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