xref: /freebsd/sys/netpfil/pf/if_pfsync.c (revision 65c318630123fcf2b6f491bf4d02a5cad3031d20)
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 struct pfsync_bucket;
114 struct pfsync_softc;
115 
116 union inet_template {
117 	struct ip	ipv4;
118 	struct ip6_hdr	ipv6;
119 };
120 
121 #define PFSYNC_MINPKT ( \
122 	sizeof(union inet_template) + \
123 	sizeof(struct pfsync_header) + \
124 	sizeof(struct pfsync_subheader) )
125 
126 static int	pfsync_upd_tcp(struct pf_kstate *, struct pf_state_peer_export *,
127 		    struct pf_state_peer_export *);
128 static int	pfsync_in_clr(struct mbuf *, int, int, int, int);
129 static int	pfsync_in_ins(struct mbuf *, int, int, int, int);
130 static int	pfsync_in_iack(struct mbuf *, int, int, int, int);
131 static int	pfsync_in_upd(struct mbuf *, int, int, int, int);
132 static int	pfsync_in_upd_c(struct mbuf *, int, int, int, int);
133 static int	pfsync_in_ureq(struct mbuf *, int, int, int, int);
134 static int	pfsync_in_del_c(struct mbuf *, int, int, int, int);
135 static int	pfsync_in_bus(struct mbuf *, int, int, int, int);
136 static int	pfsync_in_tdb(struct mbuf *, int, int, int, int);
137 static int	pfsync_in_eof(struct mbuf *, int, int, int, int);
138 static int	pfsync_in_error(struct mbuf *, int, int, int, int);
139 
140 static int (*pfsync_acts[])(struct mbuf *, int, int, int, int) = {
141 	pfsync_in_clr,			/* PFSYNC_ACT_CLR */
142 	pfsync_in_ins,			/* PFSYNC_ACT_INS_1301 */
143 	pfsync_in_iack,			/* PFSYNC_ACT_INS_ACK */
144 	pfsync_in_upd,			/* PFSYNC_ACT_UPD_1301 */
145 	pfsync_in_upd_c,		/* PFSYNC_ACT_UPD_C */
146 	pfsync_in_ureq,			/* PFSYNC_ACT_UPD_REQ */
147 	pfsync_in_error,		/* PFSYNC_ACT_DEL */
148 	pfsync_in_del_c,		/* PFSYNC_ACT_DEL_C */
149 	pfsync_in_error,		/* PFSYNC_ACT_INS_F */
150 	pfsync_in_error,		/* PFSYNC_ACT_DEL_F */
151 	pfsync_in_bus,			/* PFSYNC_ACT_BUS */
152 	pfsync_in_tdb,			/* PFSYNC_ACT_TDB */
153 	pfsync_in_eof,			/* PFSYNC_ACT_EOF */
154 	pfsync_in_ins,			/* PFSYNC_ACT_INS_1400 */
155 	pfsync_in_upd,			/* PFSYNC_ACT_UPD_1400 */
156 };
157 
158 struct pfsync_q {
159 	void		(*write)(struct pf_kstate *, void *);
160 	size_t		len;
161 	u_int8_t	action;
162 };
163 
164 /* We have the following sync queues */
165 enum pfsync_q_id {
166 	PFSYNC_Q_INS_1301,
167 	PFSYNC_Q_INS_1400,
168 	PFSYNC_Q_IACK,
169 	PFSYNC_Q_UPD_1301,
170 	PFSYNC_Q_UPD_1400,
171 	PFSYNC_Q_UPD_C,
172 	PFSYNC_Q_DEL_C,
173 	PFSYNC_Q_COUNT,
174 };
175 
176 /* Functions for building messages for given queue */
177 static void	pfsync_out_state_1301(struct pf_kstate *, void *);
178 static void	pfsync_out_state_1400(struct pf_kstate *, void *);
179 static void	pfsync_out_iack(struct pf_kstate *, void *);
180 static void	pfsync_out_upd_c(struct pf_kstate *, void *);
181 static void	pfsync_out_del_c(struct pf_kstate *, void *);
182 
183 /* Attach those functions to queue */
184 static struct pfsync_q pfsync_qs[] = {
185 	{ pfsync_out_state_1301, sizeof(struct pfsync_state_1301), PFSYNC_ACT_INS_1301 },
186 	{ pfsync_out_state_1400, sizeof(struct pfsync_state_1400), PFSYNC_ACT_INS_1400 },
187 	{ pfsync_out_iack,       sizeof(struct pfsync_ins_ack),    PFSYNC_ACT_INS_ACK },
188 	{ pfsync_out_state_1301, sizeof(struct pfsync_state_1301), PFSYNC_ACT_UPD_1301 },
189 	{ pfsync_out_state_1400, sizeof(struct pfsync_state_1400), PFSYNC_ACT_UPD_1400 },
190 	{ pfsync_out_upd_c,      sizeof(struct pfsync_upd_c),      PFSYNC_ACT_UPD_C },
191 	{ pfsync_out_del_c,      sizeof(struct pfsync_del_c),      PFSYNC_ACT_DEL_C }
192 };
193 
194 /* Map queue to pf_kstate->sync_state */
195 static u_int8_t pfsync_qid_sstate[] = {
196 	PFSYNC_S_INS,   /* PFSYNC_Q_INS_1301 */
197 	PFSYNC_S_INS,   /* PFSYNC_Q_INS_1400 */
198 	PFSYNC_S_IACK,  /* PFSYNC_Q_IACK */
199 	PFSYNC_S_UPD,   /* PFSYNC_Q_UPD_1301 */
200 	PFSYNC_S_UPD,   /* PFSYNC_Q_UPD_1400 */
201 	PFSYNC_S_UPD_C, /* PFSYNC_Q_UPD_C */
202 	PFSYNC_S_DEL_C, /* PFSYNC_Q_DEL_C */
203 };
204 
205 /* Map pf_kstate->sync_state to queue */
206 static enum pfsync_q_id pfsync_sstate_to_qid(u_int8_t);
207 
208 static void	pfsync_q_ins(struct pf_kstate *, int sync_state, bool);
209 static void	pfsync_q_del(struct pf_kstate *, bool, struct pfsync_bucket *);
210 
211 static void	pfsync_update_state(struct pf_kstate *);
212 static void	pfsync_tx(struct pfsync_softc *, struct mbuf *);
213 
214 struct pfsync_upd_req_item {
215 	TAILQ_ENTRY(pfsync_upd_req_item)	ur_entry;
216 	struct pfsync_upd_req			ur_msg;
217 };
218 
219 struct pfsync_deferral {
220 	struct pfsync_softc		*pd_sc;
221 	TAILQ_ENTRY(pfsync_deferral)	pd_entry;
222 	struct callout			pd_tmo;
223 
224 	struct pf_kstate		*pd_st;
225 	struct mbuf			*pd_m;
226 };
227 
228 struct pfsync_bucket
229 {
230 	int			b_id;
231 	struct pfsync_softc	*b_sc;
232 	struct mtx		b_mtx;
233 	struct callout		b_tmo;
234 	int			b_flags;
235 #define	PFSYNCF_BUCKET_PUSH	0x00000001
236 
237 	size_t			b_len;
238 	TAILQ_HEAD(, pf_kstate)			b_qs[PFSYNC_Q_COUNT];
239 	TAILQ_HEAD(, pfsync_upd_req_item)	b_upd_req_list;
240 	TAILQ_HEAD(, pfsync_deferral)		b_deferrals;
241 	u_int			b_deferred;
242 	uint8_t			*b_plus;
243 	size_t			b_pluslen;
244 
245 	struct  ifaltq b_snd;
246 };
247 
248 struct pfsync_softc {
249 	/* Configuration */
250 	struct ifnet		*sc_ifp;
251 	struct ifnet		*sc_sync_if;
252 	struct ip_moptions	sc_imo;
253 	struct ip6_moptions	sc_im6o;
254 	struct sockaddr_storage	sc_sync_peer;
255 	uint32_t		sc_flags;
256 	uint8_t			sc_maxupdates;
257 	union inet_template     sc_template;
258 	struct mtx		sc_mtx;
259 	uint32_t		sc_version;
260 
261 	/* Queued data */
262 	struct pfsync_bucket	*sc_buckets;
263 
264 	/* Bulk update info */
265 	struct mtx		sc_bulk_mtx;
266 	uint32_t		sc_ureq_sent;
267 	int			sc_bulk_tries;
268 	uint32_t		sc_ureq_received;
269 	int			sc_bulk_hashid;
270 	uint64_t		sc_bulk_stateid;
271 	uint32_t		sc_bulk_creatorid;
272 	struct callout		sc_bulk_tmo;
273 	struct callout		sc_bulkfail_tmo;
274 };
275 
276 #define	PFSYNC_LOCK(sc)		mtx_lock(&(sc)->sc_mtx)
277 #define	PFSYNC_UNLOCK(sc)	mtx_unlock(&(sc)->sc_mtx)
278 #define	PFSYNC_LOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_mtx, MA_OWNED)
279 
280 #define PFSYNC_BUCKET_LOCK(b)		mtx_lock(&(b)->b_mtx)
281 #define PFSYNC_BUCKET_UNLOCK(b)		mtx_unlock(&(b)->b_mtx)
282 #define PFSYNC_BUCKET_LOCK_ASSERT(b)	mtx_assert(&(b)->b_mtx, MA_OWNED)
283 
284 #define	PFSYNC_BLOCK(sc)	mtx_lock(&(sc)->sc_bulk_mtx)
285 #define	PFSYNC_BUNLOCK(sc)	mtx_unlock(&(sc)->sc_bulk_mtx)
286 #define	PFSYNC_BLOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED)
287 
288 #define PFSYNC_DEFER_TIMEOUT	20
289 
290 static const char pfsyncname[] = "pfsync";
291 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data");
292 VNET_DEFINE_STATIC(struct pfsync_softc	*, pfsyncif) = NULL;
293 #define	V_pfsyncif		VNET(pfsyncif)
294 VNET_DEFINE_STATIC(void *, pfsync_swi_cookie) = NULL;
295 #define	V_pfsync_swi_cookie	VNET(pfsync_swi_cookie)
296 VNET_DEFINE_STATIC(struct intr_event *, pfsync_swi_ie);
297 #define	V_pfsync_swi_ie		VNET(pfsync_swi_ie)
298 VNET_DEFINE_STATIC(struct pfsyncstats, pfsyncstats);
299 #define	V_pfsyncstats		VNET(pfsyncstats)
300 VNET_DEFINE_STATIC(int, pfsync_carp_adj) = CARP_MAXSKEW;
301 #define	V_pfsync_carp_adj	VNET(pfsync_carp_adj)
302 VNET_DEFINE_STATIC(unsigned int, pfsync_defer_timeout) = PFSYNC_DEFER_TIMEOUT;
303 #define	V_pfsync_defer_timeout	VNET(pfsync_defer_timeout)
304 
305 static void	pfsync_timeout(void *);
306 static void	pfsync_push(struct pfsync_bucket *);
307 static void	pfsync_push_all(struct pfsync_softc *);
308 static void	pfsyncintr(void *);
309 static int	pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *,
310 		    struct in_mfilter *, struct in6_mfilter *);
311 static void	pfsync_multicast_cleanup(struct pfsync_softc *);
312 static void	pfsync_pointers_init(void);
313 static void	pfsync_pointers_uninit(void);
314 static int	pfsync_init(void);
315 static void	pfsync_uninit(void);
316 
317 static unsigned long pfsync_buckets;
318 
319 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
320     "PFSYNC");
321 SYSCTL_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_VNET | CTLFLAG_RW,
322     &VNET_NAME(pfsyncstats), pfsyncstats,
323     "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)");
324 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_VNET | CTLFLAG_RW,
325     &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment");
326 SYSCTL_ULONG(_net_pfsync, OID_AUTO, pfsync_buckets, CTLFLAG_RDTUN,
327     &pfsync_buckets, 0, "Number of pfsync hash buckets");
328 SYSCTL_UINT(_net_pfsync, OID_AUTO, defer_delay, CTLFLAG_VNET | CTLFLAG_RW,
329     &VNET_NAME(pfsync_defer_timeout), 0, "Deferred packet timeout (in ms)");
330 
331 static int	pfsync_clone_create(struct if_clone *, int, caddr_t);
332 static void	pfsync_clone_destroy(struct ifnet *);
333 static int	pfsync_alloc_scrub_memory(struct pf_state_peer_export *,
