64_t);
348 static bool pfsync_update_state_req(struct pf_kstate *);
349
350 static void pfsync_drop_all(struct pfsync_softc *);
351 static void pfsync_drop(struct pfsync_softc *, int);
352 static void pfsync_sendout(int, int);
353 static void pfsync_send_plus(void *, size_t);
354
355 static void pfsync_bulk_start(void);
356 static void pfsync_bulk_status(u_int8_t);
357 static void pfsync_bulk_update(void *);
358 static void pfsync_bulk_fail(void *);
359
360 static void pfsync_detach_ifnet(struct ifnet *);
361
362 static int pfsync_pfsyncreq_to_kstatus(struct pfsyncreq *,
363 struct pfsync_kstatus *);
364 static int pfsync_kstatus_to_softc(struct pfsync_kstatus *,
365 struct pfsync_softc *);
366
367 #ifdef IPSEC
368 static void pfsync_update_net_tdb(struct pfsync_tdb *);
369 #endif
370 static struct pfsync_bucket *pfsync_get_bucket(struct pfsync_softc *,
371 struct pf_kstate *);
372
373 #define PFSYNC_MAX_BULKTRIES 12
374
375 VNET_DEFINE(struct if_clone *, pfsync_cloner);
376 #define V_pfsync_cloner VNET(pfsync_cloner)
377
378 const struct in6_addr in6addr_linklocal_pfsync_group =
379 {{{ 0xff, 0x12, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
380 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xf0 }}};
381 static int
pfsync_clone_create(struct if_clone * ifc,int unit,caddr_t param)382 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param)
383 {
384 struct pfsync_softc *sc;
385 struct ifnet *ifp;
386 struct pfsync_bucket *b;
387 int c;
388 enum pfsync_q_id q;
389
390 if (unit != 0)
391 return (EINVAL);
392
393 if (! pfsync_buckets)
394 pfsync_buckets = mp_ncpus * 2;
395
396 sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO);
397 sc->sc_flags |= PFSYNCF_OK;
398 sc->sc_maxupdates = 128;
399 sc->sc_version = PFSYNC_MSG_VERSION_DEFAULT;
400 sc->sc_buckets = mallocarray(pfsync_buckets, sizeof(*sc->sc_buckets),
401 M_PFSYNC, M_ZERO | M_WAITOK);
402 for (c = 0; c < pfsync_buckets; c++) {
403 b = &sc->sc_buckets[c];
404 mtx_init(&b->b_mtx, "pfsync bucket", NULL, MTX_DEF);
405
406 b->b_id = c;
407 b->b_sc = sc;
408 b->b_len = PFSYNC_MINPKT;
409
410 for (q = 0; q < PFSYNC_Q_COUNT; q++)
411 TAILQ_INIT(&b->b_qs[q]);
412
413 TAILQ_INIT(&b->b_upd_req_list);
414 TAILQ_INIT(&b->b_deferrals);
415
416 callout_init(&b->b_tmo, 1);
417
418 b->b_snd.ifq_maxlen = ifqmaxlen;
419 }
420
421 ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC);
422 if_initname(ifp, pfsyncname, unit);
423 ifp->if_softc = sc;
424 ifp->if_ioctl = pfsyncioctl;
425 ifp->if_output = pfsyncoutput;
426 ifp->if_hdrlen = sizeof(struct pfsync_header);
427 ifp->if_mtu = ETHERMTU;
428 mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF);
429 mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF);
430 callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0);
431 callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0);
432
433 if_attach(ifp);
434
435 bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
436
437 V_pfsyncif = sc;
438
439 return (0);
440 }
441
442 static void
pfsync_clone_destroy(struct ifnet * ifp)443 pfsync_clone_destroy(struct ifnet *ifp)
444 {
445 struct pfsync_softc *sc = ifp->if_softc;
446 struct pfsync_bucket *b;
447 int c, ret;
448
449 for (c = 0; c < pfsync_buckets; c++) {
450 b = &sc->sc_buckets[c];
451 /*
452 * At this stage, everything should have already been
453 * cleared by pfsync_uninit(), and we have only to
454 * drain callouts.
455 */
456 PFSYNC_BUCKET_LOCK(b);
457 while (b->b_deferred > 0) {
458 struct pfsync_deferral *pd =
459 TAILQ_FIRST(&b->b_deferrals);
460
461 ret = callout_stop(&pd->pd_tmo);
462 PFSYNC_BUCKET_UNLOCK(b);
463 if (ret > 0) {
464 pfsync_undefer(pd, 1);
465 } else {
466 callout_drain(&pd->pd_tmo);
467 }
468 PFSYNC_BUCKET_LOCK(b);
469 }
470 MPASS(b->b_deferred == 0);
471 MPASS(TAILQ_EMPTY(&b->b_deferrals));
472 PFSYNC_BUCKET_UNLOCK(b);
473
474 free(b->b_plus, M_PFSYNC);
475 b->b_plus = NULL;
476 b->b_pluslen = 0;
477
478 callout_drain(&b->b_tmo);
479 }
480
481 callout_drain(&sc->sc_bulkfail_tmo);
482 callout_drain(&sc->sc_bulk_tmo);
483
484 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
485 (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
486 bpfdetach(ifp);
487 if_detach(ifp);
488
489 pfsync_drop_all(sc);
490
491 if_free(ifp);
492 pfsync_multicast_cleanup(sc);
493 mtx_destroy(&sc->sc_mtx);
494 mtx_destroy(&sc->sc_bulk_mtx);
495
496 for (c = 0; c < pfsync_buckets; c++) {
497 b = &sc->sc_buckets[c];
498 mtx_destroy(&b->b_mtx);
499 }
500 free(sc->sc_buckets, M_PFSYNC);
501 free(sc, M_PFSYNC);
502
503 V_pfsyncif = NULL;
504 }
505
506 static int
pfsync_alloc_scrub_memory(struct pfsync_state_peer * s,struct pf_state_peer * d)507 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
508 struct pf_state_peer *d)
509 {
510 if (s->scrub.scrub_flag && d->scrub == NULL) {
511 d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO);
512 if (d->scrub == NULL)
513 return (ENOMEM);
514 }
515
516 return (0);
517 }
518
519 static int
pfsync_state_import(union pfsync_state_union * sp,int flags,int msg_version)520 pfsync_state_import(union pfsync_state_union *sp, int flags, int msg_version)
521 {
522 struct pfsync_softc *sc = V_pfsyncif;
523 #ifndef __NO_STRICT_ALIGNMENT
524 struct pfsync_state_key key[2];
525 #endif
526 struct pfsync_state_key *kw, *ks;
527 struct pf_kstate *st = NULL;
528 struct pf_state_key *skw = NULL, *sks = NULL;
529 struct pf_krule *r = NULL;
530 struct pfi_kkif *kif;
531 struct pfi_kkif *rt_kif = NULL;
532 struct pf_kpooladdr *rpool_first;
533 int error;
534 uint8_t rt = 0;
535
536 PF_RULES_RASSERT();
537
538 if (sp->pfs_1301.creatorid == 0) {
539 if (V_pf_status.debug >= PF_DEBUG_MISC)
540 printf("%s: invalid creator id: %08x\n", __func__,
541 ntohl(sp->pfs_1301.creatorid));
542 return (EINVAL);
543 }
544
545 if ((kif = pfi_kkif_find(sp->pfs_1301.ifname)) == NULL) {
546 if (V_pf_status.debug >= PF_DEBUG_MISC)
547 printf("%s: unknown interface: %s\n", __func__,
548 sp->pfs_1301.ifname);
549 if (flags & PFSYNC_SI_IOCTL)
550 return (EINVAL);
551 return (0); /* skip this state */
552 }
553
554 /*
555 * If the ruleset checksums match or the state is coming from the ioctl,
556 * it's safe to associate the state with the rule of that number.
557 */
558 if (sp->pfs_1301.rule != htonl(-1) && sp->pfs_1301.anchor == htonl(-1) &&
559 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->pfs_1301.rule) <
560 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
561 r = pf_main_ruleset.rules[
562 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->pfs_1301.rule)];
563 else
564 r = &V_pf_default_rule;
565
566 /*
567 * Check routing interface early on. Do it before allocating memory etc.
568 * because there is a high chance there will be a lot more such states.
569 */
570 switch (msg_version) {
571 case PFSYNC_MSG_VERSION_1301:
572 /*
573 * On FreeBSD <= 13 the routing interface and routing operation
574 * are not sent over pfsync. If the ruleset is identical,
575 * though, we might be able to recover the routing information
576 * from the local ruleset.
577 */
578 if (r != &V_pf_default_rule) {
579 struct pf_kpool *pool = &r->route;
580
581 /* Backwards compatibility. */
582 if (TAILQ_EMPTY(&pool->list))
583 pool = &r->rdr;
584
585 /*
586 * The ruleset is identical, try to recover. If the rule
587 * has a redirection pool with a single interface, there
588 * is a chance that this interface is identical as on
589 * the pfsync peer. If there's more than one interface,
590 * give up, as we can't be sure that we will pick the
591 * same one as the pfsync peer did.
592 */
593 rpool_first = TAILQ_FIRST(&(pool->list));
594 if ((rpool_first == NULL) ||
595 (TAILQ_NEXT(rpool_first, entries) != NULL)) {
596 DPFPRINTF(PF_DEBUG_MISC,
597 ("%s: can't recover routing information "
598 "because of empty or bad redirection pool\n",
599 __func__));
600 return ((flags & PFSYNC_SI_IOCTL) ? EINVAL : 0);
601 }
602 rt = r->rt;
603 rt_kif = rpool_first->kif;
604 } else if (!PF_AZERO(&sp->pfs_1301.rt_addr, sp->pfs_1301.af)) {
605 /*
606 * Ruleset different, routing *supposedly* requested,
607 * give up on recovering.
608 */
609 DPFPRINTF(PF_DEBUG_MISC,
610 ("%s: can't recover routing information "
611 "because of different ruleset\n", __func__));
612 return ((flags & PFSYNC_SI_IOCTL) ? EINVAL : 0);
613 }
614 break;
615 case PFSYNC_MSG_VERSION_1400:
616 /*
617 * On FreeBSD 14 and above we're not taking any chances.
618 * We use the information synced to us.
619 */
620 if (sp->pfs_1400.rt) {
621 rt_kif = pfi_kkif_find(sp->pfs_1400.rt_ifname);
622 if (rt_kif == NULL) {
623 DPFPRINTF(PF_DEBUG_MISC,
624 ("%s: unknown route interface: %s\n",
625 __func__, sp->pfs_1400.rt_ifname));
626 return ((flags & PFSYNC_SI_IOCTL) ? EINVAL : 0);
627 }
628 rt = sp->pfs_1400.rt;
629 }
630 break;
631 }
632
633 if ((r->max_states &&
634 counter_u64_fetch(r->states_cur) >= r->max_states))
635 goto cleanup;
636
637 /*
638 * XXXGL: consider M_WAITOK in ioctl path after.
639 */
640 st = pf_alloc_state(M_NOWAIT);
641 if (__predict_false(st == NULL))
642 goto cleanup;
643
644 if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
645 goto cleanup;
646
647 #ifndef __NO_STRICT_ALIGNMENT
648 bcopy(&sp->pfs_1301.key, key, sizeof(struct pfsync_state_key) * 2);
649 kw = &key[PF_SK_WIRE];
650 ks = &key[PF_SK_STACK];
651 #else
652 kw = &sp->pfs_1301.key[PF_SK_WIRE];
653 ks = &sp->pfs_1301.key[PF_SK_STACK];
654 #endif
655
656 if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->pfs_1301.af) ||
657 PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->pfs_1301.af) ||
658 kw->port[0] != ks->port[0] ||
659 kw->port[1] != ks->port[1]) {
660 sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
661 if (sks == NULL)
662 goto cleanup;
663 } else
664 sks = skw;
665
666 /* allocate memory for scrub info */
667 if (pfsync_alloc_scrub_memory(&sp->pfs_1301.src, &st->src) ||
668 pfsync_alloc_scrub_memory(&sp->pfs_1301.dst, &st->dst))
669 goto cleanup;
670
671 /* Copy to state key(s). */
672 skw->addr[0] = kw->addr[0];
673 skw->addr[1] = kw->addr[1];
674 skw->port[0] = kw->port[0];
675 skw->port[1] = kw->port[1];
676 skw->proto = sp->pfs_1301.proto;
677 skw->af = sp->pfs_1301.af;
678 if (sks != skw) {
679 sks->addr[0] = ks->addr[0];
680 sks->addr[1] = ks->addr[1];
681 sks->port[0] = ks->port[0];
682 sks->port[1] = ks->port[1];
683 sks->proto = sp->pfs_1301.proto;
684 sks->af = sp->pfs_1301.af;
685 }
686
687 /* copy to state */
688 bcopy(&sp->pfs_1301.rt_addr, &st->act.rt_addr, sizeof(st->act.rt_addr));
689 st->creation = (time_uptime - ntohl(sp->pfs_1301.creation)) * 1000;
690 st->expire = pf_get_uptime();
691 if (sp->pfs_1301.expire) {
692 uint32_t timeout;
693
694 timeout = r->timeout[sp->pfs_1301.timeout];
695 if (!timeout)
696 timeout = V_pf_default_rule.timeout[sp->pfs_1301.timeout];
697
698 /* sp->expire may have been adaptively scaled by export. */
699 st->expire -= (timeout - ntohl(sp->pfs_1301.expire)) * 1000;
700 }
701
702 st->direction = sp->pfs_1301.direction;
703 st->act.log = sp->pfs_1301.log;
704 st->timeout = sp->pfs_1301.timeout;
705
706 st->act.rt = rt;
707 st->act.rt_kif = rt_kif;
708
709 switch (msg_version) {
710 case PFSYNC_MSG_VERSION_1301:
711 st->state_flags = sp->pfs_1301.state_flags;
712 /*
713 * In FreeBSD 13 pfsync lacks many attributes. Copy them
714 * from the rule if possible. If rule can't be matched
715 * clear any set options as we can't recover their
716 * parameters.
717 */
718 if (r == &V_pf_default_rule) {
719 st->state_flags &= ~PFSTATE_SETMASK;
720 } else {
721 /*
722 * Similar to pf_rule_to_actions(). This code
723 * won't set the actions properly if they come
724 * from multiple "match" rules as only rule
725 * creating the state is send over pfsync.
726 */
727 st->act.qid = r->qid;
728 st->act.pqid = r->pqid;
729 st->act.rtableid = r->rtableid;
730 if (r->scrub_flags & PFSTATE_SETTOS)
731 st->act.set_tos = r->set_tos;
732 st->act.min_ttl = r->min_ttl;
733 st->act.max_mss = r->max_mss;
734 st->state_flags |= (r->scrub_flags &
735 (PFSTATE_NODF|PFSTATE_RANDOMID|
736 PFSTATE_SETTOS|PFSTATE_SCRUB_TCP|
737 PFSTATE_SETPRIO));
738 if (r->dnpipe || r->dnrpipe) {
739 if (r->free_flags & PFRULE_DN_IS_PIPE)
740 st->state_flags |= PFSTATE_DN_IS_PIPE;
741 else
742 st->state_flags &= ~PFSTATE_DN_IS_PIPE;
743 }
744 st->act.dnpipe = r->dnpipe;
745 st->act.dnrpipe = r->dnrpipe;
746 }
747 break;
748 case PFSYNC_MSG_VERSION_1400:
749 st->state_flags = ntohs(sp->pfs_1400.state_flags);
750 st->act.qid = ntohs(sp->pfs_1400.qid);
751 st->act.pqid = ntohs(sp->pfs_1400.pqid);
752 st->act.dnpipe = ntohs(sp->pfs_1400.dnpipe);
753 st->act.dnrpipe = ntohs(sp->pfs_1400.dnrpipe);
754 st->act.rtableid = ntohl(sp->pfs_1400.rtableid);
755 st->act.min_ttl = sp->pfs_1400.min_ttl;
756 st->act.set_tos = sp->pfs_1400.set_tos;
757 st->act.max_mss = ntohs(sp->pfs_1400.max_mss);
758 st->act.set_prio[0] = sp->pfs_1400.set_prio[0];
759 st->act.set_prio[1] = sp->pfs_1400.set_prio[1];
760 break;
761 default:
762 panic("%s: Unsupported pfsync_msg_version %d",
763 __func__, msg_version);
764 }
765
766 st->id = sp->pfs_1301.id;
767 st->creatorid = sp->pfs_1301.creatorid;
768 pf_state_peer_ntoh(&sp->pfs_1301.src, &st->src);
769 pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
770
771 st->rule = r;
772 st->nat_rule = NULL;
773 st->anchor = NULL;
774
775 st->pfsync_time = time_uptime;
776 st->sync_state = PFSYNC_S_NONE;
777
778 if (!(flags & PFSYNC_SI_IOCTL))
779 st->state_flags |= PFSTATE_NOSYNC;
780
781 if ((error = pf_state_insert(kif, kif, skw, sks, st)) != 0)
782 goto cleanup_state;
783
784 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
785 counter_u64_add(r->states_cur, 1);
786 counter_u64_add(r->states_tot, 1);
787
788 if (!(flags & PFSYNC_SI_IOCTL)) {
789 st->state_flags &= ~PFSTATE_NOSYNC;
790 if (st->state_flags & PFSTATE_ACK) {
791 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
792 PFSYNC_BUCKET_LOCK(b);
793 pfsync_q_ins(st, PFSYNC_S_IACK, true);
794 PFSYNC_BUCKET_UNLOCK(b);
795
796 pfsync_push_all(sc);
797 }
798 }
799 st->state_flags &= ~PFSTATE_ACK;
800 PF_STATE_UNLOCK(st);
801
802 return (0);
803
804 cleanup:
805 error = ENOMEM;
806
807 if (skw == sks)
808 sks = NULL;
809 uma_zfree(V_pf_state_key_z, skw);
810 uma_zfree(V_pf_state_key_z, sks);
811
812 cleanup_state: /* pf_state_insert() frees the state keys. */
813 if (st) {
814 st->timeout = PFTM_UNLINKED; /* appease an assert */
815 pf_free_state(st);
816 }
817 return (error);
818 }
819
820 #ifdef INET
821 static int
pfsync_input(struct mbuf ** mp,int * offp __unused,int proto __unused)822 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused)
823 {
824 struct pfsync_softc *sc = V_pfsyncif;
825 struct mbuf *m = *mp;
826 struct ip *ip = mtod(m, struct ip *);
827 struct pfsync_header *ph;
828 struct pfsync_subheader subh;
829
830 int offset, len, flags = 0;
831 int rv;
832 uint16_t count;
833
834 PF_RULES_RLOCK_TRACKER;
835
836 *mp = NULL;
837 V_pfsyncstats.pfsyncs_ipackets++;
838
839 /* Verify that we have a sync interface configured. */
840 if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
841 (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
842 goto done;
843
844 /* verify that the packet came in on the right interface */
845 if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
846 V_pfsyncstats.pfsyncs_badif++;
847 goto done;
848 }
849
850 if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
851 if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
852 /* verify that the IP TTL is 255. */
853 if (ip->ip_ttl != PFSYNC_DFLTTL) {
854 V_pfsyncstats.pfsyncs_badttl++;
855 goto done;
856 }
857
858 offset = ip->ip_hl << 2;
859 if (m->m_pkthdr.len < offset + sizeof(*ph)) {
860 V_pfsyncstats.pfsyncs_hdrops++;
861 goto done;
862 }
863
864 if (offset + sizeof(*ph) > m->m_len) {
865 if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
866 V_pfsyncstats.pfsyncs_hdrops++;
867 return (IPPROTO_DONE);
868 }
869 ip = mtod(m, struct ip *);
870 }
871 ph = (struct pfsync_header *)((char *)ip + offset);
872
873 /* verify the version */
874 if (ph->version != PFSYNC_VERSION) {
875 V_pfsyncstats.pfsyncs_badver++;
876 goto done;
877 }
878
879 len = ntohs(ph->len) + offset;
880 if (m->m_pkthdr.len < len) {
881 V_pfsyncstats.pfsyncs_badlen++;
882 goto done;
883 }
884
885 /*
886 * Trusting pf_chksum during packet processing, as well as seeking
887 * in interface name tree, require holding PF_RULES_RLOCK().
888 */
889 PF_RULES_RLOCK();
890 if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
891 flags = PFSYNC_SI_CKSUM;
892
893 offset += sizeof(*ph);
894 while (offset <= len - sizeof(subh)) {
895 m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
896 offset += sizeof(subh);
897
898 if (subh.action >= PFSYNC_ACT_MAX) {
899 V_pfsyncstats.pfsyncs_badact++;
900 PF_RULES_RUNLOCK();
901 goto done;
902 }
903
904 count = ntohs(subh.count);
905 V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
906 rv = (*pfsync_acts[subh.action])(m, offset, count, flags, subh.action);
907 if (rv == -1) {
908 PF_RULES_RUNLOCK();
909 return (IPPROTO_DONE);
910 }
911
912 offset += rv;
913 }
914 PF_RULES_RUNLOCK();
915
916 done:
917 m_freem(m);
918 return (IPPROTO_DONE);
919 }
920 #endif
921
922 #ifdef INET6
923 static int
pfsync6_input(struct mbuf ** mp,int * offp __unused,int proto __unused)924 pfsync6_input(struct mbuf **mp, int *offp __unused, int proto __unused)
925 {
926 struct pfsync_softc *sc = V_pfsyncif;
927 struct mbuf *m = *mp;
928 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
929 struct pfsync_header *ph;
930 struct pfsync_subheader subh;
931
932 int offset, len, flags = 0;
933 int rv;
934 uint16_t count;
935
936 PF_RULES_RLOCK_TRACKER;
937
938 *mp = NULL;
939 V_pfsyncstats.pfsyncs_ipackets++;
940
941 /* Verify that we have a sync interface configured. */
942 if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
943 (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
944 goto done;
945
946 /* verify that the packet came in on the right interface */
947 if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
948 V_pfsyncstats.pfsyncs_badif++;
949 goto done;
950 }
951
952 if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
953 if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
954 /* verify that the IP TTL is 255. */
955 if (ip6->ip6_hlim != PFSYNC_DFLTTL) {
956 V_pfsyncstats.pfsyncs_badttl++;
957 goto done;
958 }
959
960
961 offset = sizeof(*ip6);
962 if (m->m_pkthdr.len < offset + sizeof(*ph)) {
963 V_pfsyncstats.pfsyncs_hdrops++;
964 goto done;
965 }
966
967 if (offset + sizeof(*ph) > m->m_len) {
968 if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
969 V_pfsyncstats.pfsyncs_hdrops++;
970 return (IPPROTO_DONE);
971 }
972 ip6 = mtod(m, struct ip6_hdr *);
973 }
974 ph = (struct pfsync_header *)((char *)ip6 + offset);
975
976 /* verify the version */
977 if (ph->version != PFSYNC_VERSION) {
978 V_pfsyncstats.pfsyncs_badver++;
979 goto done;
980 }
981
982 len = ntohs(ph->len) + offset;
983 if (m->m_pkthdr.len < len) {
984 V_pfsyncstats.pfsyncs_badlen++;
985 goto done;
986 }
987
988 /*
989 * Trusting pf_chksum during packet processing, as well as seeking
990 * in interface name tree, require holding PF_RULES_RLOCK().
991 */
992 PF_RULES_RLOCK();
993 if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
994 flags = PFSYNC_SI_CKSUM;
995
996 offset += sizeof(*ph);
997 while (offset <= len - sizeof(subh)) {
998 m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
999 offset += sizeof(subh);
1000
1001 if (subh.action >= PFSYNC_ACT_MAX) {
1002 V_pfsyncstats.pfsyncs_badact++;
1003 PF_RULES_RUNLOCK();
1004 goto done;
1005 }
1006
1007 count = ntohs(subh.count);
1008 V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
1009 rv = (*pfsync_acts[subh.action])(m, offset, count, flags, subh.action);
1010 if (rv == -1) {
1011 PF_RULES_RUNLOCK();
1012 return (IPPROTO_DONE);
1013 }
1014
1015 offset += rv;
1016 }
1017 PF_RULES_RUNLOCK();
1018
1019 done:
1020 m_freem(m);
1021 return (IPPROTO_DONE);
1022 }
1023 #endif
1024
1025 static int
pfsync_in_clr(struct mbuf * m,int offset,int count,int flags,int action)1026 pfsync_in_clr(struct mbuf *m, int offset, int count, int flags, int action)
1027 {
1028 struct pfsync_clr *clr;
1029 struct mbuf *mp;
1030 int len = sizeof(*clr) * count;
1031 int i, offp;
1032 u_int32_t creatorid;
1033
1034 mp = m_pulldown(m, offset, len, &offp);
1035 if (mp == NULL) {
1036 V_pfsyncstats.pfsyncs_badlen++;
1037 return (-1);
1038 }
1039 clr = (struct pfsync_clr *)(mp->m_data + offp);
1040
1041 for (i = 0; i < count; i++) {
1042 creatorid = clr[i].creatorid;
1043
1044 if (clr[i].ifname[0] != '\0' &&
1045 pfi_kkif_find(clr[i].ifname) == NULL)
1046 continue;
1047
1048 for (int i = 0; i <= V_pf_hashmask; i++) {
1049 struct pf_idhash *ih = &V_pf_idhash[i];
1050 struct pf_kstate *s;
1051 relock:
1052 PF_HASHROW_LOCK(ih);
1053 LIST_FOREACH(s, &ih->states, entry) {
1054 if (s->creatorid == creatorid) {
1055 s->state_flags |= PFSTATE_NOSYNC;
1056 pf_remove_state(s);
1057 goto relock;
1058 }
1059 }
1060 PF_HASHROW_UNLOCK(ih);
1061 }
1062 }
1063
1064 return (len);
1065 }
1066
1067 static int
pfsync_in_ins(struct mbuf * m,int offset,int count,int flags,int action)1068 pfsync_in_ins(struct mbuf *m, int offset, int count, int flags, int action)
1069 {
1070 struct mbuf *mp;
1071 union pfsync_state_union *sa, *sp;
1072 int i, offp, total_len, msg_version, msg_len;
1073
1074 switch (action) {
1075 case PFSYNC_ACT_INS_1301:
1076 msg_len = sizeof(struct pfsync_state_1301);
1077 total_len = msg_len * count;
1078 msg_version = PFSYNC_MSG_VERSION_1301;
1079 break;
1080 case PFSYNC_ACT_INS_1400:
1081 msg_len = sizeof(struct pfsync_state_1400);
1082 total_len = msg_len * count;
1083 msg_version = PFSYNC_MSG_VERSION_1400;
1084 break;
1085 default:
1086 V_pfsyncstats.pfsyncs_badact++;
1087 return (-1);
1088 }
1089
1090 mp = m_pulldown(m, offset, total_len, &offp);
1091 if (mp == NULL) {
1092 V_pfsyncstats.pfsyncs_badlen++;
1093 return (-1);
1094 }
1095 sa = (union pfsync_state_union *)(mp->m_data + offp);
1096
1097 for (i = 0; i < count; i++) {
1098 sp = (union pfsync_state_union *)((char *)sa + msg_len * i);
1099
1100 /* Check for invalid values. */
1101 if (sp->pfs_1301.timeout >= PFTM_MAX ||
1102 sp->pfs_1301.src.state > PF_TCPS_PROXY_DST ||
1103 sp->pfs_1301.dst.state > PF_TCPS_PROXY_DST ||
1104 sp->pfs_1301.direction > PF_OUT ||
1105 (sp->pfs_1301.af != AF_INET &&
1106 sp->pfs_1301.af != AF_INET6)) {
1107 if (V_pf_status.debug >= PF_DEBUG_MISC)
1108 printf("%s: invalid value\n", __func__);
1109 V_pfsyncstats.pfsyncs_badval++;
1110 continue;
1111 }
1112
1113 if (pfsync_state_import(sp, flags, msg_version) == ENOMEM)
1114 /* Drop out, but process the rest of the actions. */
1115 break;
1116 }
1117
1118 return (total_len);
1119 }
1120
1121 static int
pfsync_in_iack(struct mbuf * m,int offset,int count,int flags,int action)1122 pfsync_in_iack(struct mbuf *m, int offset, int count, int flags, int action)
1123 {
1124 struct pfsync_ins_ack *ia, *iaa;
1125 struct pf_kstate *st;
1126
1127 struct mbuf *mp;
1128 int len = count * sizeof(*ia);
1129 int offp, i;
1130
1131 mp = m_pulldown(m, offset, len, &offp);
1132 if (mp == NULL) {
1133 V_pfsyncstats.pfsyncs_badlen++;
1134 return (-1);
1135 }
1136 iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
1137
1138 for (i = 0; i < count; i++) {
1139 ia = &iaa[i];
1140
1141 st = pf_find_state_byid(ia->id, ia->creatorid);
1142 if (st == NULL)
1143 continue;
1144
1145 if (st->state_flags & PFSTATE_ACK) {
1146 pfsync_undefer_state(st, 0);
1147 }
1148 PF_STATE_UNLOCK(st);
1149 }
1150 /*
1151 * XXX this is not yet implemented, but we know the size of the
1152 * message so we can skip it.