334 		    struct pf_state_peer *);
335 static int	pfsyncoutput(struct ifnet *, struct mbuf *,
336 		    const struct sockaddr *, struct route *);
337 static int	pfsyncioctl(struct ifnet *, u_long, caddr_t);
338 
339 static int	pfsync_defer(struct pf_kstate *, struct mbuf *);
340 static void	pfsync_undefer(struct pfsync_deferral *, int);
341 static void	pfsync_undefer_state_locked(struct pf_kstate *, int);
342 static void	pfsync_undefer_state(struct pf_kstate *, int);
343 static void	pfsync_defer_tmo(void *);
344 
345 static void	pfsync_request_update(u_int32_t, u_int64_t);
346 static bool	pfsync_update_state_req(struct pf_kstate *);
347 
348 static void	pfsync_drop_all(struct pfsync_softc *);
349 static void	pfsync_drop(struct pfsync_softc *, int);
350 static void	pfsync_sendout(int, int);
351 static void	pfsync_send_plus(void *, size_t);
352 
353 static void	pfsync_bulk_start(void);
354 static void	pfsync_bulk_status(u_int8_t);
355 static void	pfsync_bulk_update(void *);
356 static void	pfsync_bulk_fail(void *);
357 
358 static void	pfsync_detach_ifnet(struct ifnet *);
359 
360 static int pfsync_pfsyncreq_to_kstatus(struct pfsyncreq *,
361     struct pfsync_kstatus *);
362 static int pfsync_kstatus_to_softc(struct pfsync_kstatus *,
363     struct pfsync_softc *);
364 
365 #ifdef IPSEC
366 static void	pfsync_update_net_tdb(struct pfsync_tdb *);
367 #endif
368 static struct pfsync_bucket	*pfsync_get_bucket(struct pfsync_softc *,
369 		    struct pf_kstate *);
370 
371 #define PFSYNC_MAX_BULKTRIES	12
372 
373 VNET_DEFINE(struct if_clone *, pfsync_cloner);
374 #define	V_pfsync_cloner	VNET(pfsync_cloner)
375 
376 const struct in6_addr in6addr_linklocal_pfsync_group =
377 	{{{ 0xff, 0x12, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
378 	    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xf0 }}};
379 static int
pfsync_clone_create(struct if_clone * ifc,int unit,caddr_t param)380 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param)
381 {
382 	struct pfsync_softc *sc;
383 	struct ifnet *ifp;
384 	struct pfsync_bucket *b;
385 	int c;
386 	enum pfsync_q_id q;
387 
388 	if (unit != 0)
389 		return (EINVAL);
390 
391 	if (! pfsync_buckets)
392 		pfsync_buckets = mp_ncpus * 2;
393 
394 	sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO);
395 	sc->sc_flags |= PFSYNCF_OK;
396 	sc->sc_maxupdates = 128;
397 	sc->sc_version = PFSYNC_MSG_VERSION_DEFAULT;
398 	sc->sc_buckets = mallocarray(pfsync_buckets, sizeof(*sc->sc_buckets),
399 	    M_PFSYNC, M_ZERO | M_WAITOK);
400 	for (c = 0; c < pfsync_buckets; c++) {
401 		b = &sc->sc_buckets[c];
402 		mtx_init(&b->b_mtx, "pfsync bucket", NULL, MTX_DEF);
403 
404 		b->b_id = c;
405 		b->b_sc = sc;
406 		b->b_len = PFSYNC_MINPKT;
407 
408 		for (q = 0; q < PFSYNC_Q_COUNT; q++)
409 			TAILQ_INIT(&b->b_qs[q]);
410 
411 		TAILQ_INIT(&b->b_upd_req_list);
412 		TAILQ_INIT(&b->b_deferrals);
413 
414 		callout_init(&b->b_tmo, 1);
415 
416 		b->b_snd.ifq_maxlen = ifqmaxlen;
417 	}
418 
419 	ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC);
420 	if_initname(ifp, pfsyncname, unit);
421 	ifp->if_softc = sc;
422 	ifp->if_ioctl = pfsyncioctl;
423 	ifp->if_output = pfsyncoutput;
424 	ifp->if_hdrlen = sizeof(struct pfsync_header);
425 	ifp->if_mtu = ETHERMTU;
426 	mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF);
427 	mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF);
428 	callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0);
429 	callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0);
430 
431 	if_attach(ifp);
432 
433 	bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
434 
435 	V_pfsyncif = sc;
436 
437 	return (0);
438 }
439 
440 static void
pfsync_clone_destroy(struct ifnet * ifp)441 pfsync_clone_destroy(struct ifnet *ifp)
442 {
443 	struct pfsync_softc *sc = ifp->if_softc;
444 	struct pfsync_bucket *b;
445 	int c, ret;
446 
447 	for (c = 0; c < pfsync_buckets; c++) {
448 		b = &sc->sc_buckets[c];
449 		/*
450 		 * At this stage, everything should have already been
451 		 * cleared by pfsync_uninit(), and we have only to
452 		 * drain callouts.
453 		 */
454 		PFSYNC_BUCKET_LOCK(b);
455 		while (b->b_deferred > 0) {
456 			struct pfsync_deferral *pd =
457 			    TAILQ_FIRST(&b->b_deferrals);
458 
459 			ret = callout_stop(&pd->pd_tmo);
460 			PFSYNC_BUCKET_UNLOCK(b);
461 			if (ret > 0) {
462 				pfsync_undefer(pd, 1);
463 			} else {
464 				callout_drain(&pd->pd_tmo);
465 			}
466 			PFSYNC_BUCKET_LOCK(b);
467 		}
468 		MPASS(b->b_deferred == 0);
469 		MPASS(TAILQ_EMPTY(&b->b_deferrals));
470 		PFSYNC_BUCKET_UNLOCK(b);
471 
472 		free(b->b_plus, M_PFSYNC);
473 		b->b_plus = NULL;
474 		b->b_pluslen = 0;
475 
476 		callout_drain(&b->b_tmo);
477 	}
478 
479 	callout_drain(&sc->sc_bulkfail_tmo);
480 	callout_drain(&sc->sc_bulk_tmo);
481 
482 	if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
483 		(*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
484 	bpfdetach(ifp);
485 	if_detach(ifp);
486 
487 	pfsync_drop_all(sc);
488 
489 	if_free(ifp);
490 	pfsync_multicast_cleanup(sc);
491 	mtx_destroy(&sc->sc_mtx);
492 	mtx_destroy(&sc->sc_bulk_mtx);
493 
494 	for (c = 0; c < pfsync_buckets; c++) {
495 		b = &sc->sc_buckets[c];
496 		mtx_destroy(&b->b_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 pf_state_peer_export * s,struct pf_state_peer * d)505 pfsync_alloc_scrub_memory(struct pf_state_peer_export *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 	sa_family_t		 rt_af = 0;
533 	uint8_t			 rt = 0;
534 	int			 n = 0;
535 
536 	PF_RULES_RASSERT();
537 
538 	if (sp->pfs_1301.creatorid == 0) {
539 		if (V_pf_status.debug >= PF_DEBUG_MISC)
540 			printf("%s: invalid creator id: %08x\n", __func__,
541 			    ntohl(sp->pfs_1301.creatorid));
542 		return (EINVAL);
543 	}
544 
545 	if ((kif = pfi_kkif_find(sp->pfs_1301.ifname)) == NULL) {
546 		if (V_pf_status.debug >= PF_DEBUG_MISC)
547 			printf("%s: unknown interface: %s\n", __func__,
548 			    sp->pfs_1301.ifname);
549 		if (flags & PFSYNC_SI_IOCTL)
550 			return (EINVAL);
551 		return (0);	/* skip this state */
552 	}
553 
554 	/*
555 	 * If the ruleset checksums match or the state is coming from the ioctl,
556 	 * it's safe to associate the state with the rule of that number.
557 	 */
558 	if (sp->pfs_1301.rule != htonl(-1) && sp->pfs_1301.anchor == htonl(-1) &&
559 	    (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->pfs_1301.rule) <
560 	    pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) {
561 		TAILQ_FOREACH(r, pf_main_ruleset.rules[
562 		    PF_RULESET_FILTER].active.ptr, entries)
563 			if (ntohl(sp->pfs_1301.rule) == n++)
564 				break;
565 	} else
566 		r = &V_pf_default_rule;
567 
568 	/*
569 	 * Check routing interface early on. Do it before allocating memory etc.
570 	 * because there is a high chance there will be a lot more such states.
571 	 */
572 	switch (msg_version) {
573 	case PFSYNC_MSG_VERSION_1301:
574 		/*
575 		 * On FreeBSD <= 13 the routing interface and routing operation
576 		 * are not sent over pfsync. If the ruleset is identical,
577 		 * though, we might be able to recover the routing information
578 		 * from the local ruleset.
579 		 */
580 		if (r != &V_pf_default_rule) {
581 			struct pf_kpool		*pool = &r->route;
582 
583 			/* Backwards compatibility. */
584 			if (TAILQ_EMPTY(&pool->list))
585 				pool = &r->rdr;
586 
587 			/*
588 			 * The ruleset is identical, try to recover. If the rule
589 			 * has a redirection pool with a single interface, there
590 			 * is a chance that this interface is identical as on
591 			 * the pfsync peer. If there's more than one interface,
592 			 * give up, as we can't be sure that we will pick the
593 			 * same one as the pfsync peer did.
594 			 */
595 			rpool_first = TAILQ_FIRST(&(pool->list));
596 			if ((rpool_first == NULL) ||
597 			    (TAILQ_NEXT(rpool_first, entries) != NULL)) {
598 				DPFPRINTF(PF_DEBUG_MISC,
599 				    "%s: can't recover routing information "
600 				    "because of empty or bad redirection pool",
601 				    __func__);
602 				return ((flags & PFSYNC_SI_IOCTL) ? EINVAL : 0);
603 			}
604 			rt = r->rt;
605 			rt_kif = rpool_first->kif;
606 			/*
607 			 * Guess the AF of the route address, FreeBSD 13 does
608 			 * not support af-to nor prefer-ipv6-nexthop
609 			 * so it should be safe.
610 			 */
611 			rt_af = r->af;
612 		} else if (!PF_AZERO(&sp->pfs_1301.rt_addr, sp->pfs_1301.af)) {
613 			/*
614 			 * Ruleset different, routing *supposedly* requested,
615 			 * give up on recovering.
616 			 */
617 			DPFPRINTF(PF_DEBUG_MISC,
618 			    "%s: can't recover routing information "
619 			    "because of different ruleset", __func__);
620 			return ((flags & PFSYNC_SI_IOCTL) ? EINVAL : 0);
621 		}
622 	break;
623 	case PFSYNC_MSG_VERSION_1400:
624 		/*
625 		 * On FreeBSD 14 and above we're not taking any chances.