1153 */
1154
1155 return (count * sizeof(struct pfsync_ins_ack));
1156 }
1157
1158 static int
pfsync_upd_tcp(struct pf_kstate * st,struct pfsync_state_peer * src,struct pfsync_state_peer * dst)1159 pfsync_upd_tcp(struct pf_kstate *st, struct pfsync_state_peer *src,
1160 struct pfsync_state_peer *dst)
1161 {
1162 int sync = 0;
1163
1164 PF_STATE_LOCK_ASSERT(st);
1165
1166 /*
1167 * The state should never go backwards except
1168 * for syn-proxy states. Neither should the
1169 * sequence window slide backwards.
1170 */
1171 if ((st->src.state > src->state &&
1172 (st->src.state < PF_TCPS_PROXY_SRC ||
1173 src->state >= PF_TCPS_PROXY_SRC)) ||
1174
1175 (st->src.state == src->state &&
1176 SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
1177 sync++;
1178 else
1179 pf_state_peer_ntoh(src, &st->src);
1180
1181 if ((st->dst.state > dst->state) ||
1182
1183 (st->dst.state >= TCPS_SYN_SENT &&
1184 SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
1185 sync++;
1186 else
1187 pf_state_peer_ntoh(dst, &st->dst);
1188
1189 return (sync);
1190 }
1191
1192 static int
pfsync_in_upd(struct mbuf * m,int offset,int count,int flags,int action)1193 pfsync_in_upd(struct mbuf *m, int offset, int count, int flags, int action)
1194 {
1195 struct pfsync_softc *sc = V_pfsyncif;
1196 union pfsync_state_union *sa, *sp;
1197 struct pf_kstate *st;
1198 struct mbuf *mp;
1199 int sync, offp, i, total_len, msg_len, msg_version;
1200
1201 switch (action) {
1202 case PFSYNC_ACT_UPD_1301:
1203 msg_len = sizeof(struct pfsync_state_1301);
1204 total_len = msg_len * count;
1205 msg_version = PFSYNC_MSG_VERSION_1301;
1206 break;
1207 case PFSYNC_ACT_UPD_1400:
1208 msg_len = sizeof(struct pfsync_state_1400);
1209 total_len = msg_len * count;
1210 msg_version = PFSYNC_MSG_VERSION_1400;
1211 break;
1212 default:
1213 V_pfsyncstats.pfsyncs_badact++;
1214 return (-1);
1215 }
1216
1217 mp = m_pulldown(m, offset, total_len, &offp);
1218 if (mp == NULL) {
1219 V_pfsyncstats.pfsyncs_badlen++;
1220 return (-1);
1221 }
1222 sa = (union pfsync_state_union *)(mp->m_data + offp);
1223
1224 for (i = 0; i < count; i++) {
1225 sp = (union pfsync_state_union *)((char *)sa + msg_len * i);
1226
1227 /* check for invalid values */
1228 if (sp->pfs_1301.timeout >= PFTM_MAX ||
1229 sp->pfs_1301.src.state > PF_TCPS_PROXY_DST ||
1230 sp->pfs_1301.dst.state > PF_TCPS_PROXY_DST) {
1231 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1232 printf("pfsync_input: PFSYNC_ACT_UPD: "
1233 "invalid value\n");
1234 }
1235 V_pfsyncstats.pfsyncs_badval++;
1236 continue;
1237 }
1238
1239 st = pf_find_state_byid(sp->pfs_1301.id, sp->pfs_1301.creatorid);
1240 if (st == NULL) {
1241 /* insert the update */
1242 if (pfsync_state_import(sp, flags, msg_version))
1243 V_pfsyncstats.pfsyncs_badstate++;
1244 continue;
1245 }
1246
1247 if (st->state_flags & PFSTATE_ACK) {
1248 pfsync_undefer_state(st, 1);
1249 }
1250
1251 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1252 sync = pfsync_upd_tcp(st, &sp->pfs_1301.src, &sp->pfs_1301.dst);
1253 else {
1254 sync = 0;
1255
1256 /*
1257 * Non-TCP protocol state machine always go
1258 * forwards
1259 */
1260 if (st->src.state > sp->pfs_1301.src.state)
1261 sync++;
1262 else
1263 pf_state_peer_ntoh(&sp->pfs_1301.src, &st->src);
1264 if (st->dst.state > sp->pfs_1301.dst.state)
1265 sync++;
1266 else
1267 pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
1268 }
1269 if (sync < 2) {
1270 pfsync_alloc_scrub_memory(&sp->pfs_1301.dst, &st->dst);
1271 pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
1272 st->expire = pf_get_uptime();
1273 st->timeout = sp->pfs_1301.timeout;
1274 }
1275 st->pfsync_time = time_uptime;
1276
1277 if (sync) {
1278 V_pfsyncstats.pfsyncs_stale++;
1279
1280 pfsync_update_state(st);
1281 PF_STATE_UNLOCK(st);
1282 pfsync_push_all(sc);
1283 continue;
1284 }
1285 PF_STATE_UNLOCK(st);
1286 }
1287
1288 return (total_len);
1289 }
1290
1291 static int
pfsync_in_upd_c(struct mbuf * m,int offset,int count,int flags,int action)1292 pfsync_in_upd_c(struct mbuf *m, int offset, int count, int flags, int action)
1293 {
1294 struct pfsync_softc *sc = V_pfsyncif;
1295 struct pfsync_upd_c *ua, *up;
1296 struct pf_kstate *st;
1297 int len = count * sizeof(*up);
1298 int sync;
1299 struct mbuf *mp;
1300 int offp, i;
1301
1302 mp = m_pulldown(m, offset, len, &offp);
1303 if (mp == NULL) {
1304 V_pfsyncstats.pfsyncs_badlen++;
1305 return (-1);
1306 }
1307 ua = (struct pfsync_upd_c *)(mp->m_data + offp);
1308
1309 for (i = 0; i < count; i++) {
1310 up = &ua[i];
1311
1312 /* check for invalid values */
1313 if (up->timeout >= PFTM_MAX ||
1314 up->src.state > PF_TCPS_PROXY_DST ||
1315 up->dst.state > PF_TCPS_PROXY_DST) {
1316 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1317 printf("pfsync_input: "
1318 "PFSYNC_ACT_UPD_C: "
1319 "invalid value\n");
1320 }
1321 V_pfsyncstats.pfsyncs_badval++;
1322 continue;
1323 }
1324
1325 st = pf_find_state_byid(up->id, up->creatorid);
1326 if (st == NULL) {
1327 /* We don't have this state. Ask for it. */
1328 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
1329 pfsync_request_update(up->creatorid, up->id);
1330 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
1331 continue;
1332 }
1333
1334 if (st->state_flags & PFSTATE_ACK) {
1335 pfsync_undefer_state(st, 1);
1336 }
1337
1338 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1339 sync = pfsync_upd_tcp(st, &up->src, &up->dst);
1340 else {
1341 sync = 0;
1342
1343 /*
1344 * Non-TCP protocol state machine always go
1345 * forwards
1346 */
1347 if (st->src.state > up->src.state)
1348 sync++;
1349 else
1350 pf_state_peer_ntoh(&up->src, &st->src);
1351 if (st->dst.state > up->dst.state)
1352 sync++;
1353 else
1354 pf_state_peer_ntoh(&up->dst, &st->dst);
1355 }
1356 if (sync < 2) {
1357 pfsync_alloc_scrub_memory(&up->dst, &st->dst);
1358 pf_state_peer_ntoh(&up->dst, &st->dst);
1359 st->expire = pf_get_uptime();
1360 st->timeout = up->timeout;
1361 }
1362 st->pfsync_time = time_uptime;
1363
1364 if (sync) {
1365 V_pfsyncstats.pfsyncs_stale++;
1366
1367 pfsync_update_state(st);
1368 PF_STATE_UNLOCK(st);
1369 pfsync_push_all(sc);
1370 continue;
1371 }
1372 PF_STATE_UNLOCK(st);
1373 }
1374
1375 return (len);
1376 }
1377
1378 static int
pfsync_in_ureq(struct mbuf * m,int offset,int count,int flags,int action)1379 pfsync_in_ureq(struct mbuf *m, int offset, int count, int flags, int action)
1380 {
1381 struct pfsync_upd_req *ur, *ura;
1382 struct mbuf *mp;
1383 int len = count * sizeof(*ur);
1384 int i, offp;
1385
1386 struct pf_kstate *st;
1387
1388 mp = m_pulldown(m, offset, len, &offp);
1389 if (mp == NULL) {
1390 V_pfsyncstats.pfsyncs_badlen++;
1391 return (-1);
1392 }
1393 ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1394
1395 for (i = 0; i < count; i++) {
1396 ur = &ura[i];
1397
1398 if (ur->id == 0 && ur->creatorid == 0)
1399 pfsync_bulk_start();
1400 else {
1401 st = pf_find_state_byid(ur->id, ur->creatorid);
1402 if (st == NULL) {
1403 V_pfsyncstats.pfsyncs_badstate++;
1404 continue;
1405 }
1406 if (st->state_flags & PFSTATE_NOSYNC) {
1407 PF_STATE_UNLOCK(st);
1408 continue;
1409 }
1410
1411 pfsync_update_state_req(st);
1412 PF_STATE_UNLOCK(st);
1413 }
1414 }
1415
1416 return (len);
1417 }
1418
1419 static int
pfsync_in_del_c(struct mbuf * m,int offset,int count,int flags,int action)1420 pfsync_in_del_c(struct mbuf *m, int offset, int count, int flags, int action)
1421 {
1422 struct mbuf *mp;
1423 struct pfsync_del_c *sa, *sp;
1424 struct pf_kstate *st;
1425 int len = count * sizeof(*sp);
1426 int offp, i;
1427
1428 mp = m_pulldown(m, offset, len, &offp);
1429 if (mp == NULL) {
1430 V_pfsyncstats.pfsyncs_badlen++;
1431 return (-1);
1432 }
1433 sa = (struct pfsync_del_c *)(mp->m_data + offp);
1434
1435 for (i = 0; i < count; i++) {
1436 sp = &sa[i];
1437
1438 st = pf_find_state_byid(sp->id, sp->creatorid);
1439 if (st == NULL) {
1440 V_pfsyncstats.pfsyncs_badstate++;
1441 continue;
1442 }
1443
1444 st->state_flags |= PFSTATE_NOSYNC;
1445 pf_remove_state(st);
1446 }
1447
1448 return (len);
1449 }
1450
1451 static int
pfsync_in_bus(struct mbuf * m,int offset,int count,int flags,int action)1452 pfsync_in_bus(struct mbuf *m, int offset, int count, int flags, int action)
1453 {
1454 struct pfsync_softc *sc = V_pfsyncif;
1455 struct pfsync_bus *bus;
1456 struct mbuf *mp;
1457 int len = count * sizeof(*bus);
1458 int offp;
1459
1460 PFSYNC_BLOCK(sc);
1461
1462 /* If we're not waiting for a bulk update, who cares. */
1463 if (sc->sc_ureq_sent == 0) {
1464 PFSYNC_BUNLOCK(sc);
1465 return (len);
1466 }
1467
1468 mp = m_pulldown(m, offset, len, &offp);
1469 if (mp == NULL) {
1470 PFSYNC_BUNLOCK(sc);
1471 V_pfsyncstats.pfsyncs_badlen++;
1472 return (-1);
1473 }
1474 bus = (struct pfsync_bus *)(mp->m_data + offp);
1475
1476 switch (bus->status) {
1477 case PFSYNC_BUS_START:
1478 callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1479 V_pf_limits[PF_LIMIT_STATES].limit /
1480 ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1481 sizeof(union pfsync_state_union)),
1482 pfsync_bulk_fail, sc);
1483 if (V_pf_status.debug >= PF_DEBUG_MISC)
1484 printf("pfsync: received bulk update start\n");
1485 break;
1486
1487 case PFSYNC_BUS_END:
1488 if (time_uptime - ntohl(bus->endtime) >=
1489 sc->sc_ureq_sent) {
1490 /* that's it, we're happy */
1491 sc->sc_ureq_sent = 0;
1492 sc->sc_bulk_tries = 0;
1493 callout_stop(&sc->sc_bulkfail_tmo);
1494 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1495 (*carp_demote_adj_p)(-V_pfsync_carp_adj,
1496 "pfsync bulk done");
1497 sc->sc_flags |= PFSYNCF_OK;
1498 if (V_pf_status.debug >= PF_DEBUG_MISC)
1499 printf("pfsync: received valid "
1500 "bulk update end\n");
1501 } else {
1502 if (V_pf_status.debug >= PF_DEBUG_MISC)
1503 printf("pfsync: received invalid "
1504 "bulk update end: bad timestamp\n");
1505 }
1506 break;
1507 }
1508 PFSYNC_BUNLOCK(sc);
1509
1510 return (len);
1511 }
1512
1513 static int
pfsync_in_tdb(struct mbuf * m,int offset,int count,int flags,int action)1514 pfsync_in_tdb(struct mbuf *m, int offset, int count, int flags, int action)
1515 {
1516 int len = count * sizeof(struct pfsync_tdb);
1517
1518 #if defined(IPSEC)
1519 struct pfsync_tdb *tp;
1520 struct mbuf *mp;
1521 int offp;
1522 int i;
1523 int s;
1524
1525 mp = m_pulldown(m, offset, len, &offp);
1526 if (mp == NULL) {
1527 V_pfsyncstats.pfsyncs_badlen++;
1528 return (-1);
1529 }
1530 tp = (struct pfsync_tdb *)(mp->m_data + offp);
1531
1532 for (i = 0; i < count; i++)
1533 pfsync_update_net_tdb(&tp[i]);
1534 #endif
1535
1536 return (len);
1537 }
1538
1539 #if defined(IPSEC)
1540 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1541 static void
pfsync_update_net_tdb(struct pfsync_tdb * pt)1542 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1543 {
1544 struct tdb *tdb;
1545 int s;
1546
1547 /* check for invalid values */
1548 if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1549 (pt->dst.sa.sa_family != AF_INET &&
1550 pt->dst.sa.sa_family != AF_INET6))
1551 goto bad;
1552
1553 tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1554 if (tdb) {
1555 pt->rpl = ntohl(pt->rpl);
1556 pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1557
1558 /* Neither replay nor byte counter should ever decrease. */
1559 if (pt->rpl < tdb->tdb_rpl ||
1560 pt->cur_bytes < tdb->tdb_cur_bytes) {
1561 goto bad;
1562 }
1563
1564 tdb->tdb_rpl = pt->rpl;
1565 tdb->tdb_cur_bytes = pt->cur_bytes;
1566 }
1567 return;
1568
1569 bad:
1570 if (V_pf_status.debug >= PF_DEBUG_MISC)
1571 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1572 "invalid value\n");
1573 V_pfsyncstats.pfsyncs_badstate++;
1574 return;
1575 }
1576 #endif
1577
1578 static int
pfsync_in_eof(struct mbuf * m,int offset,int count,int flags,int action)1579 pfsync_in_eof(struct mbuf *m, int offset, int count, int flags, int action)
1580 {
1581 /* check if we are at the right place in the packet */
1582 if (offset != m->m_pkthdr.len)
1583 V_pfsyncstats.pfsyncs_badlen++;
1584
1585 /* we're done. free and let the caller return */
1586 m_freem(m);
1587 return (-1);
1588 }
1589
1590 static int
pfsync_in_error(struct mbuf * m,int offset,int count,int flags,int action)1591 pfsync_in_error(struct mbuf *m, int offset, int count, int flags, int action)
1592 {
1593 V_pfsyncstats.pfsyncs_badact++;
1594
1595 m_freem(m);
1596 return (-1);
1597 }
1598
1599 static int
pfsyncoutput(struct ifnet * ifp,struct mbuf * m,const struct sockaddr * dst,struct route * rt)1600 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1601 struct route *rt)
1602 {
1603 m_freem(m);
1604 return (0);
1605 }
1606
1607 /* ARGSUSED */
1608 static int
pfsyncioctl(struct ifnet * ifp,u_long cmd,caddr_t data)1609 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1610 {
1611 struct pfsync_softc *sc = ifp->if_softc;
1612 struct ifreq *ifr = (struct ifreq *)data;
1613 struct pfsyncreq pfsyncr;
1614 size_t nvbuflen;
1615 int error;
1616 int c;
1617
1618 switch (cmd) {
1619 case SIOCSIFFLAGS:
1620 PFSYNC_LOCK(sc);
1621 if (ifp->if_flags & IFF_UP) {
1622 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1623 PFSYNC_UNLOCK(sc);
1624 pfsync_pointers_init();
1625 } else {
1626 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1627 PFSYNC_UNLOCK(sc);
1628 pfsync_pointers_uninit();
1629 }
1630 break;
1631 case SIOCSIFMTU:
1632 if (!sc->sc_sync_if ||
1633 ifr->ifr_mtu <= PFSYNC_MINPKT ||
1634 ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1635 return (EINVAL);
1636 if (ifr->ifr_mtu < ifp->if_mtu) {
1637 for (c = 0; c < pfsync_buckets; c++) {
1638 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
1639 if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT)
1640 pfsync_sendout(1, c);
1641 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
1642 }
1643 }
1644 ifp->if_mtu = ifr->ifr_mtu;
1645 break;
1646 case SIOCGETPFSYNC:
1647 bzero(&pfsyncr, sizeof(pfsyncr));
1648 PFSYNC_LOCK(sc);
1649 if (sc->sc_sync_if) {
1650 strlcpy(pfsyncr.pfsyncr_syncdev,
1651 sc->sc_sync_if->if_xname, IFNAMSIZ);
1652 }
1653 pfsyncr.pfsyncr_syncpeer = ((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr;
1654 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1655 pfsyncr.pfsyncr_defer = sc->sc_flags;
1656 PFSYNC_UNLOCK(sc);
1657 return (copyout(&pfsyncr, ifr_data_get_ptr(ifr),
1658 sizeof(pfsyncr)));
1659
1660 case SIOCGETPFSYNCNV:
1661 {
1662 nvlist_t *nvl_syncpeer;
1663 nvlist_t *nvl = nvlist_create(0);
1664
1665 if (nvl == NULL)
1666 return (ENOMEM);
1667
1668 if (sc->sc_sync_if)
1669 nvlist_add_string(nvl, "syncdev", sc->sc_sync_if->if_xname);
1670 nvlist_add_number(nvl, "maxupdates", sc->sc_maxupdates);
1671 nvlist_add_number(nvl, "flags", sc->sc_flags);
1672 nvlist_add_number(nvl, "version", sc->sc_version);
1673 if ((nvl_syncpeer = pfsync_sockaddr_to_syncpeer_nvlist(&sc->sc_sync_peer)) != NULL)
1674 nvlist_add_nvlist(nvl, "syncpeer", nvl_syncpeer);
1675
1676 void *packed = NULL;
1677 packed = nvlist_pack(nvl, &nvbuflen);
1678 if (packed == NULL) {
1679 free(packed, M_NVLIST);
1680 nvlist_destroy(nvl);
1681 return (ENOMEM);
1682 }
1683
1684 if (nvbuflen > ifr->ifr_cap_nv.buf_length) {
1685 ifr->ifr_cap_nv.length = nvbuflen;
1686 ifr->ifr_cap_nv.buffer = NULL;
1687 free(packed, M_NVLIST);
1688 nvlist_destroy(nvl);
1689 return (EFBIG);
1690 }
1691
1692 ifr->ifr_cap_nv.length = nvbuflen;
1693 error = copyout(packed, ifr->ifr_cap_nv.buffer, nvbuflen);
1694
1695 nvlist_destroy(nvl);
1696 nvlist_destroy(nvl_syncpeer);
1697 free(packed, M_NVLIST);
1698 break;
1699 }
1700
1701 case SIOCSETPFSYNC:
1702 {
1703 struct pfsync_kstatus status;
1704
1705 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1706 return (error);
1707 if ((error = copyin(ifr_data_get_ptr(ifr), &pfsyncr,
1708 sizeof(pfsyncr))))
1709 return (error);
1710
1711 memset((char *)&status, 0, sizeof(struct pfsync_kstatus));
1712 pfsync_pfsyncreq_to_kstatus(&pfsyncr, &status);
1713
1714 error = pfsync_kstatus_to_softc(&status, sc);
1715 return (error);
1716 }
1717 case SIOCSETPFSYNCNV:
1718 {
1719 struct pfsync_kstatus status;
1720 void *data;
1721 nvlist_t *nvl;
1722
1723 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1724 return (error);
1725 if (ifr->ifr_cap_nv.length > IFR_CAP_NV_MAXBUFSIZE)
1726 return (EINVAL);
1727
1728 data = malloc(ifr->ifr_cap_nv.length, M_TEMP, M_WAITOK);
1729
1730 if ((error = copyin(ifr->ifr_cap_nv.buffer, data,
1731 ifr->ifr_cap_nv.length)) != 0) {
1732 free(data, M_TEMP);
1733 return (error);
1734 }
1735
1736 if ((nvl = nvlist_unpack(data, ifr->ifr_cap_nv.length, 0)) == NULL) {
1737 free(data, M_TEMP);
1738 return (EINVAL);
1739 }
1740
1741 memset((char *)&status, 0, sizeof(struct pfsync_kstatus));
1742 pfsync_nvstatus_to_kstatus(nvl, &status);
1743
1744 nvlist_destroy(nvl);
1745 free(data, M_TEMP);
1746
1747 error = pfsync_kstatus_to_softc(&status, sc);
1748 return (error);
1749 }
1750 default:
1751 return (ENOTTY);
1752 }
1753
1754 return (0);
1755 }
1756
1757 static void
pfsync_out_state_1301(struct pf_kstate * st,void * buf)1758 pfsync_out_state_1301(struct pf_kstate *st, void *buf)
1759 {
1760 union pfsync_state_union *sp = buf;
1761
1762 pfsync_state_export(sp, st, PFSYNC_MSG_VERSION_1301);
1763 }
1764
1765 static void
pfsync_out_state_1400(struct pf_kstate * st,void * buf)1766 pfsync_out_state_1400(struct pf_kstate *st, void *buf)
1767 {
1768 union pfsync_state_union *sp = buf;
1769
1770 pfsync_state_export(sp, st, PFSYNC_MSG_VERSION_1400);
1771 }
1772
1773 static void
pfsync_out_iack(struct pf_kstate * st,void * buf)1774 pfsync_out_iack(struct pf_kstate *st, void *buf)
1775 {
1776 struct pfsync_ins_ack *iack = buf;
1777
1778 iack->id = st->id;
1779 iack->creatorid = st->creatorid;
1780 }
1781
1782 static void
pfsync_out_upd_c(struct pf_kstate * st,void * buf)1783 pfsync_out_upd_c(struct pf_kstate *st, void *buf)
1784 {
1785 struct pfsync_upd_c *up = buf;
1786
1787 bzero(up, sizeof(*up));
1788 up->id = st->id;
1789 pf_state_peer_hton(&st->src, &up->src);
1790 pf_state_peer_hton(&st->dst, &up->dst);
1791 up->creatorid = st->creatorid;
1792 up->timeout = st->timeout;
1793 }
1794
1795 static void
pfsync_out_del_c(struct pf_kstate * st,void * buf)1796 pfsync_out_del_c(struct pf_kstate *st, void *buf)
1797 {
1798 struct pfsync_del_c *dp = buf;
1799
1800 dp->id = st->id;
1801 dp->creatorid = st->creatorid;
1802 st->state_flags |= PFSTATE_NOSYNC;
1803 }
1804
1805 static void
pfsync_drop_all(struct pfsync_softc * sc)1806 pfsync_drop_all(struct pfsync_softc *sc)
1807 {
1808 struct pfsync_bucket *b;
1809 int c;
1810
1811 for (c = 0; c < pfsync_buckets; c++) {
1812 b = &sc->sc_buckets[c];
1813
1814 PFSYNC_BUCKET_LOCK(b);
1815 pfsync_drop(sc, c);
1816 PFSYNC_BUCKET_UNLOCK(b);
1817 }
1818 }
1819
1820 static void
pfsync_drop(struct pfsync_softc * sc,int c)1821 pfsync_drop(struct pfsync_softc *sc, int c)
1822 {
1823 struct pf_kstate *st, *next;
1824 struct pfsync_upd_req_item *ur;
1825 struct pfsync_bucket *b;
1826 enum pfsync_q_id q;
1827
1828 b = &sc->sc_buckets[c];
1829 PFSYNC_BUCKET_LOCK_ASSERT(b);
1830
1831 for (q = 0; q < PFSYNC_Q_COUNT; q++) {
1832 if (TAILQ_EMPTY(&b->b_qs[q]))
1833 continue;
1834
1835 TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, next) {
1836 KASSERT(st->sync_state == pfsync_qid_sstate[q],
1837 ("%s: st->sync_state %d == q %d",
1838 __func__, st->sync_state, q));
1839 st->sync_state = PFSYNC_S_NONE;
1840 pf_release_state(st);
1841 }
1842 TAILQ_INIT(&b->b_qs[q]);
1843 }
1844
1845 while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1846 TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1847 free(ur, M_PFSYNC);
1848 }
1849
1850 b->b_len = PFSYNC_MINPKT;
1851 free(b->b_plus, M_PFSYNC);
1852 b->b_plus = NULL;
1853 b->b_pluslen = 0;
1854 }
1855
1856 static void
pfsync_sendout(int schedswi,int c)1857 pfsync_sendout(int schedswi, int c)
1858 {
1859 struct pfsync_softc *sc = V_pfsyncif;
1860 struct ifnet *ifp = sc->sc_ifp;
1861 struct mbuf *m;
1862 struct pfsync_header *ph;
1863 struct pfsync_subheader *subh;
1864 struct pf_kstate *st, *st_next;
1865 struct pfsync_upd_req_item *ur;
1866 struct pfsync_bucket *b = &sc->sc_buckets[c];
1867 size_t len;
1868 int aflen, offset, count = 0;
1869 enum pfsync_q_id q;
1870
1871 KASSERT(sc != NULL, ("%s: null sc", __func__));
1872 KASSERT(b->b_len > PFSYNC_MINPKT,
1873 ("%s: sc_len %zu", __func__, b->b_len));
1874 PFSYNC_BUCKET_LOCK_ASSERT(b);
1875
1876 if (!bpf_peers_present(ifp->if_bpf) && sc->sc_sync_if == NULL) {
1877 pfsync_drop(sc, c);
1878 return;
1879 }
1880
1881 m = m_get2(max_linkhdr + b->b_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1882 if (m == NULL) {
1883 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
1884 V_pfsyncstats.pfsyncs_onomem++;
1885 return;
1886 }
1887 m->m_data += max_linkhdr;
1888 bzero(m->m_data, b->b_len);
1889
1890 len = b->b_len;
1891
1892 /* build the ip header */
1893 switch (sc->sc_sync_peer.ss_family) {
1894 #ifdef INET
1895 case AF_INET:
1896 {
1897 struct ip *ip;
1898
1899 ip = mtod(m, struct ip *);
1900 bcopy(&sc->sc_template.ipv4, ip, sizeof(*ip));
1901 aflen = offset = sizeof(*ip);
1902
1903 len -= sizeof(union inet_template) - sizeof(struct ip);