626 		 * We use the information synced to us.
627 		 */
628 		if (sp->pfs_1400.rt) {
629 			rt_kif = pfi_kkif_find(sp->pfs_1400.rt_ifname);
630 			if (rt_kif == NULL) {
631 				DPFPRINTF(PF_DEBUG_MISC,
632 				    "%s: unknown route interface: %s",
633 				    __func__, sp->pfs_1400.rt_ifname);
634 				return ((flags & PFSYNC_SI_IOCTL) ? EINVAL : 0);
635 			}
636 			rt = sp->pfs_1400.rt;
637 			/*
638 			 * Guess the AF of the route address, FreeBSD 14 does
639 			 * not support af-to nor prefer-ipv6-nexthop
640 			 * so it should be safe.
641 			 */
642 			rt_af = sp->pfs_1400.af;
643 		}
644 	break;
645 	}
646 
647 	if ((r->max_states &&
648 	    counter_u64_fetch(r->states_cur) >= r->max_states))
649 		goto cleanup;
650 
651 	/*
652 	 * XXXGL: consider M_WAITOK in ioctl path after.
653 	 */
654 	st = pf_alloc_state(M_NOWAIT);
655 	if (__predict_false(st == NULL))
656 		goto cleanup;
657 
658 	if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
659 		goto cleanup;
660 
661 #ifndef	__NO_STRICT_ALIGNMENT
662 	bcopy(&sp->pfs_1301.key, key, sizeof(struct pfsync_state_key) * 2);
663 	kw = &key[PF_SK_WIRE];
664 	ks = &key[PF_SK_STACK];
665 #else
666 	kw = &sp->pfs_1301.key[PF_SK_WIRE];
667 	ks = &sp->pfs_1301.key[PF_SK_STACK];
668 #endif
669 
670 	if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->pfs_1301.af) ||
671 	    PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->pfs_1301.af) ||
672 	    kw->port[0] != ks->port[0] ||
673 	    kw->port[1] != ks->port[1]) {
674 		sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
675 		if (sks == NULL)
676 			goto cleanup;
677 	} else
678 		sks = skw;
679 
680 	/* allocate memory for scrub info */
681 	if (pfsync_alloc_scrub_memory(&sp->pfs_1301.src, &st->src) ||
682 	    pfsync_alloc_scrub_memory(&sp->pfs_1301.dst, &st->dst))
683 		goto cleanup;
684 
685 	/* Copy to state key(s). */
686 	skw->addr[0] = kw->addr[0];
687 	skw->addr[1] = kw->addr[1];
688 	skw->port[0] = kw->port[0];
689 	skw->port[1] = kw->port[1];
690 	skw->proto = sp->pfs_1301.proto;
691 	skw->af = sp->pfs_1301.af;
692 	if (sks != skw) {
693 		sks->addr[0] = ks->addr[0];
694 		sks->addr[1] = ks->addr[1];
695 		sks->port[0] = ks->port[0];
696 		sks->port[1] = ks->port[1];
697 		sks->proto = sp->pfs_1301.proto;
698 		sks->af = sp->pfs_1301.af;
699 	}
700 
701 	/* copy to state */
702 	bcopy(&sp->pfs_1301.rt_addr, &st->act.rt_addr, sizeof(st->act.rt_addr));
703 	st->creation = (time_uptime - ntohl(sp->pfs_1301.creation)) * 1000;
704 	st->expire = pf_get_uptime();
705 	if (sp->pfs_1301.expire) {
706 		uint32_t timeout;
707 
708 		timeout = r->timeout[sp->pfs_1301.timeout];
709 		if (!timeout)
710 			timeout = V_pf_default_rule.timeout[sp->pfs_1301.timeout];
711 
712 		/* sp->expire may have been adaptively scaled by export. */
713 		st->expire -= (timeout - ntohl(sp->pfs_1301.expire)) * 1000;
714 	}
715 
716 	st->direction = sp->pfs_1301.direction;
717 	st->act.log = sp->pfs_1301.log;
718 	st->timeout = sp->pfs_1301.timeout;
719 
720 	st->act.rt = rt;
721 	st->act.rt_kif = rt_kif;
722 	st->act.rt_af = rt_af;
723 
724 	switch (msg_version) {
725 		case PFSYNC_MSG_VERSION_1301:
726 			st->state_flags = sp->pfs_1301.state_flags;
727 			/*
728 			 * In FreeBSD 13 pfsync lacks many attributes. Copy them
729 			 * from the rule if possible. If rule can't be matched
730 			 * clear any set options as we can't recover their
731 			 * parameters.
732 			*/
733 			if (r == &V_pf_default_rule) {
734 				st->state_flags &= ~PFSTATE_SETMASK;
735 			} else {
736 				/*
737 				 * Similar to pf_rule_to_actions(). This code
738 				 * won't set the actions properly if they come
739 				 * from multiple "match" rules as only rule
740 				 * creating the state is send over pfsync.
741 				 */
742 				st->act.qid = r->qid;
743 				st->act.pqid = r->pqid;
744 				st->act.rtableid = r->rtableid;
745 				if (r->scrub_flags & PFSTATE_SETTOS)
746 					st->act.set_tos = r->set_tos;
747 				st->act.min_ttl = r->min_ttl;
748 				st->act.max_mss = r->max_mss;
749 				st->state_flags |= (r->scrub_flags &
750 				    (PFSTATE_NODF|PFSTATE_RANDOMID|
751 				    PFSTATE_SETTOS|PFSTATE_SCRUB_TCP|
752 				    PFSTATE_SETPRIO));
753 				if (r->dnpipe || r->dnrpipe) {
754 					if (r->free_flags & PFRULE_DN_IS_PIPE)
755 						st->state_flags |= PFSTATE_DN_IS_PIPE;
756 					else
757 						st->state_flags &= ~PFSTATE_DN_IS_PIPE;
758 				}
759 				st->act.dnpipe = r->dnpipe;
760 				st->act.dnrpipe = r->dnrpipe;
761 			}
762 			break;
763 		case PFSYNC_MSG_VERSION_1400:
764 			st->state_flags = ntohs(sp->pfs_1400.state_flags);
765 			st->act.qid = ntohs(sp->pfs_1400.qid);
766 			st->act.pqid = ntohs(sp->pfs_1400.pqid);
767 			st->act.dnpipe = ntohs(sp->pfs_1400.dnpipe);
768 			st->act.dnrpipe = ntohs(sp->pfs_1400.dnrpipe);
769 			st->act.rtableid = ntohl(sp->pfs_1400.rtableid);
770 			st->act.min_ttl = sp->pfs_1400.min_ttl;
771 			st->act.set_tos = sp->pfs_1400.set_tos;
772 			st->act.max_mss = ntohs(sp->pfs_1400.max_mss);
773 			st->act.set_prio[0] = sp->pfs_1400.set_prio[0];
774 			st->act.set_prio[1] = sp->pfs_1400.set_prio[1];
775 			break;
776 		default:
777 			panic("%s: Unsupported pfsync_msg_version %d",
778 			    __func__, msg_version);
779 	}
780 
781 	if (! (st->act.rtableid == -1 ||
782 	    (st->act.rtableid >= 0 && st->act.rtableid < rt_numfibs)))
783 		goto cleanup;
784 
785 	st->id = sp->pfs_1301.id;
786 	st->creatorid = sp->pfs_1301.creatorid;
787 	pf_state_peer_ntoh(&sp->pfs_1301.src, &st->src);
788 	pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
789 
790 	st->rule = r;
791 	st->nat_rule = NULL;
792 	st->anchor = NULL;
793 
794 	st->pfsync_time = time_uptime;
795 	st->sync_state = PFSYNC_S_NONE;
796 
797 	if (!(flags & PFSYNC_SI_IOCTL))
798 		st->state_flags |= PFSTATE_NOSYNC;
799 
800 	if ((error = pf_state_insert(kif, kif, skw, sks, st)) != 0)
801 		goto cleanup_state;
802 
803 	/* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
804 	counter_u64_add(r->states_cur, 1);
805 	counter_u64_add(r->states_tot, 1);
806 
807 	if (!(flags & PFSYNC_SI_IOCTL)) {
808 		st->state_flags &= ~PFSTATE_NOSYNC;
809 		if (st->state_flags & PFSTATE_ACK) {
810 			struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
811 			PFSYNC_BUCKET_LOCK(b);
812 			pfsync_q_ins(st, PFSYNC_S_IACK, true);
813 			PFSYNC_BUCKET_UNLOCK(b);
814 
815 			pfsync_push_all(sc);
816 		}
817 	}
818 	st->state_flags &= ~PFSTATE_ACK;
819 	PF_STATE_UNLOCK(st);
820 
821 	return (0);
822 
823 cleanup:
824 	error = ENOMEM;
825 
826 	if (skw == sks)
827 		sks = NULL;
828 	uma_zfree(V_pf_state_key_z, skw);
829 	uma_zfree(V_pf_state_key_z, sks);
830 
831 cleanup_state:	/* pf_state_insert() frees the state keys. */
832 	if (st) {
833 		st->timeout = PFTM_UNLINKED; /* appease an assert */
834 		pf_free_state(st);
835 	}
836 	return (error);
837 }
838 
839 #ifdef INET
840 static int
pfsync_input(struct mbuf ** mp,int * offp __unused,int proto __unused)841 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused)
842 {
843 	struct pfsync_softc *sc = V_pfsyncif;
844 	struct mbuf *m = *mp;
845 	struct ip *ip = mtod(m, struct ip *);
846 	struct pfsync_header *ph;
847 	struct pfsync_subheader subh;
848 
849 	int offset, len, flags = 0;
850 	int rv;
851 	uint16_t count;
852 
853 	PF_RULES_RLOCK_TRACKER;
854 
855 	*mp = NULL;
856 	V_pfsyncstats.pfsyncs_ipackets++;
857 
858 	/* Verify that we have a sync interface configured. */
859 	if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
860 	    (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
861 		goto done;
862 
863 	/* verify that the packet came in on the right interface */
864 	if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
865 		V_pfsyncstats.pfsyncs_badif++;
866 		goto done;
867 	}
868 
869 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
870 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
871 	/* verify that the IP TTL is 255. */
872 	if (ip->ip_ttl != PFSYNC_DFLTTL) {
873 		V_pfsyncstats.pfsyncs_badttl++;
874 		goto done;
875 	}
876 
877 	offset = ip->ip_hl << 2;
878 	if (m->m_pkthdr.len < offset + sizeof(*ph)) {
879 		V_pfsyncstats.pfsyncs_hdrops++;
880 		goto done;
881 	}
882 
883 	if (offset + sizeof(*ph) > m->m_len) {
884 		if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
885 			V_pfsyncstats.pfsyncs_hdrops++;
886 			return (IPPROTO_DONE);
887 		}
888 		ip = mtod(m, struct ip *);
889 	}
890 	ph = (struct pfsync_header *)((char *)ip + offset);
891 
892 	/* verify the version */
893 	if (ph->version != PFSYNC_VERSION) {
894 		V_pfsyncstats.pfsyncs_badver++;
895 		goto done;
896 	}
897 
898 	len = ntohs(ph->len) + offset;
899 	if (m->m_pkthdr.len < len) {
900 		V_pfsyncstats.pfsyncs_badlen++;
901 		goto done;
902 	}
903 
904 	/*
905 	 * Trusting pf_chksum during packet processing, as well as seeking
906 	 * in interface name tree, require holding PF_RULES_RLOCK().