1904 ip->ip_len = htons(len);
1905 ip_fillid(ip, V_ip_random_id);
1906 break;
1907 }
1908 #endif
1909 #ifdef INET6
1910 case AF_INET6:
1911 {
1912 struct ip6_hdr *ip6;
1913
1914 ip6 = mtod(m, struct ip6_hdr *);
1915 bcopy(&sc->sc_template.ipv6, ip6, sizeof(*ip6));
1916 aflen = offset = sizeof(*ip6);
1917
1918 len -= sizeof(union inet_template) - sizeof(struct ip6_hdr);
1919 ip6->ip6_plen = htons(len);
1920 break;
1921 }
1922 #endif
1923 default:
1924 m_freem(m);
1925 pfsync_drop(sc, c);
1926 return;
1927 }
1928 m->m_len = m->m_pkthdr.len = len;
1929
1930 /* build the pfsync header */
1931 ph = (struct pfsync_header *)(m->m_data + offset);
1932 offset += sizeof(*ph);
1933
1934 ph->version = PFSYNC_VERSION;
1935 ph->len = htons(len - aflen);
1936 bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1937
1938 /* walk the queues */
1939 for (q = 0; q < PFSYNC_Q_COUNT; q++) {
1940 if (TAILQ_EMPTY(&b->b_qs[q]))
1941 continue;
1942
1943 subh = (struct pfsync_subheader *)(m->m_data + offset);
1944 offset += sizeof(*subh);
1945
1946 count = 0;
1947 TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, st_next) {
1948 KASSERT(st->sync_state == pfsync_qid_sstate[q],
1949 ("%s: st->sync_state == q",
1950 __func__));
1951 /*
1952 * XXXGL: some of write methods do unlocked reads
1953 * of state data :(
1954 */
1955 pfsync_qs[q].write(st, m->m_data + offset);
1956 offset += pfsync_qs[q].len;
1957 st->sync_state = PFSYNC_S_NONE;
1958 pf_release_state(st);
1959 count++;
1960 }
1961 TAILQ_INIT(&b->b_qs[q]);
1962
1963 subh->action = pfsync_qs[q].action;
1964 subh->count = htons(count);
1965 V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1966 }
1967
1968 if (!TAILQ_EMPTY(&b->b_upd_req_list)) {
1969 subh = (struct pfsync_subheader *)(m->m_data + offset);
1970 offset += sizeof(*subh);
1971
1972 count = 0;
1973 while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1974 TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1975
1976 bcopy(&ur->ur_msg, m->m_data + offset,
1977 sizeof(ur->ur_msg));
1978 offset += sizeof(ur->ur_msg);
1979 free(ur, M_PFSYNC);
1980 count++;
1981 }
1982
1983 subh->action = PFSYNC_ACT_UPD_REQ;
1984 subh->count = htons(count);
1985 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1986 }
1987
1988 /* has someone built a custom region for us to add? */
1989 if (b->b_plus != NULL) {
1990 bcopy(b->b_plus, m->m_data + offset, b->b_pluslen);
1991 offset += b->b_pluslen;
1992
1993 free(b->b_plus, M_PFSYNC);
1994 b->b_plus = NULL;
1995 b->b_pluslen = 0;
1996 }
1997
1998 subh = (struct pfsync_subheader *)(m->m_data + offset);
1999 offset += sizeof(*subh);
2000
2001 subh->action = PFSYNC_ACT_EOF;
2002 subh->count = htons(1);
2003 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
2004
2005 /* we're done, let's put it on the wire */
2006 if (bpf_peers_present(ifp->if_bpf)) {
2007 m->m_data += aflen;
2008 m->m_len = m->m_pkthdr.len = len - aflen;
2009 bpf_mtap(ifp->if_bpf, m);
2010 m->m_data -= aflen;
2011 m->m_len = m->m_pkthdr.len = len;
2012 }
2013
2014 if (sc->sc_sync_if == NULL) {
2015 b->b_len = PFSYNC_MINPKT;
2016 m_freem(m);
2017 return;
2018 }
2019
2020 if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1);
2021 if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len);
2022 b->b_len = PFSYNC_MINPKT;
2023
2024 if (!_IF_QFULL(&b->b_snd))
2025 _IF_ENQUEUE(&b->b_snd, m);
2026 else {
2027 m_freem(m);
2028 if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1);
2029 }
2030 if (schedswi)
2031 swi_sched(V_pfsync_swi_cookie, 0);
2032 }
2033
2034 static void
pfsync_insert_state(struct pf_kstate * st)2035 pfsync_insert_state(struct pf_kstate *st)
2036 {
2037 struct pfsync_softc *sc = V_pfsyncif;
2038 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2039
2040 if (st->state_flags & PFSTATE_NOSYNC)
2041 return;
2042
2043 if ((st->rule->rule_flag & PFRULE_NOSYNC) ||
2044 st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
2045 st->state_flags |= PFSTATE_NOSYNC;
2046 return;
2047 }
2048
2049 KASSERT(st->sync_state == PFSYNC_S_NONE,
2050 ("%s: st->sync_state %u", __func__, st->sync_state));
2051
2052 PFSYNC_BUCKET_LOCK(b);
2053 if (b->b_len == PFSYNC_MINPKT)
2054 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2055
2056 pfsync_q_ins(st, PFSYNC_S_INS, true);
2057 PFSYNC_BUCKET_UNLOCK(b);
2058
2059 st->sync_updates = 0;
2060 }
2061
2062 static int
pfsync_defer(struct pf_kstate * st,struct mbuf * m)2063 pfsync_defer(struct pf_kstate *st, struct mbuf *m)
2064 {
2065 struct pfsync_softc *sc = V_pfsyncif;
2066 struct pfsync_deferral *pd;
2067 struct pfsync_bucket *b;
2068
2069 if (m->m_flags & (M_BCAST|M_MCAST))
2070 return (0);
2071
2072 if (sc == NULL)
2073 return (0);
2074
2075 b = pfsync_get_bucket(sc, st);
2076
2077 PFSYNC_LOCK(sc);
2078
2079 if (!(sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) ||
2080 !(sc->sc_flags & PFSYNCF_DEFER)) {
2081 PFSYNC_UNLOCK(sc);
2082 return (0);
2083 }
2084
2085 PFSYNC_BUCKET_LOCK(b);
2086 PFSYNC_UNLOCK(sc);
2087
2088 if (b->b_deferred >= 128)
2089 pfsync_undefer(TAILQ_FIRST(&b->b_deferrals), 0);
2090
2091 pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
2092 if (pd == NULL) {
2093 PFSYNC_BUCKET_UNLOCK(b);
2094 return (0);
2095 }
2096 b->b_deferred++;
2097
2098 m->m_flags |= M_SKIP_FIREWALL;
2099 st->state_flags |= PFSTATE_ACK;
2100
2101 pd->pd_sc = sc;
2102 pd->pd_st = st;
2103 pf_ref_state(st);
2104 pd->pd_m = m;
2105
2106 TAILQ_INSERT_TAIL(&b->b_deferrals, pd, pd_entry);
2107 callout_init_mtx(&pd->pd_tmo, &b->b_mtx, CALLOUT_RETURNUNLOCKED);
2108 callout_reset(&pd->pd_tmo, (V_pfsync_defer_timeout * hz) / 1000,
2109 pfsync_defer_tmo, pd);
2110
2111 pfsync_push(b);
2112 PFSYNC_BUCKET_UNLOCK(b);
2113
2114 return (1);
2115 }
2116
2117 static void
pfsync_undefer(struct pfsync_deferral * pd,int drop)2118 pfsync_undefer(struct pfsync_deferral *pd, int drop)
2119 {
2120 struct pfsync_softc *sc = pd->pd_sc;
2121 struct mbuf *m = pd->pd_m;
2122 struct pf_kstate *st = pd->pd_st;
2123 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2124
2125 PFSYNC_BUCKET_LOCK_ASSERT(b);
2126
2127 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
2128 b->b_deferred--;
2129 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */
2130 free(pd, M_PFSYNC);
2131 pf_release_state(st);
2132
2133 if (drop)
2134 m_freem(m);
2135 else {
2136 _IF_ENQUEUE(&b->b_snd, m);
2137 pfsync_push(b);
2138 }
2139 }
2140
2141 static void
pfsync_defer_tmo(void * arg)2142 pfsync_defer_tmo(void *arg)
2143 {
2144 struct epoch_tracker et;
2145 struct pfsync_deferral *pd = arg;
2146 struct pfsync_softc *sc = pd->pd_sc;
2147 struct mbuf *m = pd->pd_m;
2148 struct pf_kstate *st = pd->pd_st;
2149 struct pfsync_bucket *b;
2150
2151 CURVNET_SET(sc->sc_ifp->if_vnet);
2152
2153 b = pfsync_get_bucket(sc, st);
2154
2155 PFSYNC_BUCKET_LOCK_ASSERT(b);
2156
2157 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
2158 b->b_deferred--;
2159 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */
2160 PFSYNC_BUCKET_UNLOCK(b);
2161 free(pd, M_PFSYNC);
2162
2163 if (sc->sc_sync_if == NULL) {
2164 pf_release_state(st);
2165 m_freem(m);
2166 CURVNET_RESTORE();
2167 return;
2168 }
2169
2170 NET_EPOCH_ENTER(et);
2171
2172 pfsync_tx(sc, m);
2173
2174 pf_release_state(st);
2175
2176 CURVNET_RESTORE();
2177 NET_EPOCH_EXIT(et);
2178 }
2179
2180 static void
pfsync_undefer_state_locked(struct pf_kstate * st,int drop)2181 pfsync_undefer_state_locked(struct pf_kstate *st, int drop)
2182 {
2183 struct pfsync_softc *sc = V_pfsyncif;
2184 struct pfsync_deferral *pd;
2185 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2186
2187 PFSYNC_BUCKET_LOCK_ASSERT(b);
2188
2189 TAILQ_FOREACH(pd, &b->b_deferrals, pd_entry) {
2190 if (pd->pd_st == st) {
2191 if (callout_stop(&pd->pd_tmo) > 0)
2192 pfsync_undefer(pd, drop);
2193
2194 return;
2195 }
2196 }
2197
2198 panic("%s: unable to find deferred state", __func__);
2199 }
2200
2201 static void
pfsync_undefer_state(struct pf_kstate * st,int drop)2202 pfsync_undefer_state(struct pf_kstate *st, int drop)
2203 {
2204 struct pfsync_softc *sc = V_pfsyncif;
2205 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2206
2207 PFSYNC_BUCKET_LOCK(b);
2208 pfsync_undefer_state_locked(st, drop);
2209 PFSYNC_BUCKET_UNLOCK(b);
2210 }
2211
2212 static struct pfsync_bucket*
pfsync_get_bucket(struct pfsync_softc * sc,struct pf_kstate * st)2213 pfsync_get_bucket(struct pfsync_softc *sc, struct pf_kstate *st)
2214 {
2215 int c = PF_IDHASH(st) % pfsync_buckets;
2216 return &sc->sc_buckets[c];
2217 }
2218
2219 static void
pfsync_update_state(struct pf_kstate * st)2220 pfsync_update_state(struct pf_kstate *st)
2221 {
2222 struct pfsync_softc *sc = V_pfsyncif;
2223 bool sync = false, ref = true;
2224 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2225
2226 PF_STATE_LOCK_ASSERT(st);
2227 PFSYNC_BUCKET_LOCK(b);
2228
2229 if (st->state_flags & PFSTATE_ACK)
2230 pfsync_undefer_state_locked(st, 0);
2231 if (st->state_flags & PFSTATE_NOSYNC) {
2232 if (st->sync_state != PFSYNC_S_NONE)
2233 pfsync_q_del(st, true, b);
2234 PFSYNC_BUCKET_UNLOCK(b);
2235 return;
2236 }
2237
2238 if (b->b_len == PFSYNC_MINPKT)
2239 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2240
2241 switch (st->sync_state) {
2242 case PFSYNC_S_UPD_C:
2243 case PFSYNC_S_UPD:
2244 case PFSYNC_S_INS:
2245 /* we're already handling it */
2246
2247 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
2248 st->sync_updates++;
2249 if (st->sync_updates >= sc->sc_maxupdates)
2250 sync = true;
2251 }
2252 break;
2253
2254 case PFSYNC_S_IACK:
2255 pfsync_q_del(st, false, b);
2256 ref = false;
2257 /* FALLTHROUGH */
2258
2259 case PFSYNC_S_NONE:
2260 pfsync_q_ins(st, PFSYNC_S_UPD_C, ref);
2261 st->sync_updates = 0;
2262 break;
2263
2264 default:
2265 panic("%s: unexpected sync state %d", __func__, st->sync_state);
2266 }
2267
2268 if (sync || (time_uptime - st->pfsync_time) < 2)
2269 pfsync_push(b);
2270
2271 PFSYNC_BUCKET_UNLOCK(b);
2272 }
2273
2274 static void
pfsync_request_update(u_int32_t creatorid,u_int64_t id)2275 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
2276 {
2277 struct pfsync_softc *sc = V_pfsyncif;
2278 struct pfsync_bucket *b = &sc->sc_buckets[0];
2279 struct pfsync_upd_req_item *item;
2280 size_t nlen = sizeof(struct pfsync_upd_req);
2281
2282 PFSYNC_BUCKET_LOCK_ASSERT(b);
2283
2284 /*
2285 * This code does a bit to prevent multiple update requests for the
2286 * same state being generated. It searches current subheader queue,
2287 * but it doesn't lookup into queue of already packed datagrams.