907 	 */
908 	PF_RULES_RLOCK();
909 	if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
910 		flags = PFSYNC_SI_CKSUM;
911 
912 	offset += sizeof(*ph);
913 	while (offset <= len - sizeof(subh)) {
914 		m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
915 		offset += sizeof(subh);
916 
917 		if (subh.action >= PFSYNC_ACT_MAX) {
918 			V_pfsyncstats.pfsyncs_badact++;
919 			PF_RULES_RUNLOCK();
920 			goto done;
921 		}
922 
923 		count = ntohs(subh.count);
924 		V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
925 		rv = (*pfsync_acts[subh.action])(m, offset, count, flags, subh.action);
926 		if (rv == -1) {
927 			PF_RULES_RUNLOCK();
928 			return (IPPROTO_DONE);
929 		}
930 
931 		offset += rv;
932 	}
933 	PF_RULES_RUNLOCK();
934 
935 done:
936 	m_freem(m);
937 	return (IPPROTO_DONE);
938 }
939 #endif
940 
941 #ifdef INET6
942 static int
pfsync6_input(struct mbuf ** mp,int * offp __unused,int proto __unused)943 pfsync6_input(struct mbuf **mp, int *offp __unused, int proto __unused)
944 {
945 	struct pfsync_softc *sc = V_pfsyncif;
946 	struct mbuf *m = *mp;
947 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
948 	struct pfsync_header *ph;
949 	struct pfsync_subheader subh;
950 
951 	int offset, len, flags = 0;
952 	int rv;
953 	uint16_t count;
954 
955 	PF_RULES_RLOCK_TRACKER;
956 
957 	*mp = NULL;
958 	V_pfsyncstats.pfsyncs_ipackets++;
959 
960 	/* Verify that we have a sync interface configured. */
961 	if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
962 	    (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
963 		goto done;
964 
965 	/* verify that the packet came in on the right interface */
966 	if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
967 		V_pfsyncstats.pfsyncs_badif++;
968 		goto done;
969 	}
970 
971 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
972 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
973 	/* verify that the IP TTL is 255. */
974 	if (ip6->ip6_hlim != PFSYNC_DFLTTL) {
975 		V_pfsyncstats.pfsyncs_badttl++;
976 		goto done;
977 	}
978 
979 
980 	offset = sizeof(*ip6);
981 	if (m->m_pkthdr.len < offset + sizeof(*ph)) {
982 		V_pfsyncstats.pfsyncs_hdrops++;
983 		goto done;
984 	}
985 
986 	if (offset + sizeof(*ph) > m->m_len) {
987 		if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
988 			V_pfsyncstats.pfsyncs_hdrops++;
989 			return (IPPROTO_DONE);
990 		}
991 		ip6 = mtod(m, struct ip6_hdr *);
992 	}
993 	ph = (struct pfsync_header *)((char *)ip6 + offset);
994 
995 	/* verify the version */
996 	if (ph->version != PFSYNC_VERSION) {
997 		V_pfsyncstats.pfsyncs_badver++;
998 		goto done;
999 	}
1000 
1001 	len = ntohs(ph->len) + offset;
1002 	if (m->m_pkthdr.len < len) {
1003 		V_pfsyncstats.pfsyncs_badlen++;
1004 		goto done;
1005 	}
1006 
1007 	/*
1008 	 * Trusting pf_chksum during packet processing, as well as seeking
1009 	 * in interface name tree, require holding PF_RULES_RLOCK().
1010 	 */
1011 	PF_RULES_RLOCK();
1012 	if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
1013 		flags = PFSYNC_SI_CKSUM;
1014 
1015 	offset += sizeof(*ph);
1016 	while (offset <= len - sizeof(subh)) {
1017 		m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
1018 		offset += sizeof(subh);
1019 
1020 		if (subh.action >= PFSYNC_ACT_MAX) {
1021 			V_pfsyncstats.pfsyncs_badact++;
1022 			PF_RULES_RUNLOCK();
1023 			goto done;
1024 		}
1025 
1026 		count = ntohs(subh.count);
1027 		V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
1028 		rv = (*pfsync_acts[subh.action])(m, offset, count, flags, subh.action);
1029 		if (rv == -1) {
1030 			PF_RULES_RUNLOCK();
1031 			return (IPPROTO_DONE);
1032 		}
1033 
1034 		offset += rv;
1035 	}
1036 	PF_RULES_RUNLOCK();
1037 
1038 done:
1039 	m_freem(m);
1040 	return (IPPROTO_DONE);
1041 }
1042 #endif
1043 
1044 static int
pfsync_in_clr(struct mbuf * m,int offset,int count,int flags,int action)1045 pfsync_in_clr(struct mbuf *m, int offset, int count, int flags, int action)
1046 {
1047 	struct pfsync_clr *clr;
1048 	struct mbuf *mp;
1049 	int len = sizeof(*clr) * count;
1050 	int i, offp;
1051 	u_int32_t creatorid;
1052 
1053 	mp = m_pulldown(m, offset, len, &offp);
1054 	if (mp == NULL) {
1055 		V_pfsyncstats.pfsyncs_badlen++;
1056 		return (-1);
1057 	}
1058 	clr = (struct pfsync_clr *)(mp->m_data + offp);
1059 
1060 	for (i = 0; i < count; i++) {
1061 		creatorid = clr[i].creatorid;
1062 
1063 		if (clr[i].ifname[0] != '\0' &&
1064 		    pfi_kkif_find(clr[i].ifname) == NULL)
1065 			continue;
1066 
1067 		for (int i = 0; i <= V_pf_hashmask; i++) {
1068 			struct pf_idhash *ih = &V_pf_idhash[i];
1069 			struct pf_kstate *s;
1070 relock:
1071 			PF_HASHROW_LOCK(ih);
1072 			LIST_FOREACH(s, &ih->states, entry) {
1073 				if (s->creatorid == creatorid) {
1074 					s->state_flags |= PFSTATE_NOSYNC;
1075 					pf_remove_state(s);
1076 					goto relock;
1077 				}
1078 			}
1079 			PF_HASHROW_UNLOCK(ih);
1080 		}
1081 	}
1082 
1083 	return (len);
1084 }
1085 
1086 static int
pfsync_in_ins(struct mbuf * m,int offset,int count,int flags,int action)1087 pfsync_in_ins(struct mbuf *m, int offset, int count, int flags, int action)
1088 {
1089 	struct mbuf *mp;
1090 	union pfsync_state_union *sa, *sp;
1091 	int i, offp, total_len, msg_version, msg_len;
1092 
1093 	switch (action) {
1094 		case PFSYNC_ACT_INS_1301:
1095 			msg_len = sizeof(struct pfsync_state_1301);
1096 			total_len = msg_len * count;
1097 			msg_version = PFSYNC_MSG_VERSION_1301;
1098 			break;
1099 		case PFSYNC_ACT_INS_1400:
1100 			msg_len = sizeof(struct pfsync_state_1400);
1101 			total_len = msg_len * count;
1102 			msg_version = PFSYNC_MSG_VERSION_1400;
1103 			break;
1104 		default:
1105 			V_pfsyncstats.pfsyncs_badver++;
1106 			return (-1);
1107 	}
1108 
1109 	mp = m_pulldown(m, offset, total_len, &offp);
1110 	if (mp == NULL) {
1111 		V_pfsyncstats.pfsyncs_badlen++;
1112 		return (-1);
1113 	}
1114 	sa = (union pfsync_state_union *)(mp->m_data + offp);
1115 
1116 	for (i = 0; i < count; i++) {
1117 		sp = (union pfsync_state_union *)((char *)sa + msg_len * i);
1118 
1119 		/* Check for invalid values. */
1120 		if (sp->pfs_1301.timeout >= PFTM_MAX ||
1121 		    sp->pfs_1301.src.state > PF_TCPS_PROXY_DST ||
1122 		    sp->pfs_1301.dst.state > PF_TCPS_PROXY_DST ||
1123 		    sp->pfs_1301.direction > PF_OUT ||
1124 		    (sp->pfs_1301.af != AF_INET &&
1125 		    sp->pfs_1301.af != AF_INET6)) {
1126 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1127 				printf("%s: invalid value\n", __func__);
1128 			V_pfsyncstats.pfsyncs_badval++;
1129 			continue;
1130 		}
1131 
1132 		if (pfsync_state_import(sp, flags, msg_version) != 0)
1133 			V_pfsyncstats.pfsyncs_badact++;
1134 	}
1135 
1136 	return (total_len);
1137 }
1138 
1139 static int
pfsync_in_iack(struct mbuf * m,int offset,int count,int flags,int action)1140 pfsync_in_iack(struct mbuf *m, int offset, int count, int flags, int action)
1141 {
1142 	struct pfsync_ins_ack *ia, *iaa;
1143 	struct pf_kstate *st;
1144 
1145 	struct mbuf *mp;
1146 	int len = count * sizeof(*ia);
1147 	int offp, i;
1148 
1149 	mp = m_pulldown(m, offset, len, &offp);
1150 	if (mp == NULL) {
1151 		V_pfsyncstats.pfsyncs_badlen++;
1152 		return (-1);
1153 	}
1154 	iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
1155 
1156 	for (i = 0; i < count; i++) {
1157 		ia = &iaa[i];
1158 
1159 		st = pf_find_state_byid(ia->id, ia->creatorid);
1160 		if (st == NULL)
1161 			continue;
1162 
1163 		if (st->state_flags & PFSTATE_ACK) {
1164 			pfsync_undefer_state(st, 0);
1165 		}
1166 		PF_STATE_UNLOCK(st);
1167 	}
1168 	/*
1169 	 * XXX this is not yet implemented, but we know the size of the
1170 	 * message so we can skip it.
1171 	 */
1172 
1173 	return (count * sizeof(struct pfsync_ins_ack));
1174 }
1175 
1176 static int
pfsync_upd_tcp(struct pf_kstate * st,struct pf_state_peer_export * src,struct pf_state_peer_export * dst)1177 pfsync_upd_tcp(struct pf_kstate *st, struct pf_state_peer_export *src,
1178     struct pf_state_peer_export *dst)
1179 {
1180 	int sync = 0;
1181 
1182 	PF_STATE_LOCK_ASSERT(st);
1183 
1184 	/*
1185 	 * The state should never go backwards except
1186 	 * for syn-proxy states.  Neither should the
1187 	 * sequence window slide backwards.