2288 */
2289 TAILQ_FOREACH(item, &b->b_upd_req_list, ur_entry)
2290 if (item->ur_msg.id == id &&
2291 item->ur_msg.creatorid == creatorid)
2292 return;
2293
2294 item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
2295 if (item == NULL)
2296 return; /* XXX stats */
2297
2298 item->ur_msg.id = id;
2299 item->ur_msg.creatorid = creatorid;
2300
2301 if (TAILQ_EMPTY(&b->b_upd_req_list))
2302 nlen += sizeof(struct pfsync_subheader);
2303
2304 if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2305 pfsync_sendout(0, 0);
2306
2307 nlen = sizeof(struct pfsync_subheader) +
2308 sizeof(struct pfsync_upd_req);
2309 }
2310
2311 TAILQ_INSERT_TAIL(&b->b_upd_req_list, item, ur_entry);
2312 b->b_len += nlen;
2313
2314 pfsync_push(b);
2315 }
2316
2317 static bool
pfsync_update_state_req(struct pf_kstate * st)2318 pfsync_update_state_req(struct pf_kstate *st)
2319 {
2320 struct pfsync_softc *sc = V_pfsyncif;
2321 bool ref = true, full = false;
2322 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2323
2324 PF_STATE_LOCK_ASSERT(st);
2325 PFSYNC_BUCKET_LOCK(b);
2326
2327 if (st->state_flags & PFSTATE_NOSYNC) {
2328 if (st->sync_state != PFSYNC_S_NONE)
2329 pfsync_q_del(st, true, b);
2330 PFSYNC_BUCKET_UNLOCK(b);
2331 return (full);
2332 }
2333
2334 switch (st->sync_state) {
2335 case PFSYNC_S_UPD_C:
2336 case PFSYNC_S_IACK:
2337 pfsync_q_del(st, false, b);
2338 ref = false;
2339 /* FALLTHROUGH */
2340
2341 case PFSYNC_S_NONE:
2342 pfsync_q_ins(st, PFSYNC_S_UPD, ref);
2343 pfsync_push(b);
2344 break;
2345
2346 case PFSYNC_S_INS:
2347 case PFSYNC_S_UPD:
2348 case PFSYNC_S_DEL_C:
2349 /* we're already handling it */
2350 break;
2351
2352 default:
2353 panic("%s: unexpected sync state %d", __func__, st->sync_state);
2354 }
2355
2356 if ((sc->sc_ifp->if_mtu - b->b_len) < sizeof(union pfsync_state_union))
2357 full = true;
2358
2359 PFSYNC_BUCKET_UNLOCK(b);
2360
2361 return (full);
2362 }
2363
2364 static void
pfsync_delete_state(struct pf_kstate * st)2365 pfsync_delete_state(struct pf_kstate *st)
2366 {
2367 struct pfsync_softc *sc = V_pfsyncif;
2368 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2369 bool ref = true;
2370
2371 PFSYNC_BUCKET_LOCK(b);
2372 if (st->state_flags & PFSTATE_ACK)
2373 pfsync_undefer_state_locked(st, 1);
2374 if (st->state_flags & PFSTATE_NOSYNC) {
2375 if (st->sync_state != PFSYNC_S_NONE)
2376 pfsync_q_del(st, true, b);
2377 PFSYNC_BUCKET_UNLOCK(b);
2378 return;
2379 }
2380
2381 if (b->b_len == PFSYNC_MINPKT)
2382 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2383
2384 switch (st->sync_state) {
2385 case PFSYNC_S_INS:
2386 /* We never got to tell the world so just forget about it. */
2387 pfsync_q_del(st, true, b);
2388 break;
2389
2390 case PFSYNC_S_UPD_C:
2391 case PFSYNC_S_UPD:
2392 case PFSYNC_S_IACK:
2393 pfsync_q_del(st, false, b);
2394 ref = false;
2395 /* FALLTHROUGH */
2396
2397 case PFSYNC_S_NONE:
2398 pfsync_q_ins(st, PFSYNC_S_DEL_C, ref);
2399 break;
2400
2401 default:
2402 panic("%s: unexpected sync state %d", __func__, st->sync_state);
2403 }
2404
2405 PFSYNC_BUCKET_UNLOCK(b);
2406 }
2407
2408 static void
pfsync_clear_states(u_int32_t creatorid,const char * ifname)2409 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
2410 {
2411 struct {
2412 struct pfsync_subheader subh;
2413 struct pfsync_clr clr;
2414 } __packed r;
2415
2416 bzero(&r, sizeof(r));
2417
2418 r.subh.action = PFSYNC_ACT_CLR;
2419 r.subh.count = htons(1);
2420 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
2421
2422 strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
2423 r.clr.creatorid = creatorid;
2424
2425 pfsync_send_plus(&r, sizeof(r));
2426 }
2427
2428 static enum pfsync_q_id
pfsync_sstate_to_qid(u_int8_t sync_state)2429 pfsync_sstate_to_qid(u_int8_t sync_state)
2430 {
2431 struct pfsync_softc *sc = V_pfsyncif;
2432
2433 switch (sync_state) {
2434 case PFSYNC_S_INS:
2435 switch (sc->sc_version) {
2436 case PFSYNC_MSG_VERSION_1301:
2437 return PFSYNC_Q_INS_1301;
2438 case PFSYNC_MSG_VERSION_1400:
2439 return PFSYNC_Q_INS_1400;
2440 }
2441 break;
2442 case PFSYNC_S_IACK:
2443 return PFSYNC_Q_IACK;
2444 case PFSYNC_S_UPD:
2445 switch (sc->sc_version) {
2446 case PFSYNC_MSG_VERSION_1301:
2447 return PFSYNC_Q_UPD_1301;
2448 case PFSYNC_MSG_VERSION_1400:
2449 return PFSYNC_Q_UPD_1400;
2450 }
2451 break;
2452 case PFSYNC_S_UPD_C:
2453 return PFSYNC_Q_UPD_C;
2454 case PFSYNC_S_DEL_C:
2455 return PFSYNC_Q_DEL_C;
2456 default:
2457 panic("%s: Unsupported st->sync_state 0x%02x",
2458 __func__, sync_state);
2459 }
2460
2461 panic("%s: Unsupported pfsync_msg_version %d",
2462 __func__, sc->sc_version);
2463 }
2464
2465 static void
pfsync_q_ins(struct pf_kstate * st,int sync_state,bool ref)2466 pfsync_q_ins(struct pf_kstate *st, int sync_state, bool ref)
2467 {
2468 enum pfsync_q_id q = pfsync_sstate_to_qid(sync_state);
2469 struct pfsync_softc *sc = V_pfsyncif;
2470 size_t nlen = pfsync_qs[q].len;
2471 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2472
2473 PFSYNC_BUCKET_LOCK_ASSERT(b);
2474
2475 KASSERT(st->sync_state == PFSYNC_S_NONE,
2476 ("%s: st->sync_state %u", __func__, st->sync_state));
2477 KASSERT(b->b_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
2478 b->b_len));
2479
2480 if (TAILQ_EMPTY(&b->b_qs[q]))
2481 nlen += sizeof(struct pfsync_subheader);
2482
2483 if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2484 pfsync_sendout(1, b->b_id);
2485
2486 nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
2487 }
2488
2489 b->b_len += nlen;
2490 st->sync_state = pfsync_qid_sstate[q];
2491 TAILQ_INSERT_TAIL(&b->b_qs[q], st, sync_list);
2492 if (ref)
2493 pf_ref_state(st);
2494 }
2495
2496 static void
pfsync_q_del(struct pf_kstate * st,bool unref,struct pfsync_bucket * b)2497 pfsync_q_del(struct pf_kstate *st, bool unref, struct pfsync_bucket *b)
2498 {
2499 enum pfsync_q_id q;
2500
2501 PFSYNC_BUCKET_LOCK_ASSERT(b);
2502 KASSERT(st->sync_state != PFSYNC_S_NONE,
2503 ("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2504
2505 q = pfsync_sstate_to_qid(st->sync_state);
2506 b->b_len -= pfsync_qs[q].len;
2507 TAILQ_REMOVE(&b->b_qs[q], st, sync_list);
2508 st->sync_state = PFSYNC_S_NONE;
2509 if (unref)
2510 pf_release_state(st);
2511
2512 if (TAILQ_EMPTY(&b->b_qs[q]))
2513 b->b_len -= sizeof(struct pfsync_subheader);
2514 }
2515
2516 static void
pfsync_bulk_start(void)2517 pfsync_bulk_start(void)
2518 {
2519 struct pfsync_softc *sc = V_pfsyncif;
2520
2521 if (V_pf_status.debug >= PF_DEBUG_MISC)
2522 printf("pfsync: received bulk update request\n");
2523
2524 PFSYNC_BLOCK(sc);
2525
2526 sc->sc_ureq_received = time_uptime;
2527 sc->sc_bulk_hashid = 0;
2528 sc->sc_bulk_stateid = 0;
2529 pfsync_bulk_status(PFSYNC_BUS_START);
2530 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2531 PFSYNC_BUNLOCK(sc);
2532 }
2533
2534 static void
pfsync_bulk_update(void * arg)2535 pfsync_bulk_update(void *arg)
2536 {
2537 struct pfsync_softc *sc = arg;
2538 struct pf_kstate *s;
2539 int i;
2540
2541 PFSYNC_BLOCK_ASSERT(sc);
2542 CURVNET_SET(sc->sc_ifp->if_vnet);
2543
2544 /*
2545 * Start with last state from previous invocation.
2546 * It may had gone, in this case start from the
2547 * hash slot.
2548 */
2549 s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2550
2551 if (s != NULL)
2552 i = PF_IDHASH(s);
2553 else
2554 i = sc->sc_bulk_hashid;
2555
2556 for (; i <= V_pf_hashmask; i++) {
2557 struct pf_idhash *ih = &V_pf_idhash[i];
2558
2559 if (s != NULL)
2560 PF_HASHROW_ASSERT(ih);
2561 else {
2562 PF_HASHROW_LOCK(ih);
2563 s = LIST_FIRST(&ih->states);
2564 }
2565
2566 for (; s; s = LIST_NEXT(s, entry)) {
2567 if (s->sync_state == PFSYNC_S_NONE &&
2568 s->timeout < PFTM_MAX &&
2569 s->pfsync_time <= sc->sc_ureq_received) {
2570 if (pfsync_update_state_req(s)) {
2571 /* We've filled a packet. */
2572 sc->sc_bulk_hashid = i;
2573 sc->sc_bulk_stateid = s->id;
2574 sc->sc_bulk_creatorid = s->creatorid;
2575 PF_HASHROW_UNLOCK(ih);
2576 callout_reset(&sc->sc_bulk_tmo, 1,
2577 pfsync_bulk_update, sc);
2578 goto full;
2579 }
2580 }
2581 }
2582 PF_HASHROW_UNLOCK(ih);
2583 }
2584
2585 /* We're done. */
2586 pfsync_bulk_status(PFSYNC_BUS_END);
2587 full:
2588 CURVNET_RESTORE();
2589 }
2590
2591 static void
pfsync_bulk_status(u_int8_t status)2592 pfsync_bulk_status(u_int8_t status)
2593 {
2594 struct {
2595 struct pfsync_subheader subh;
2596 struct pfsync_bus bus;
2597 } __packed r;
2598
2599 struct pfsync_softc *sc = V_pfsyncif;
2600
2601 bzero(&r, sizeof(r));
2602
2603 r.subh.action = PFSYNC_ACT_BUS;
2604 r.subh.count = htons(1);
2605 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2606
2607 r.bus.creatorid = V_pf_status.hostid;
2608 r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2609 r.bus.status = status;
2610
2611 pfsync_send_plus(&r, sizeof(r));
2612 }
2613
2614 static void
pfsync_bulk_fail(void * arg)2615 pfsync_bulk_fail(void *arg)
2616 {
2617 struct pfsync_softc *sc = arg;
2618 struct pfsync_bucket *b = &sc->sc_buckets[0];
2619
2620 CURVNET_SET(sc->sc_ifp->if_vnet);
2621
2622 PFSYNC_BLOCK_ASSERT(sc);
2623
2624 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2625 /* Try again */
2626 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2627 pfsync_bulk_fail, V_pfsyncif);
2628 PFSYNC_BUCKET_LOCK(b);
2629 pfsync_request_update(0, 0);
2630 PFSYNC_BUCKET_UNLOCK(b);
2631 } else {
2632 /* Pretend like the transfer was ok. */
2633 sc->sc_ureq_sent = 0;
2634 sc->sc_bulk_tries = 0;
2635 PFSYNC_LOCK(sc);
2636 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2637 (*carp_demote_adj_p)(-V_pfsync_carp_adj,
2638 "pfsync bulk fail");
2639 sc->sc_flags |= PFSYNCF_OK;
2640 PFSYNC_UNLOCK(sc);
2641 if (V_pf_status.debug >= PF_DEBUG_MISC)
2642 printf("pfsync: failed to receive bulk update\n");
2643 }
2644
2645 CURVNET_RESTORE();
2646 }
2647
2648 static void
pfsync_send_plus(void * plus,size_t pluslen)2649 pfsync_send_plus(void *plus, size_t pluslen)
2650 {
2651 struct pfsync_softc *sc = V_pfsyncif;
2652 struct pfsync_bucket *b = &sc->sc_buckets[0];
2653 uint8_t *newplus;
2654
2655 PFSYNC_BUCKET_LOCK(b);
2656
2657 if (b->b_len + pluslen > sc->sc_ifp->if_mtu)
2658 pfsync_sendout(1, b->b_id);
2659
2660 newplus = malloc(pluslen + b->b_pluslen, M_PFSYNC, M_NOWAIT);
2661 if (newplus == NULL)
2662 goto out;
2663
2664 if (b->b_plus != NULL) {
2665 memcpy(newplus, b->b_plus, b->b_pluslen);
2666 free(b->b_plus, M_PFSYNC);
2667 } else {
2668 MPASS(b->b_pluslen == 0);
2669 }
2670 memcpy(newplus + b->b_pluslen, plus, pluslen);
2671
2672 b->b_plus = newplus;
2673 b->b_pluslen += pluslen;
2674 b->b_len += pluslen;
2675
2676 pfsync_sendout(1, b->b_id);
2677
2678 out:
2679 PFSYNC_BUCKET_UNLOCK(b);
2680 }
2681
2682 static void
pfsync_timeout(void * arg)2683 pfsync_timeout(void *arg)
2684 {
2685 struct pfsync_bucket *b = arg;
2686
2687 CURVNET_SET(b->b_sc->sc_ifp->if_vnet);
2688 PFSYNC_BUCKET_LOCK(b);
2689 pfsync_push(b);
2690 PFSYNC_BUCKET_UNLOCK(b);
2691 CURVNET_RESTORE();
2692 }
2693
2694 static void
pfsync_push(struct pfsync_bucket * b)2695 pfsync_push(struct pfsync_bucket *b)
2696 {
2697
2698 PFSYNC_BUCKET_LOCK_ASSERT(b);
2699
2700 b->b_flags |= PFSYNCF_BUCKET_PUSH;
2701 swi_sched(V_pfsync_swi_cookie, 0);
2702 }
2703
2704 static void
pfsync_push_all(struct pfsync_softc * sc)2705 pfsync_push_all(struct pfsync_softc *sc)
2706 {
2707 int c;
2708 struct pfsync_bucket *b;
2709
2710 for (c = 0; c < pfsync_buckets; c++) {
2711 b = &sc->sc_buckets[c];
2712
2713 PFSYNC_BUCKET_LOCK(b);
2714 pfsync_push(b);
2715 PFSYNC_BUCKET_UNLOCK(b);
2716 }
2717 }
2718
2719 static void
pfsync_tx(struct pfsync_softc * sc,struct mbuf * m)2720 pfsync_tx(struct pfsync_softc *sc, struct mbuf *m)
2721 {
2722 struct ip *ip;
2723 int af, error = 0;
2724
2725 ip = mtod(m, struct ip *);
2726 MPASS(ip->ip_v == IPVERSION || ip->ip_v == (IPV6_VERSION >> 4));
2727
2728 af = ip->ip_v == IPVERSION ? AF_INET : AF_INET6;
2729
2730 /*
2731 * We distinguish between a deferral packet and our
2732 * own pfsync packet based on M_SKIP_FIREWALL
2733 * flag. This is XXX.
2734 */
2735 switch (af) {
2736 #ifdef INET
2737 case AF_INET:
2738 if (m->m_flags & M_SKIP_FIREWALL) {
2739 error = ip_output(m, NULL, NULL, 0,
2740 NULL, NULL);
2741 } else {
2742 error = ip_output(m, NULL, NULL,
2743 IP_RAWOUTPUT, &sc->sc_imo, NULL);
2744 }
2745 break;
2746 #endif
2747 #ifdef INET6
2748 case AF_INET6:
2749 if (m->m_flags & M_SKIP_FIREWALL) {
2750 error = ip6_output(m, NULL, NULL, 0,
2751 NULL, NULL, NULL);
2752 } else {
2753 error = ip6_output(m, NULL, NULL, 0,
2754 &sc->sc_im6o, NULL, NULL);
2755 }
2756 break;
2757 #endif
2758 }
2759
2760 if (error == 0)
2761 V_pfsyncstats.pfsyncs_opackets++;
2762 else
2763 V_pfsyncstats.pfsyncs_oerrors++;
2764
2765 }
2766
2767 static void
pfsyncintr(void * arg)2768 pfsyncintr(void *arg)
2769 {
2770 struct epoch_tracker et;
2771 struct pfsync_softc *sc = arg;
2772 struct pfsync_bucket *b;
2773 struct mbuf *m, *n;
2774 int c;
2775
2776 NET_EPOCH_ENTER(et);
2777 CURVNET_SET(sc->sc_ifp->if_vnet);
2778
2779 for (c = 0; c < pfsync_buckets; c++) {
2780 b = &sc->sc_buckets[c];
2781
2782 PFSYNC_BUCKET_LOCK(b);
2783 if ((b->b_flags & PFSYNCF_BUCKET_PUSH) && b->b_len > PFSYNC_MINPKT) {
2784 pfsync_sendout(0, b->b_id);
2785 b->b_flags &= ~PFSYNCF_BUCKET_PUSH;
2786 }
2787 _IF_DEQUEUE_ALL(&b->b_snd, m);
2788 PFSYNC_BUCKET_UNLOCK(b);
2789
2790 for (; m != NULL; m = n) {
2791 n = m->m_nextpkt;
2792 m->m_nextpkt = NULL;
2793
2794 pfsync_tx(sc, m);
2795 }
2796 }
2797 CURVNET_RESTORE();
2798 NET_EPOCH_EXIT(et);
2799 }
2800
2801 static int
pfsync_multicast_setup(struct pfsync_softc * sc,struct ifnet * ifp,struct in_mfilter * imf,struct in6_mfilter * im6f)2802 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp,
2803 struct in_mfilter* imf, struct in6_mfilter* im6f)
2804 {
2805 #ifdef INET
2806 struct ip_moptions *imo = &sc->sc_imo;
2807 #endif
2808 #ifdef INET6
2809 struct ip6_moptions *im6o = &sc->sc_im6o;
2810 struct sockaddr_in6 *syncpeer_sa6 = NULL;
2811 #endif
2812
2813 if (!(ifp->if_flags & IFF_MULTICAST))
2814 return (EADDRNOTAVAIL);
2815
2816 switch (sc->sc_sync_peer.ss_family) {
2817 #ifdef INET
2818 case AF_INET:
2819 {
2820 int error;
2821
2822 ip_mfilter_init(&imo->imo_head);
2823 imo->imo_multicast_vif = -1;
2824 if ((error = in_joingroup(ifp,
2825 &((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr, NULL,
2826 &imf->imf_inm)) != 0)
2827 return (error);
2828
2829 ip_mfilter_insert(&imo->imo_head, imf);
2830 imo->imo_multicast_ifp = ifp;
2831 imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2832 imo->imo_multicast_loop = 0;
2833 break;
2834 }
2835 #endif
2836 #ifdef INET6
2837 case AF_INET6:
2838 {
2839 int error;
2840
2841 syncpeer_sa6 = (struct sockaddr_in6 *)&sc->sc_sync_peer;
2842 if ((error = in6_setscope(&syncpeer_sa6->sin6_addr, ifp, NULL)))
2843 return (error);
2844
2845 ip6_mfilter_init(&im6o->im6o_head);
2846 if ((error = in6_joingroup(ifp, &syncpeer_sa6->sin6_addr, NULL,
2847 &(im6f->im6f_in6m), 0)) != 0)
2848 return (error);
2849
2850 ip6_mfilter_insert(&im6o->im6o_head, im6f);
2851 im6o->im6o_multicast_ifp = ifp;
2852 im6o->im6o_multicast_hlim = PFSYNC_DFLTTL;
2853 im6o->im6o_multicast_loop = 0;
2854 break;
2855 }
2856 #endif
2857 }
2858
2859 return (0);
2860 }
2861
2862 static void
pfsync_multicast_cleanup(struct pfsync_softc * sc)2863 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2864 {
2865 #ifdef INET
2866 struct ip_moptions *imo = &sc->sc_imo;
2867 struct in_mfilter *imf;
2868
2869 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
2870 ip_mfilter_remove(&imo->imo_head, imf);
2871 in_leavegroup(imf->imf_inm, NULL);
2872 ip_mfilter_free(imf);
2873 }
2874 imo->imo_multicast_ifp = NULL;
2875 #endif
2876
2877 #ifdef INET6
2878 struct ip6_moptions *im6o = &sc->sc_im6o;
2879 struct in6_mfilter *im6f;
2880
2881 while ((im6f = ip6_mfilter_first(&im6o->im6o_head)) != NULL) {
2882 ip6_mfilter_remove(&im6o->im6o_head, im6f);
2883 in6_leavegroup(im6f->im6f_in6m, NULL);
2884 ip6_mfilter_free(im6f);
2885 }
2886 im6o->im6o_multicast_ifp = NULL;
2887 #endif
2888 }
2889
2890 void
pfsync_detach_ifnet(struct ifnet * ifp)2891 pfsync_detach_ifnet(struct ifnet *ifp)
2892 {
2893 struct pfsync_softc *sc = V_pfsyncif;
2894
2895 if (sc == NULL)
2896 return;
2897
2898 PFSYNC_LOCK(sc);
2899
2900 if (sc->sc_sync_if == ifp) {
2901 /* We don't need mutlicast cleanup here, because the interface
2902 * is going away. We do need to ensure we don't try to do
2903 * cleanup later.