1188 	 */
1189 	if ((st->src.state > src->state &&
1190 	    (st->src.state < PF_TCPS_PROXY_SRC ||
1191 	    src->state >= PF_TCPS_PROXY_SRC)) ||
1192 
1193 	    (st->src.state == src->state &&
1194 	    SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
1195 		sync++;
1196 	else
1197 		pf_state_peer_ntoh(src, &st->src);
1198 
1199 	if ((st->dst.state > dst->state) ||
1200 
1201 	    (st->dst.state >= TCPS_SYN_SENT &&
1202 	    SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
1203 		sync++;
1204 	else
1205 		pf_state_peer_ntoh(dst, &st->dst);
1206 
1207 	return (sync);
1208 }
1209 
1210 static int
pfsync_in_upd(struct mbuf * m,int offset,int count,int flags,int action)1211 pfsync_in_upd(struct mbuf *m, int offset, int count, int flags, int action)
1212 {
1213 	struct pfsync_softc *sc = V_pfsyncif;
1214 	union pfsync_state_union *sa, *sp;
1215 	struct pf_kstate *st;
1216 	struct mbuf *mp;
1217 	int sync, offp, i, total_len, msg_len, msg_version;
1218 
1219 	switch (action) {
1220 		case PFSYNC_ACT_UPD_1301:
1221 			msg_len = sizeof(struct pfsync_state_1301);
1222 			total_len = msg_len * count;
1223 			msg_version = PFSYNC_MSG_VERSION_1301;
1224 			break;
1225 		case PFSYNC_ACT_UPD_1400:
1226 			msg_len = sizeof(struct pfsync_state_1400);
1227 			total_len = msg_len * count;
1228 			msg_version = PFSYNC_MSG_VERSION_1400;
1229 			break;
1230 		default:
1231 			V_pfsyncstats.pfsyncs_badact++;
1232 			return (-1);
1233 	}
1234 
1235 	mp = m_pulldown(m, offset, total_len, &offp);
1236 	if (mp == NULL) {
1237 		V_pfsyncstats.pfsyncs_badlen++;
1238 		return (-1);
1239 	}
1240 	sa = (union pfsync_state_union *)(mp->m_data + offp);
1241 
1242 	for (i = 0; i < count; i++) {
1243 		sp = (union pfsync_state_union *)((char *)sa + msg_len * i);
1244 
1245 		/* check for invalid values */
1246 		if (sp->pfs_1301.timeout >= PFTM_MAX ||
1247 		    sp->pfs_1301.src.state > PF_TCPS_PROXY_DST ||
1248 		    sp->pfs_1301.dst.state > PF_TCPS_PROXY_DST) {
1249 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1250 				printf("pfsync_input: PFSYNC_ACT_UPD: "
1251 				    "invalid value\n");
1252 			}
1253 			V_pfsyncstats.pfsyncs_badval++;
1254 			continue;
1255 		}
1256 
1257 		st = pf_find_state_byid(sp->pfs_1301.id, sp->pfs_1301.creatorid);
1258 		if (st == NULL) {
1259 			/* insert the update */
1260 			if (pfsync_state_import(sp, flags, msg_version))
1261 				V_pfsyncstats.pfsyncs_badstate++;
1262 			continue;
1263 		}
1264 
1265 		if (st->state_flags & PFSTATE_ACK) {
1266 			pfsync_undefer_state(st, 1);
1267 		}
1268 
1269 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1270 			sync = pfsync_upd_tcp(st, &sp->pfs_1301.src, &sp->pfs_1301.dst);
1271 		else {
1272 			sync = 0;
1273 
1274 			/*
1275 			 * Non-TCP protocol state machine always go
1276 			 * forwards
1277 			 */
1278 			if (st->src.state > sp->pfs_1301.src.state)
1279 				sync++;
1280 			else
1281 				pf_state_peer_ntoh(&sp->pfs_1301.src, &st->src);
1282 			if (st->dst.state > sp->pfs_1301.dst.state)
1283 				sync++;
1284 			else
1285 				pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
1286 		}
1287 		if (sync < 2) {
1288 			pfsync_alloc_scrub_memory(&sp->pfs_1301.dst, &st->dst);
1289 			pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
1290 			st->expire = pf_get_uptime();
1291 			st->timeout = sp->pfs_1301.timeout;
1292 		}
1293 		st->pfsync_time = time_uptime;
1294 
1295 		if (sync) {
1296 			V_pfsyncstats.pfsyncs_stale++;
1297 
1298 			pfsync_update_state(st);
1299 			PF_STATE_UNLOCK(st);
1300 			pfsync_push_all(sc);
1301 			continue;
1302 		}
1303 		PF_STATE_UNLOCK(st);
1304 	}
1305 
1306 	return (total_len);
1307 }
1308 
1309 static int
pfsync_in_upd_c(struct mbuf * m,int offset,int count,int flags,int action)1310 pfsync_in_upd_c(struct mbuf *m, int offset, int count, int flags, int action)
1311 {
1312 	struct pfsync_softc *sc = V_pfsyncif;
1313 	struct pfsync_upd_c *ua, *up;
1314 	struct pf_kstate *st;
1315 	int len = count * sizeof(*up);
1316 	int sync;
1317 	struct mbuf *mp;
1318 	int offp, i;
1319 
1320 	mp = m_pulldown(m, offset, len, &offp);
1321 	if (mp == NULL) {
1322 		V_pfsyncstats.pfsyncs_badlen++;
1323 		return (-1);
1324 	}
1325 	ua = (struct pfsync_upd_c *)(mp->m_data + offp);
1326 
1327 	for (i = 0; i < count; i++) {
1328 		up = &ua[i];
1329 
1330 		/* check for invalid values */
1331 		if (up->timeout >= PFTM_MAX ||
1332 		    up->src.state > PF_TCPS_PROXY_DST ||
1333 		    up->dst.state > PF_TCPS_PROXY_DST) {
1334 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1335 				printf("pfsync_input: "
1336 				    "PFSYNC_ACT_UPD_C: "
1337 				    "invalid value\n");
1338 			}
1339 			V_pfsyncstats.pfsyncs_badval++;
1340 			continue;
1341 		}
1342 
1343 		st = pf_find_state_byid(up->id, up->creatorid);
1344 		if (st == NULL) {
1345 			/* We don't have this state. Ask for it. */
1346 			PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
1347 			pfsync_request_update(up->creatorid, up->id);
1348 			PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
1349 			continue;
1350 		}
1351 
1352 		if (st->state_flags & PFSTATE_ACK) {
1353 			pfsync_undefer_state(st, 1);
1354 		}
1355 
1356 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1357 			sync = pfsync_upd_tcp(st, &up->src, &up->dst);
1358 		else {
1359 			sync = 0;
1360 
1361 			/*
1362 			 * Non-TCP protocol state machine always go
1363 			 * forwards
1364 			 */
1365 			if (st->src.state > up->src.state)
1366 				sync++;
1367 			else
1368 				pf_state_peer_ntoh(&up->src, &st->src);
1369 			if (st->dst.state > up->dst.state)
1370 				sync++;
1371 			else
1372 				pf_state_peer_ntoh(&up->dst, &st->dst);
1373 		}
1374 		if (sync < 2) {
1375 			pfsync_alloc_scrub_memory(&up->dst, &st->dst);
1376 			pf_state_peer_ntoh(&up->dst, &st->dst);
1377 			st->expire = pf_get_uptime();
1378 			st->timeout = up->timeout;
1379 		}
1380 		st->pfsync_time = time_uptime;
1381 
1382 		if (sync) {
1383 			V_pfsyncstats.pfsyncs_stale++;
1384 
1385 			pfsync_update_state(st);
1386 			PF_STATE_UNLOCK(st);
1387 			pfsync_push_all(sc);
1388 			continue;
1389 		}
1390 		PF_STATE_UNLOCK(st);
1391 	}
1392 
1393 	return (len);
1394 }
1395 
1396 static int
pfsync_in_ureq(struct mbuf * m,int offset,int count,int flags,int action)1397 pfsync_in_ureq(struct mbuf *m, int offset, int count, int flags, int action)
1398 {
1399 	struct pfsync_upd_req *ur, *ura;
1400 	struct mbuf *mp;
1401 	int len = count * sizeof(*ur);
1402 	int i, offp;
1403 
1404 	struct pf_kstate *st;
1405 
1406 	mp = m_pulldown(m, offset, len, &offp);
1407 	if (mp == NULL) {
1408 		V_pfsyncstats.pfsyncs_badlen++;
1409 		return (-1);
1410 	}
1411 	ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1412 
1413 	for (i = 0; i < count; i++) {
1414 		ur = &ura[i];
1415 
1416 		if (ur->id == 0 && ur->creatorid == 0)
1417 			pfsync_bulk_start();
1418 		else {
1419 			st = pf_find_state_byid(ur->id, ur->creatorid);
1420 			if (st == NULL) {
1421 				V_pfsyncstats.pfsyncs_badstate++;
1422 				continue;
1423 			}
1424 			if (st->state_flags & PFSTATE_NOSYNC) {
1425 				PF_STATE_UNLOCK(st);
1426 				continue;
1427 			}
1428 
1429 			pfsync_update_state_req(st);
1430 			PF_STATE_UNLOCK(st);
1431 		}
1432 	}
1433 
1434 	return (len);
1435 }
1436 
1437 static int
pfsync_in_del_c(struct mbuf * m,int offset,int count,int flags,int action)1438 pfsync_in_del_c(struct mbuf *m, int offset, int count, int flags, int action)
1439 {
1440 	struct mbuf *mp;
1441 	struct pfsync_del_c *sa, *sp;
1442 	struct pf_kstate *st;
1443 	int len = count * sizeof(*sp);
1444 	int offp, i;
1445 
1446 	mp = m_pulldown(m, offset, len, &offp);
1447 	if (mp == NULL) {
1448 		V_pfsyncstats.pfsyncs_badlen++;
1449 		return (-1);
1450 	}
1451 	sa = (struct pfsync_del_c *)(mp->m_data + offp);
1452 
1453 	for (i = 0; i < count; i++) {
1454 		sp = &sa[i];
1455 
1456 		st = pf_find_state_byid(sp->id, sp->creatorid);
1457 		if (st == NULL) {
1458 			V_pfsyncstats.pfsyncs_badstate++;
1459 			continue;
1460 		}
1461 
1462 		st->state_flags |= PFSTATE_NOSYNC;
1463 		pf_remove_state(st);
1464 	}
1465 
1466 	return (len);
1467 }
1468 
1469 static int
pfsync_in_bus(struct mbuf * m,int offset,int count,int flags,int action)1470 pfsync_in_bus(struct mbuf *m, int offset, int count, int flags, int action)
1471 {
1472 	struct pfsync_softc *sc = V_pfsyncif;
1473 	struct pfsync_bus *bus;
1474 	struct mbuf *mp;
1475 	int len = count * sizeof(*bus);
1476 	int offp;
1477 
1478 	PFSYNC_BLOCK(sc);
1479 
1480 	/* If we're not waiting for a bulk update, who cares. */
1481 	if (sc->sc_ureq_sent == 0) {
1482 		PFSYNC_BUNLOCK(sc);
1483 		return (len);
1484 	}
1485 
1486 	mp = m_pulldown(m, offset, len, &offp);
1487 	if (mp == NULL) {
1488 		PFSYNC_BUNLOCK(sc);
1489 		V_pfsyncstats.pfsyncs_badlen++;
1490 		return (-1);
1491 	}
1492 	bus = (struct pfsync_bus *)(mp->m_data + offp);
1493 
1494 	switch (bus->status) {
1495 	case PFSYNC_BUS_START:
1496 		callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1497 		    V_pf_limits[PF_LIMIT_STATES].limit /
1498 		    ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1499 		    sizeof(union pfsync_state_union)),
1500 		    pfsync_bulk_fail, sc);
1501 		if (V_pf_status.debug >= PF_DEBUG_MISC)
1502 			printf("pfsync: received bulk update start\n");
1503 		break;
1504 
1505 	case PFSYNC_BUS_END:
1506 		if (time_uptime - ntohl(bus->endtime) >=
1507 		    sc->sc_ureq_sent) {
1508 			/* that's it, we're happy */
1509 			sc->sc_ureq_sent = 0;
1510 			sc->sc_bulk_tries = 0;
1511 			callout_stop(&sc->sc_bulkfail_tmo);
1512 			if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1513 				(*carp_demote_adj_p)(-V_pfsync_carp_adj,
1514 				    "pfsync bulk done");
1515 			sc->sc_flags |= PFSYNCF_OK;
1516 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1517 				printf("pfsync: received valid "
1518 				    "bulk update end\n");
1519 		} else {
1520 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1521 				printf("pfsync: received invalid "
1522 				    "bulk update end: bad timestamp\n");
1523 		}
1524 		break;
1525 	}
1526 	PFSYNC_BUNLOCK(sc);
1527 
1528 	return (len);
1529 }
1530 
1531 static int
pfsync_in_tdb(struct mbuf * m,int offset,int count,int flags,int action)1532 pfsync_in_tdb(struct mbuf *m, int offset, int count, int flags, int action)
1533 {
1534 	int len = count * sizeof(struct pfsync_tdb);
1535 
1536 #if defined(IPSEC)
1537 	struct pfsync_tdb *tp;
1538 	struct mbuf *mp;
1539 	int offp;
1540 	int i;
1541 	int s;
1542 
1543 	mp = m_pulldown(m, offset, len, &offp);
1544 	if (mp == NULL) {
1545 		V_pfsyncstats.pfsyncs_badlen++;
1546 		return (-1);
1547 	}
1548 	tp = (struct pfsync_tdb *)(mp->m_data + offp);
1549 
1550 	for (i = 0; i < count; i++)
1551 		pfsync_update_net_tdb(&tp[i]);
1552 #endif
1553 
1554 	return (len);
1555 }
1556 
1557 #if defined(IPSEC)
1558 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1559 static void
pfsync_update_net_tdb(struct pfsync_tdb * pt)1560 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1561 {
1562 	struct tdb		*tdb;
1563 	int			 s;
1564 
1565 	/* check for invalid values */
1566 	if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1567 	    (pt->dst.sa.sa_family != AF_INET &&
1568 	    pt->dst.sa.sa_family != AF_INET6))
1569 		goto bad;
1570 
1571 	tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1572 	if (tdb) {
1573 		pt->rpl = ntohl(pt->rpl);
1574 		pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1575 
1576 		/* Neither replay nor byte counter should ever decrease. */
1577 		if (pt->rpl < tdb->tdb_rpl ||
1578 		    pt->cur_bytes < tdb->tdb_cur_bytes) {
1579 			goto bad;
1580 		}
1581 
1582 		tdb->tdb_rpl = pt->rpl;
1583 		tdb->tdb_cur_bytes = pt->cur_bytes;
1584 	}
1585 	return;
1586 
1587 bad:
1588 	if (V_pf_status.debug >= PF_DEBUG_MISC)
1589 		printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1590 		    "invalid value\n");
1591 	V_pfsyncstats.pfsyncs_badstate++;
1592 	return;
1593 }
1594 #endif
1595 
1596 static int
pfsync_in_eof(struct mbuf * m,int offset,int count,int flags,int action)1597 pfsync_in_eof(struct mbuf *m, int offset, int count, int flags, int action)
1598 {
1599 	/* check if we are at the right place in the packet */
1600 	if (offset != m->m_pkthdr.len)
1601 		V_pfsyncstats.pfsyncs_badlen++;
1602 
1603 	/* we're done. free and let the caller return */
1604 	m_freem(m);
1605 	return (-1);
1606 }
1607 
1608 static int
pfsync_in_error(struct mbuf * m,int offset,int count,int flags,int action)1609 pfsync_in_error(struct mbuf *m, int offset, int count, int flags, int action)
1610 {
1611 	V_pfsyncstats.pfsyncs_badact++;
1612 
1613 	m_freem(m);
1614 	return (-1);
1615 }
1616 
1617 static int
pfsyncoutput(struct ifnet * ifp,struct mbuf * m,const struct sockaddr * dst,struct route * rt)1618 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1619 	struct route *rt)
1620 {
1621 	m_freem(m);
1622 	return (0);
1623 }
1624 
1625 /* ARGSUSED */
1626 static int
pfsyncioctl(struct ifnet * ifp,u_long cmd,caddr_t data)1627 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1628 {
1629 	struct pfsync_softc *sc = ifp->if_softc;
1630 	struct ifreq *ifr = (struct ifreq *)data;
1631 	struct pfsyncreq pfsyncr;
1632 	size_t nvbuflen;
1633 	int error;
1634 	int c;
1635 
1636 	switch (cmd) {
1637 	case SIOCSIFFLAGS:
1638 		PFSYNC_LOCK(sc);
1639 		if (ifp->if_flags & IFF_UP) {
1640 			ifp->if_drv_flags |= IFF_DRV_RUNNING;
1641 			PFSYNC_UNLOCK(sc);
1642 			pfsync_pointers_init();
1643 		} else {
1644 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1645 			PFSYNC_UNLOCK(sc);
1646 			pfsync_pointers_uninit();
1647 		}
1648 		break;
1649 	case SIOCSIFMTU:
1650 		if (!sc->sc_sync_if ||
1651 		    ifr->ifr_mtu <= PFSYNC_MINPKT ||
1652 		    ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1653 			return (EINVAL);
1654 		if (ifr->ifr_mtu < ifp->if_mtu) {
1655 			for (c = 0; c < pfsync_buckets; c++) {
1656 				PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
1657 				if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT)
1658 					pfsync_sendout(1, c);
1659 				PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
1660 			}
1661 		}
1662 		ifp->if_mtu = ifr->ifr_mtu;
1663 		break;
1664 	case SIOCGETPFSYNC:
1665 		bzero(&pfsyncr, sizeof(pfsyncr));
1666 		PFSYNC_LOCK(sc);
1667 		if (sc->sc_sync_if) {
1668 			strlcpy(pfsyncr.pfsyncr_syncdev,
1669 			    sc->sc_sync_if->if_xname, IFNAMSIZ);
1670 		}
1671 		pfsyncr.pfsyncr_syncpeer = ((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr;
1672 		pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1673 		pfsyncr.pfsyncr_defer = sc->sc_flags;
1674 		PFSYNC_UNLOCK(sc);
1675 		return (copyout(&pfsyncr, ifr_data_get_ptr(ifr),
1676 		    sizeof(pfsyncr)));
1677 
1678 	case SIOCGETPFSYNCNV:
1679 	    {
1680 		nvlist_t *nvl_syncpeer;
1681 		nvlist_t *nvl = nvlist_create(0);
1682 
1683 		if (nvl == NULL)
1684 			return (ENOMEM);
1685 
1686 		if (sc->sc_sync_if)
1687 			nvlist_add_string(nvl, "syncdev", sc->sc_sync_if->if_xname);
1688 		nvlist_add_number(nvl, "maxupdates", sc->sc_maxupdates);
1689 		nvlist_add_number(nvl, "flags", sc->sc_flags);
1690 		nvlist_add_number(nvl, "version", sc->sc_version);
1691 		if ((nvl_syncpeer = pfsync_sockaddr_to_syncpeer_nvlist(&sc->sc_sync_peer)) != NULL)
1692 			nvlist_add_nvlist(nvl, "syncpeer", nvl_syncpeer);
1693 
1694 		void *packed = NULL;
1695 		packed = nvlist_pack(nvl, &nvbuflen);
1696 		if (packed == NULL) {
1697 			free(packed, M_NVLIST);
1698 			nvlist_destroy(nvl);
1699 			return (ENOMEM);
1700 		}
1701 
1702 		if (nvbuflen > ifr->ifr_cap_nv.buf_length) {
1703 			ifr->ifr_cap_nv.length = nvbuflen;
1704 			ifr->ifr_cap_nv.buffer = NULL;
1705 			free(packed, M_NVLIST);
1706 			nvlist_destroy(nvl);
1707 			return (EFBIG);
1708 		}
1709 
1710 		ifr->ifr_cap_nv.length = nvbuflen;
1711 		error = copyout(packed, ifr->ifr_cap_nv.buffer, nvbuflen);
1712 
1713 		nvlist_destroy(nvl);
1714 		nvlist_destroy(nvl_syncpeer);
1715 		free(packed, M_NVLIST);
1716 		break;
1717 	    }
1718 
1719 	case SIOCSETPFSYNC:
1720 	    {
1721 		struct pfsync_kstatus status;
1722 
1723 		if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1724 			return (error);
1725 		if ((error = copyin(ifr_data_get_ptr(ifr), &pfsyncr,
1726 		    sizeof(pfsyncr))))
1727 			return (error);
1728 
1729 		memset((char *)&status, 0, sizeof(struct pfsync_kstatus));
1730 		pfsync_pfsyncreq_to_kstatus(&pfsyncr, &status);
1731 
1732 		error = pfsync_kstatus_to_softc(&status, sc);
1733 		return (error);
1734 	    }
1735 	case SIOCSETPFSYNCNV:
1736 	    {
1737 		struct pfsync_kstatus status;
1738 		void *data;
1739 		nvlist_t *nvl;
1740 
1741 		if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1742 			return (error);
1743 		if (ifr->ifr_cap_nv.length > IFR_CAP_NV_MAXBUFSIZE)
1744 			return (EINVAL);
1745 
1746 		data = malloc(ifr->ifr_cap_nv.length, M_TEMP, M_WAITOK);
1747 
1748 		if ((error = copyin(ifr->ifr_cap_nv.buffer, data,
1749 		    ifr->ifr_cap_nv.length)) != 0) {
1750 			free(data, M_TEMP);
1751 			return (error);
1752 		}
1753 
1754 		if ((nvl = nvlist_unpack(data, ifr->ifr_cap_nv.length, 0)) == NULL) {
1755 			free(data, M_TEMP);
1756 			return (EINVAL);
1757 		}
1758 
1759 		memset((char *)&status, 0, sizeof(struct pfsync_kstatus));
1760 		pfsync_nvstatus_to_kstatus(nvl, &status);
1761 
1762 		nvlist_destroy(nvl);
1763 		free(data, M_TEMP);
1764 
1765 		error = pfsync_kstatus_to_softc(&status, sc);
1766 		return (error);
1767 	    }
1768 	default:
1769 		return (ENOTTY);
1770 	}
1771 
1772 	return (0);
1773 }
1774 
1775 static void
pfsync_out_state_1301(struct pf_kstate * st,void * buf)1776 pfsync_out_state_1301(struct pf_kstate *st, void *buf)
1777 {
1778 	union pfsync_state_union *sp = buf;
1779 
1780 	pfsync_state_export(sp, st, PFSYNC_MSG_VERSION_1301);
1781 }
1782 
1783 static void
pfsync_out_state_1400(struct pf_kstate * st,void * buf)1784 pfsync_out_state_1400(struct pf_kstate *st, void *buf)
1785 {
1786 	union pfsync_state_union *sp = buf;
1787 
1788 	pfsync_state_export(sp, st, PFSYNC_MSG_VERSION_1400);
1789 }
1790 
1791 static void
pfsync_out_iack(struct pf_kstate * st,void * buf)1792 pfsync_out_iack(struct pf_kstate *st, void *buf)
1793 {
1794 	struct pfsync_ins_ack *iack = buf;
1795 
1796 	iack->id = st->id;
1797 	iack->creatorid = st->creatorid;
1798 }
1799 
1800 static void
pfsync_out_upd_c(struct pf_kstate * st,void * buf)1801 pfsync_out_upd_c(struct pf_kstate *st, void *buf)
1802 {
1803 	struct pfsync_upd_c *up = buf;
1804 
1805 	bzero(up, sizeof(*up));
1806 	up->id = st->id;
1807 	pf_state_peer_hton(&st->src, &up->src);