2904 */
2905 ip_mfilter_init(&sc->sc_imo.imo_head);
2906 sc->sc_imo.imo_multicast_ifp = NULL;
2907 sc->sc_im6o.im6o_multicast_ifp = NULL;
2908 sc->sc_sync_if = NULL;
2909 }
2910
2911 PFSYNC_UNLOCK(sc);
2912 }
2913
2914 static int
pfsync_pfsyncreq_to_kstatus(struct pfsyncreq * pfsyncr,struct pfsync_kstatus * status)2915 pfsync_pfsyncreq_to_kstatus(struct pfsyncreq *pfsyncr, struct pfsync_kstatus *status)
2916 {
2917 struct sockaddr_storage sa;
2918 status->maxupdates = pfsyncr->pfsyncr_maxupdates;
2919 status->flags = pfsyncr->pfsyncr_defer;
2920
2921 strlcpy(status->syncdev, pfsyncr->pfsyncr_syncdev, IFNAMSIZ);
2922
2923 memset(&sa, 0, sizeof(sa));
2924 if (pfsyncr->pfsyncr_syncpeer.s_addr != 0) {
2925 struct sockaddr_in *in = (struct sockaddr_in *)&sa;
2926 in->sin_family = AF_INET;
2927 in->sin_len = sizeof(*in);
2928 in->sin_addr.s_addr = pfsyncr->pfsyncr_syncpeer.s_addr;
2929 }
2930 status->syncpeer = sa;
2931
2932 return 0;
2933 }
2934
2935 static int
pfsync_kstatus_to_softc(struct pfsync_kstatus * status,struct pfsync_softc * sc)2936 pfsync_kstatus_to_softc(struct pfsync_kstatus *status, struct pfsync_softc *sc)
2937 {
2938 struct ifnet *sifp;
2939 struct in_mfilter *imf = NULL;
2940 struct in6_mfilter *im6f = NULL;
2941 int error;
2942 int c;
2943
2944 if ((status->maxupdates < 0) || (status->maxupdates > 255))
2945 return (EINVAL);
2946
2947 if (status->syncdev[0] == '\0')
2948 sifp = NULL;
2949 else if ((sifp = ifunit_ref(status->syncdev)) == NULL)
2950 return (EINVAL);
2951
2952 switch (status->syncpeer.ss_family) {
2953 #ifdef INET
2954 case AF_UNSPEC:
2955 case AF_INET: {
2956 struct sockaddr_in *status_sin;
2957 status_sin = (struct sockaddr_in *)&(status->syncpeer);
2958 if (sifp != NULL) {
2959 if (status_sin->sin_addr.s_addr == 0 ||
2960 status_sin->sin_addr.s_addr ==
2961 htonl(INADDR_PFSYNC_GROUP)) {
2962 status_sin->sin_family = AF_INET;
2963 status_sin->sin_len = sizeof(*status_sin);
2964 status_sin->sin_addr.s_addr =
2965 htonl(INADDR_PFSYNC_GROUP);
2966 }
2967
2968 if (IN_MULTICAST(ntohl(status_sin->sin_addr.s_addr))) {
2969 imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
2970 }
2971 }
2972 break;
2973 }
2974 #endif
2975 #ifdef INET6
2976 case AF_INET6: {
2977 struct sockaddr_in6 *status_sin6;
2978 status_sin6 = (struct sockaddr_in6*)&(status->syncpeer);
2979 if (sifp != NULL) {
2980 if (IN6_IS_ADDR_UNSPECIFIED(&status_sin6->sin6_addr) ||
2981 IN6_ARE_ADDR_EQUAL(&status_sin6->sin6_addr,
2982 &in6addr_linklocal_pfsync_group)) {
2983 status_sin6->sin6_family = AF_INET6;
2984 status_sin6->sin6_len = sizeof(*status_sin6);
2985 status_sin6->sin6_addr =
2986 in6addr_linklocal_pfsync_group;
2987 }
2988
2989 if (IN6_IS_ADDR_MULTICAST(&status_sin6->sin6_addr)) {
2990 im6f = ip6_mfilter_alloc(M_WAITOK, 0, 0);
2991 }
2992 }
2993 break;
2994 }
2995 #endif
2996 }
2997
2998 PFSYNC_LOCK(sc);
2999
3000 switch (status->version) {
3001 case PFSYNC_MSG_VERSION_UNSPECIFIED:
3002 sc->sc_version = PFSYNC_MSG_VERSION_DEFAULT;
3003 break;
3004 case PFSYNC_MSG_VERSION_1301:
3005 case PFSYNC_MSG_VERSION_1400:
3006 sc->sc_version = status->version;
3007 break;
3008 default:
3009 PFSYNC_UNLOCK(sc);
3010 return (EINVAL);
3011 }
3012
3013 switch (status->syncpeer.ss_family) {
3014 case AF_INET: {
3015 struct sockaddr_in *status_sin = (struct sockaddr_in *)&(status->syncpeer);
3016 struct sockaddr_in *sc_sin = (struct sockaddr_in *)&sc->sc_sync_peer;
3017 sc_sin->sin_family = AF_INET;
3018 sc_sin->sin_len = sizeof(*sc_sin);
3019 if (status_sin->sin_addr.s_addr == 0) {
3020 sc_sin->sin_addr.s_addr = htonl(INADDR_PFSYNC_GROUP);
3021 } else {
3022 sc_sin->sin_addr.s_addr = status_sin->sin_addr.s_addr;
3023 }
3024 break;
3025 }
3026 case AF_INET6: {
3027 struct sockaddr_in6 *status_sin = (struct sockaddr_in6 *)&(status->syncpeer);
3028 struct sockaddr_in6 *sc_sin = (struct sockaddr_in6 *)&sc->sc_sync_peer;
3029 sc_sin->sin6_family = AF_INET6;
3030 sc_sin->sin6_len = sizeof(*sc_sin);
3031 if(IN6_IS_ADDR_UNSPECIFIED(&status_sin->sin6_addr)) {
3032 sc_sin->sin6_addr = in6addr_linklocal_pfsync_group;
3033 } else {
3034 sc_sin->sin6_addr = status_sin->sin6_addr;
3035 }
3036 break;
3037 }
3038 }
3039
3040 sc->sc_maxupdates = status->maxupdates;
3041 if (status->flags & PFSYNCF_DEFER) {
3042 sc->sc_flags |= PFSYNCF_DEFER;
3043 V_pfsync_defer_ptr = pfsync_defer;
3044 } else {
3045 sc->sc_flags &= ~PFSYNCF_DEFER;
3046 V_pfsync_defer_ptr = NULL;
3047 }
3048
3049 if (sifp == NULL) {
3050 if (sc->sc_sync_if)
3051 if_rele(sc->sc_sync_if);
3052 sc->sc_sync_if = NULL;
3053 pfsync_multicast_cleanup(sc);
3054 PFSYNC_UNLOCK(sc);
3055 return (0);
3056 }
3057
3058 for (c = 0; c < pfsync_buckets; c++) {
3059 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
3060 if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT &&
3061 (sifp->if_mtu < sc->sc_ifp->if_mtu ||
3062 (sc->sc_sync_if != NULL &&
3063 sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
3064 sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
3065 pfsync_sendout(1, c);
3066 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
3067 }
3068
3069 pfsync_multicast_cleanup(sc);
3070
3071 if (((sc->sc_sync_peer.ss_family == AF_INET) &&
3072 IN_MULTICAST(ntohl(((struct sockaddr_in *)
3073 &sc->sc_sync_peer)->sin_addr.s_addr))) ||
3074 ((sc->sc_sync_peer.ss_family == AF_INET6) &&
3075 IN6_IS_ADDR_MULTICAST(&((struct sockaddr_in6*)
3076 &sc->sc_sync_peer)->sin6_addr))) {
3077 error = pfsync_multicast_setup(sc, sifp, imf, im6f);
3078 if (error) {
3079 if_rele(sifp);
3080 PFSYNC_UNLOCK(sc);
3081 #ifdef INET
3082 if (imf != NULL)
3083 ip_mfilter_free(imf);
3084 #endif
3085 #ifdef INET6
3086 if (im6f != NULL)
3087 ip6_mfilter_free(im6f);
3088 #endif
3089 return (error);
3090 }
3091 }
3092 if (sc->sc_sync_if)
3093 if_rele(sc->sc_sync_if);
3094 sc->sc_sync_if = sifp;
3095
3096 switch (sc->sc_sync_peer.ss_family) {
3097 #ifdef INET
3098 case AF_INET: {
3099 struct ip *ip;
3100 ip = &sc->sc_template.ipv4;
3101 bzero(ip, sizeof(*ip));
3102 ip->ip_v = IPVERSION;
3103 ip->ip_hl = sizeof(sc->sc_template.ipv4) >> 2;
3104 ip->ip_tos = IPTOS_LOWDELAY;
3105 /* len and id are set later. */
3106 ip->ip_off = htons(IP_DF);
3107 ip->ip_ttl = PFSYNC_DFLTTL;
3108 ip->ip_p = IPPROTO_PFSYNC;
3109 ip->ip_src.s_addr = INADDR_ANY;
3110 ip->ip_dst = ((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr;
3111 break;
3112 }
3113 #endif
3114 #ifdef INET6
3115 case AF_INET6: {
3116 struct ip6_hdr *ip6;
3117 ip6 = &sc->sc_template.ipv6;
3118 bzero(ip6, sizeof(*ip6));
3119 ip6->ip6_vfc = IPV6_VERSION;
3120 ip6->ip6_hlim = PFSYNC_DFLTTL;
3121 ip6->ip6_nxt = IPPROTO_PFSYNC;
3122 ip6->ip6_dst = ((struct sockaddr_in6 *)&sc->sc_sync_peer)->sin6_addr;
3123
3124 struct epoch_tracker et;
3125 NET_EPOCH_ENTER(et);
3126 in6_selectsrc_addr(if_getfib(sc->sc_sync_if), &ip6->ip6_dst, 0,
3127 sc->sc_sync_if, &ip6->ip6_src, NULL);
3128 NET_EPOCH_EXIT(et);
3129 break;
3130 }
3131 #endif
3132 }
3133
3134 /* Request a full state table update. */
3135 if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
3136 (*carp_demote_adj_p)(V_pfsync_carp_adj,
3137 "pfsync bulk start");
3138 sc->sc_flags &= ~PFSYNCF_OK;
3139 if (V_pf_status.debug >= PF_DEBUG_MISC)
3140 printf("pfsync: requesting bulk update\n");
3141 PFSYNC_UNLOCK(sc);
3142 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
3143 pfsync_request_update(0, 0);
3144 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
3145 PFSYNC_BLOCK(sc);
3146 sc->sc_ureq_sent = time_uptime;
3147 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail, sc);
3148 PFSYNC_BUNLOCK(sc);
3149 return (0);
3150 }
3151
3152 static void
pfsync_pointers_init(void)3153 pfsync_pointers_init(void)
3154 {
3155
3156 PF_RULES_WLOCK();
3157 V_pfsync_state_import_ptr = pfsync_state_import;
3158 V_pfsync_insert_state_ptr = pfsync_insert_state;
3159 V_pfsync_update_state_ptr = pfsync_update_state;
3160 V_pfsync_delete_state_ptr = pfsync_delete_state;
3161 V_pfsync_clear_states_ptr = pfsync_clear_states;
3162 V_pfsync_defer_ptr = pfsync_defer;
3163 PF_RULES_WUNLOCK();
3164 }
3165
3166 static void
pfsync_pointers_uninit(void)3167 pfsync_pointers_uninit(void)
3168 {
3169
3170 PF_RULES_WLOCK();
3171 V_pfsync_state_import_ptr = NULL;
3172 V_pfsync_insert_state_ptr = NULL;
3173 V_pfsync_update_state_ptr = NULL;
3174 V_pfsync_delete_state_ptr = NULL;
3175 V_pfsync_clear_states_ptr = NULL;
3176 V_pfsync_defer_ptr = NULL;
3177 PF_RULES_WUNLOCK();
3178 }
3179
3180 static void
vnet_pfsync_init(const void * unused __unused)3181 vnet_pfsync_init(const void *unused __unused)
3182 {
3183 int error;
3184
3185 V_pfsync_cloner = if_clone_simple(pfsyncname,
3186 pfsync_clone_create, pfsync_clone_destroy, 1);
3187 error = swi_add(&V_pfsync_swi_ie, pfsyncname, pfsyncintr, V_pfsyncif,
3188 SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
3189 if (error) {
3190 if_clone_detach(V_pfsync_cloner);
3191 log(LOG_INFO, "swi_add() failed in %s\n", __func__);
3192 }
3193
3194 pfsync_pointers_init();
3195 }
3196 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY,
3197 vnet_pfsync_init, NULL);
3198
3199 static void
vnet_pfsync_uninit(const void * unused __unused)3200 vnet_pfsync_uninit(const void *unused __unused)
3201 {
3202 int ret __diagused;
3203
3204 pfsync_pointers_uninit();
3205
3206 if_clone_detach(V_pfsync_cloner);
3207 ret = swi_remove(V_pfsync_swi_cookie);
3208 MPASS(ret == 0);
3209 ret = intr_event_destroy(V_pfsync_swi_ie);
3210 MPASS(ret == 0);
3211 }
3212
3213 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_FOURTH,
3214 vnet_pfsync_uninit, NULL);
3215
3216 static int
pfsync_init(void)3217 pfsync_init(void)
3218 {
3219 int error;
3220
3221 pfsync_detach_ifnet_ptr = pfsync_detach_ifnet;
3222
3223 #ifdef INET
3224 error = ipproto_register(IPPROTO_PFSYNC, pfsync_input, NULL);
3225 if (error)
3226 return (error);
3227 #endif
3228 #ifdef INET6
3229 error = ip6proto_register(IPPROTO_PFSYNC, pfsync6_input, NULL);
3230 if (error) {
3231 ipproto_unregister(IPPROTO_PFSYNC);
3232 return (error);
3233 }
3234 #endif
3235
3236 return (0);
3237 }
3238
3239 static void
pfsync_uninit(void)3240 pfsync_uninit(void)
3241 {
3242 pfsync_detach_ifnet_ptr = NULL;
3243
3244 #ifdef INET
3245 ipproto_unregister(IPPROTO_PFSYNC);
3246 #endif
3247 #ifdef INET6
3248 ip6proto_unregister(IPPROTO_PFSYNC);
3249 #endif
3250 }
3251
3252 static int
pfsync_modevent(module_t mod,int type,void * data)3253 pfsync_modevent(module_t mod, int type, void *data)
3254 {
3255 int error = 0;
3256
3257 switch (type) {
3258 case MOD_LOAD:
3259 error = pfsync_init();
3260 break;
3261 case MOD_UNLOAD:
3262 pfsync_uninit();
3263 break;
3264 default:
3265 error = EINVAL;
3266 break;
3267 }
3268
3269 return (error);
3270 }
3271
3272 static moduledata_t pfsync_mod = {
3273 pfsyncname,
3274 pfsync_modevent,
3275 0
3276 };
3277
3278 #define PFSYNC_MODVER 1
3279
3280 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */
3281 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
3282 MODULE_VERSION(pfsync, PFSYNC_MODVER);
3283 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);
3284