1808 	pf_state_peer_hton(&st->dst, &up->dst);
1809 	up->creatorid = st->creatorid;
1810 	up->timeout = st->timeout;
1811 }
1812 
1813 static void
pfsync_out_del_c(struct pf_kstate * st,void * buf)1814 pfsync_out_del_c(struct pf_kstate *st, void *buf)
1815 {
1816 	struct pfsync_del_c *dp = buf;
1817 
1818 	dp->id = st->id;
1819 	dp->creatorid = st->creatorid;
1820 	st->state_flags |= PFSTATE_NOSYNC;
1821 }
1822 
1823 static void
pfsync_drop_all(struct pfsync_softc * sc)1824 pfsync_drop_all(struct pfsync_softc *sc)
1825 {
1826 	struct pfsync_bucket *b;
1827 	int c;
1828 
1829 	for (c = 0; c < pfsync_buckets; c++) {
1830 		b = &sc->sc_buckets[c];
1831 
1832 		PFSYNC_BUCKET_LOCK(b);
1833 		pfsync_drop(sc, c);
1834 		PFSYNC_BUCKET_UNLOCK(b);
1835 	}
1836 }
1837 
1838 static void
pfsync_drop(struct pfsync_softc * sc,int c)1839 pfsync_drop(struct pfsync_softc *sc, int c)
1840 {
1841 	struct pf_kstate *st, *next;
1842 	struct pfsync_upd_req_item *ur;
1843 	struct pfsync_bucket *b;
1844 	enum pfsync_q_id q;
1845 
1846 	b = &sc->sc_buckets[c];
1847 	PFSYNC_BUCKET_LOCK_ASSERT(b);
1848 
1849 	for (q = 0; q < PFSYNC_Q_COUNT; q++) {
1850 		if (TAILQ_EMPTY(&b->b_qs[q]))
1851 			continue;
1852 
1853 		TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, next) {
1854 			KASSERT(st->sync_state == pfsync_qid_sstate[q],
1855 				("%s: st->sync_state %d == q %d",
1856 					__func__, st->sync_state, q));
1857 			st->sync_state = PFSYNC_S_NONE;
1858 			pf_release_state(st);
1859 		}
1860 		TAILQ_INIT(&b->b_qs[q]);
1861 	}
1862 
1863 	while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1864 		TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1865 		free(ur, M_PFSYNC);
1866 	}
1867 
1868 	b->b_len = PFSYNC_MINPKT;
1869 	free(b->b_plus, M_PFSYNC);
1870 	b->b_plus = NULL;
1871 	b->b_pluslen = 0;
1872 }
1873 
1874 static void
pfsync_sendout(int schedswi,int c)1875 pfsync_sendout(int schedswi, int c)
1876 {
1877 	struct pfsync_softc *sc = V_pfsyncif;
1878 	struct ifnet *ifp = sc->sc_ifp;
1879 	struct mbuf *m;
1880 	struct pfsync_header *ph;
1881 	struct pfsync_subheader *subh;
1882 	struct pf_kstate *st, *st_next;
1883 	struct pfsync_upd_req_item *ur;
1884 	struct pfsync_bucket *b = &sc->sc_buckets[c];
1885 	size_t len;
1886 	int aflen, offset, count = 0;
1887 	enum pfsync_q_id q;
1888 
1889 	KASSERT(sc != NULL, ("%s: null sc", __func__));
1890 	KASSERT(b->b_len > PFSYNC_MINPKT,
1891 	    ("%s: sc_len %zu", __func__, b->b_len));
1892 	PFSYNC_BUCKET_LOCK_ASSERT(b);
1893 
1894 	if (!bpf_peers_present(ifp->if_bpf) && sc->sc_sync_if == NULL) {
1895 		pfsync_drop(sc, c);
1896 		return;
1897 	}
1898 
1899 	m = m_get2(max_linkhdr + b->b_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1900 	if (m == NULL) {
1901 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
1902 		V_pfsyncstats.pfsyncs_onomem++;
1903 		return;
1904 	}
1905 	m->m_data += max_linkhdr;
1906 	bzero(m->m_data, b->b_len);
1907 
1908 	len = b->b_len;
1909 
1910 	/* build the ip header */
1911 	switch (sc->sc_sync_peer.ss_family) {
1912 #ifdef INET
1913 	case AF_INET:
1914 	    {
1915 		struct ip *ip;
1916 
1917 		ip = mtod(m, struct ip *);
1918 		bcopy(&sc->sc_template.ipv4, ip, sizeof(*ip));
1919 		aflen = offset = sizeof(*ip);
1920 
1921 		len -= sizeof(union inet_template) - sizeof(struct ip);
1922 		ip->ip_len = htons(len);
1923 		ip_fillid(ip, V_ip_random_id);
1924 		break;
1925 	    }
1926 #endif
1927 #ifdef INET6
1928 	case AF_INET6:
1929 		{
1930 		struct ip6_hdr *ip6;
1931 
1932 		ip6 = mtod(m, struct ip6_hdr *);
1933 		bcopy(&sc->sc_template.ipv6, ip6, sizeof(*ip6));
1934 		aflen = offset = sizeof(*ip6);
1935 
1936 		len -= sizeof(union inet_template) - sizeof(struct ip6_hdr);
1937 		ip6->ip6_plen = htons(len);
1938 		break;
1939 		}
1940 #endif
1941 	default:
1942 		m_freem(m);
1943 		pfsync_drop(sc, c);
1944 		return;
1945 	}
1946 	m->m_len = m->m_pkthdr.len = len;
1947 
1948 	/* build the pfsync header */
1949 	ph = (struct pfsync_header *)(m->m_data + offset);
1950 	offset += sizeof(*ph);
1951 
1952 	ph->version = PFSYNC_VERSION;
1953 	ph->len = htons(len - aflen);
1954 	bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1955 
1956 	/* walk the queues */
1957 	for (q = 0; q < PFSYNC_Q_COUNT; q++) {
1958 		if (TAILQ_EMPTY(&b->b_qs[q]))
1959 			continue;
1960 
1961 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1962 		offset += sizeof(*subh);
1963 
1964 		count = 0;
1965 		TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, st_next) {
1966 			KASSERT(st->sync_state == pfsync_qid_sstate[q],
1967 				("%s: st->sync_state == q",
1968 					__func__));
1969 			/*
1970 			 * XXXGL: some of write methods do unlocked reads
1971 			 * of state data :(
1972 			 */
1973 			pfsync_qs[q].write(st, m->m_data + offset);
1974 			offset += pfsync_qs[q].len;
1975 			st->sync_state = PFSYNC_S_NONE;
1976 			pf_release_state(st);
1977 			count++;
1978 		}
1979 		TAILQ_INIT(&b->b_qs[q]);
1980 
1981 		subh->action = pfsync_qs[q].action;
1982 		subh->count = htons(count);
1983 		V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1984 	}
1985 
1986 	if (!TAILQ_EMPTY(&b->b_upd_req_list)) {
1987 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1988 		offset += sizeof(*subh);
1989 
1990 		count = 0;
1991 		while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1992 			TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1993 
1994 			bcopy(&ur->ur_msg, m->m_data + offset,
1995 			    sizeof(ur->ur_msg));
1996 			offset += sizeof(ur->ur_msg);
1997 			free(ur, M_PFSYNC);
1998 			count++;
1999 		}
2000 
2001 		subh->action = PFSYNC_ACT_UPD_REQ;
2002 		subh->count = htons(count);
2003 		V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
2004 	}
2005 
2006 	/* has someone built a custom region for us to add? */
2007 	if (b->b_plus != NULL) {
2008 		bcopy(b->b_plus, m->m_data + offset, b->b_pluslen);
2009 		offset += b->b_pluslen;
2010 
2011 		free(b->b_plus, M_PFSYNC);
2012 		b->b_plus = NULL;
2013 		b->b_pluslen = 0;
2014 	}
2015 
2016 	subh = (struct pfsync_subheader *)(m->m_data + offset);
2017 	offset += sizeof(*subh);
2018 
2019 	subh->action = PFSYNC_ACT_EOF;
2020 	subh->count = htons(1);
2021 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
2022 
2023 	/* we're done, let's put it on the wire */
2024 	if (bpf_peers_present(ifp->if_bpf)) {
2025 		m->m_data += aflen;
2026 		m->m_len = m->m_pkthdr.len = len - aflen;
2027 		bpf_mtap(ifp->if_bpf, m);
2028 		m->m_data -= aflen;
2029 		m->m_len = m->m_pkthdr.len = len;
2030 	}
2031 
2032 	if (sc->sc_sync_if == NULL) {
2033 		b->b_len = PFSYNC_MINPKT;
2034 		m_freem(m);
2035 		return;
2036 	}
2037 
2038 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1);
2039 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len);
2040 	b->b_len = PFSYNC_MINPKT;
2041 
2042 	if (!_IF_QFULL(&b->b_snd))
2043 		_IF_ENQUEUE(&b->b_snd, m);
2044 	else {
2045 		m_freem(m);
2046 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1);
2047 	}
2048 	if (schedswi)
2049 		swi_sched(V_pfsync_swi_cookie, 0);
2050 }
2051 
2052 static void
pfsync_insert_state(struct pf_kstate * st)2053 pfsync_insert_state(struct pf_kstate *st)
2054 {
2055 	struct pfsync_softc *sc = V_pfsyncif;
2056 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2057 
2058 	if (st->state_flags & PFSTATE_NOSYNC)
2059 		return;
2060 
2061 	if ((st->rule->rule_flag & PFRULE_NOSYNC) ||
2062 	    st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
2063 		st->state_flags |= PFSTATE_NOSYNC;
2064 		return;
2065 	}
2066 
2067 	KASSERT(st->sync_state == PFSYNC_S_NONE,
2068 		("%s: st->sync_state %u", __func__, st->sync_state));
2069 
2070 	PFSYNC_BUCKET_LOCK(b);
2071 	if (b->b_len == PFSYNC_MINPKT)
2072 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2073 
2074 	pfsync_q_ins(st, PFSYNC_S_INS, true);
2075 	PFSYNC_BUCKET_UNLOCK(b);
2076 
2077 	st->sync_updates = 0;
2078 }
2079 
2080 static int
pfsync_defer(struct pf_kstate * st,struct mbuf * m)2081 pfsync_defer(struct pf_kstate *st, struct mbuf *m)
2082 {
2083 	struct pfsync_softc *sc = V_pfsyncif;
2084 	struct pfsync_deferral *pd;
2085 	struct pfsync_bucket *b;
2086 
2087 	if (m->m_flags & (M_BCAST|M_MCAST))
2088 		return (0);
2089 
2090 	if (sc == NULL)
2091 		return (0);
2092 
2093 	b = pfsync_get_bucket(sc, st);
2094 
2095 	PFSYNC_LOCK(sc);
2096 
2097 	if (!(sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) ||
2098 	    !(sc->sc_flags & PFSYNCF_DEFER)) {
2099 		PFSYNC_UNLOCK(sc);
2100 		return (0);
2101 	}
2102 
2103 	PFSYNC_BUCKET_LOCK(b);
2104 	PFSYNC_UNLOCK(sc);
2105 
2106 	if (b->b_deferred >= 128)
2107 		pfsync_undefer(TAILQ_FIRST(&b->b_deferrals), 0);
2108 
2109 	pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
2110 	if (pd == NULL) {
2111 		PFSYNC_BUCKET_UNLOCK(b);
2112 		return (0);
2113 	}
2114 	b->b_deferred++;
2115 
2116 	m->m_flags |= M_SKIP_FIREWALL;
2117 	st->state_flags |= PFSTATE_ACK;
2118 
2119 	pd->pd_sc = sc;
2120 	pd->pd_st = st;
2121 	pf_ref_state(st);
2122 	pd->pd_m = m;
2123 
2124 	TAILQ_INSERT_TAIL(&b->b_deferrals, pd, pd_entry);
2125 	callout_init_mtx(&pd->pd_tmo, &b->b_mtx, CALLOUT_RETURNUNLOCKED);
2126 	callout_reset(&pd->pd_tmo, (V_pfsync_defer_timeout * hz) / 1000,
2127 	    pfsync_defer_tmo, pd);
2128 
2129 	pfsync_push(b);
2130 	PFSYNC_BUCKET_UNLOCK(b);
2131 
2132 	return (1);
2133 }
2134 
2135 static void
pfsync_undefer(struct pfsync_deferral * pd,int drop)2136 pfsync_undefer(struct pfsync_deferral *pd, int drop)
2137 {
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 = pfsync_get_bucket(sc, st);
2142 
2143 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2144 
2145 	TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
2146 	b->b_deferred--;
2147 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
2148 	free(pd, M_PFSYNC);
2149 	pf_release_state(st);
2150 
2151 	if (drop)
2152 		m_freem(m);
2153 	else {
2154 		_IF_ENQUEUE(&b->b_snd, m);
2155 		pfsync_push(b);
2156 	}
2157 }
2158 
2159 static void
pfsync_defer_tmo(void * arg)2160 pfsync_defer_tmo(void *arg)
2161 {
2162 	struct epoch_tracker et;
2163 	struct pfsync_deferral *pd = arg;
2164 	struct pfsync_softc *sc = pd->pd_sc;
2165 	struct mbuf *m = pd->pd_m;
2166 	struct pf_kstate *st = pd->pd_st;
2167 	struct pfsync_bucket *b;
2168 
2169 	CURVNET_SET(sc->sc_ifp->if_vnet);
2170 
2171 	b = pfsync_get_bucket(sc, st);
2172 
2173 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2174 
2175 	TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
2176 	b->b_deferred--;
2177 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
2178 	PFSYNC_BUCKET_UNLOCK(b);
2179 	free(pd, M_PFSYNC);
2180 
2181 	if (sc->sc_sync_if == NULL) {
2182 		pf_release_state(st);
2183 		m_freem(m);
2184 		CURVNET_RESTORE();
2185 		return;
2186 	}
2187 
2188 	NET_EPOCH_ENTER(et);
2189 
2190 	pfsync_tx(sc, m);
2191 
2192 	pf_release_state(st);
2193 
2194 	CURVNET_RESTORE();
2195 	NET_EPOCH_EXIT(et);
2196 }
2197 
2198 static void
pfsync_undefer_state_locked(struct pf_kstate * st,int drop)2199 pfsync_undefer_state_locked(struct pf_kstate *st, int drop)
2200 {
2201 	struct pfsync_softc *sc = V_pfsyncif;
2202 	struct pfsync_deferral *pd;
2203 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2204 
2205 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2206 
2207 	TAILQ_FOREACH(pd, &b->b_deferrals, pd_entry) {
2208 		 if (pd->pd_st == st) {
2209 			if (callout_stop(&pd->pd_tmo) > 0)
2210 				pfsync_undefer(pd, drop);
2211 
2212 			return;
2213 		}
2214 	}
2215 
2216 	panic("%s: unable to find deferred state", __func__);
2217 }
2218 
2219 static void
pfsync_undefer_state(struct pf_kstate * st,int drop)2220 pfsync_undefer_state(struct pf_kstate *st, int drop)
2221 {
2222 	struct pfsync_softc *sc = V_pfsyncif;
2223 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2224 
2225 	PFSYNC_BUCKET_LOCK(b);
2226 	pfsync_undefer_state_locked(st, drop);
2227 	PFSYNC_BUCKET_UNLOCK(b);
2228 }
2229 
2230 static struct pfsync_bucket*
pfsync_get_bucket(struct pfsync_softc * sc,struct pf_kstate * st)2231 pfsync_get_bucket(struct pfsync_softc *sc, struct pf_kstate *st)
2232 {
2233 	int c = PF_IDHASH(st) % pfsync_buckets;
2234 	return &sc->sc_buckets[c];
2235 }
2236 
2237 static void
pfsync_update_state(struct pf_kstate * st)2238 pfsync_update_state(struct pf_kstate *st)
2239 {
2240 	struct pfsync_softc *sc = V_pfsyncif;
2241 	bool sync = false, ref = true;
2242 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2243 
2244 	PF_STATE_LOCK_ASSERT(st);
2245 	PFSYNC_BUCKET_LOCK(b);
2246 
2247 	if (st->state_flags & PFSTATE_ACK)
2248 		pfsync_undefer_state_locked(st, 0);
2249 	if (st->state_flags & PFSTATE_NOSYNC) {
2250 		if (st->sync_state != PFSYNC_S_NONE)
2251 			pfsync_q_del(st, true, b);
2252 		PFSYNC_BUCKET_UNLOCK(b);
2253 		return;
2254 	}
2255 
2256 	if (b->b_len == PFSYNC_MINPKT)
2257 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2258 
2259 	switch (st->sync_state) {
2260 	case PFSYNC_S_UPD_C:
2261 	case PFSYNC_S_UPD:
2262 	case PFSYNC_S_INS:
2263 		/* we're already handling it */
2264 
2265 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
2266 			st->sync_updates++;
2267 			if (st->sync_updates >= sc->sc_maxupdates)
2268 				sync = true;
2269 		}
2270 		break;
2271 
2272 	case PFSYNC_S_IACK:
2273 		pfsync_q_del(st, false, b);
2274 		ref = false;
2275 		/* FALLTHROUGH */
2276 
2277 	case PFSYNC_S_NONE:
2278 		pfsync_q_ins(st, PFSYNC_S_UPD_C, ref);
2279 		st->sync_updates = 0;
2280 		break;
2281 
2282 	default:
2283 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2284 	}
2285 
2286 	if (sync || (time_uptime - st->pfsync_time) < 2)
2287 		pfsync_push(b);
2288 
2289 	PFSYNC_BUCKET_UNLOCK(b);
2290 }
2291 
2292 static void
pfsync_request_update(u_int32_t creatorid,u_int64_t id)2293 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
2294 {
2295 	struct pfsync_softc *sc = V_pfsyncif;
2296 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2297 	struct pfsync_upd_req_item *item;
2298 	size_t nlen = sizeof(struct pfsync_upd_req);
2299 
2300 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2301 
2302 	/*
2303 	 * This code does a bit to prevent multiple update requests for the
2304 	 * same state being generated. It searches current subheader queue,
2305 	 * but it doesn't lookup into queue of already packed datagrams.
2306 	 */
2307 	TAILQ_FOREACH(item, &b->b_upd_req_list, ur_entry)
2308 		if (item->ur_msg.id == id &&
2309 		    item->ur_msg.creatorid == creatorid)
2310 			return;
2311 
2312 	item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
2313 	if (item == NULL)
2314 		return; /* XXX stats */
2315 
2316 	item->ur_msg.id = id;
2317 	item->ur_msg.creatorid = creatorid;
2318 
2319 	if (TAILQ_EMPTY(&b->b_upd_req_list))
2320 		nlen += sizeof(struct pfsync_subheader);
2321 
2322 	if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2323 		pfsync_sendout(0, 0);
2324 
2325 		nlen = sizeof(struct pfsync_subheader) +
2326 		    sizeof(struct pfsync_upd_req);
2327 	}
2328 
2329 	TAILQ_INSERT_TAIL(&b->b_upd_req_list, item, ur_entry);
2330 	b->b_len += nlen;
2331 
2332 	pfsync_push(b);
2333 }
2334 
2335 static bool
pfsync_update_state_req(struct pf_kstate * st)2336 pfsync_update_state_req(struct pf_kstate *st)
2337 {
2338 	struct pfsync_softc *sc = V_pfsyncif;
2339 	bool ref = true, full = false;
2340 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2341 
2342 	PF_STATE_LOCK_ASSERT(st);
2343 	PFSYNC_BUCKET_LOCK(b);
2344 
2345 	if (st->state_flags & PFSTATE_NOSYNC) {
2346 		if (st->sync_state != PFSYNC_S_NONE)
2347 			pfsync_q_del(st, true, b);
2348 		PFSYNC_BUCKET_UNLOCK(b);
2349 		return (full);
2350 	}
2351 
2352 	switch (st->sync_state) {
2353 	case PFSYNC_S_UPD_C:
2354 	case PFSYNC_S_IACK:
2355 		pfsync_q_del(st, false, b);
2356 		ref = false;
2357 		/* FALLTHROUGH */
2358 
2359 	case PFSYNC_S_NONE:
2360 		pfsync_q_ins(st, PFSYNC_S_UPD, ref);
2361 		pfsync_push(b);
2362 		break;
2363 
2364 	case PFSYNC_S_INS:
2365 	case PFSYNC_S_UPD:
2366 	case PFSYNC_S_DEL_C:
2367 		/* we're already handling it */
2368 		break;
2369 
2370 	default:
2371 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2372 	}
2373 
2374 	if ((sc->sc_ifp->if_mtu - b->b_len) < sizeof(union pfsync_state_union))
2375 		full = true;
2376 
2377 	PFSYNC_BUCKET_UNLOCK(b);
2378 
2379 	return (full);
2380 }
2381 
2382 static void
pfsync_delete_state(struct pf_kstate * st)2383 pfsync_delete_state(struct pf_kstate *st)
2384 {
2385 	struct pfsync_softc *sc = V_pfsyncif;
2386 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2387 	bool ref = true;
2388 
2389 	PFSYNC_BUCKET_LOCK(b);
2390 	if (st->state_flags & PFSTATE_ACK)
2391 		pfsync_undefer_state_locked(st, 1);
2392 	if (st->state_flags & PFSTATE_NOSYNC) {
2393 		if (st->sync_state != PFSYNC_S_NONE)
2394 			pfsync_q_del(st, true, b);
2395 		PFSYNC_BUCKET_UNLOCK(b);
2396 		return;
2397 	}
2398 
2399 	if (b->b_len == PFSYNC_MINPKT)
2400 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2401 
2402 	switch (st->sync_state) {
2403 	case PFSYNC_S_INS:
2404 		/* We never got to tell the world so just forget about it. */
2405 		pfsync_q_del(st, true, b);
2406 		break;
2407 
2408 	case PFSYNC_S_UPD_C:
2409 	case PFSYNC_S_UPD:
2410 	case PFSYNC_S_IACK:
2411 		pfsync_q_del(st, false, b);
2412 		ref = false;
2413 		/* FALLTHROUGH */
2414 
2415 	case PFSYNC_S_NONE:
2416 		pfsync_q_ins(st, PFSYNC_S_DEL_C, ref);
2417 		break;
2418 
2419 	default:
2420 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2421 	}
2422 
2423 	PFSYNC_BUCKET_UNLOCK(b);
2424 }
2425 
2426 static void
pfsync_clear_states(u_int32_t creatorid,const char * ifname)2427 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
2428 {
2429 	struct {
2430 		struct pfsync_subheader subh;
2431 		struct pfsync_clr clr;
2432 	} __packed r;
2433 
2434 	bzero(&r, sizeof(r));
2435 
2436 	r.subh.action = PFSYNC_ACT_CLR;
2437 	r.subh.count = htons(1);
2438 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
2439 
2440 	strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
2441 	r.clr.creatorid = creatorid;
2442 
2443 	pfsync_send_plus(&r, sizeof(r));
2444 }
2445 
2446 static enum pfsync_q_id
pfsync_sstate_to_qid(u_int8_t sync_state)2447 pfsync_sstate_to_qid(u_int8_t sync_state)
2448 {
2449 	struct pfsync_softc *sc = V_pfsyncif;
2450 
2451 	switch (sync_state) {
2452 		case PFSYNC_S_INS:
2453 			switch (sc->sc_version) {
2454 				case PFSYNC_MSG_VERSION_1301:
2455 					return PFSYNC_Q_INS_1301;
2456 				case PFSYNC_MSG_VERSION_1400:
2457 					return PFSYNC_Q_INS_1400;
2458 			}
2459 			break;
2460 		case PFSYNC_S_IACK:
2461 			return PFSYNC_Q_IACK;
2462 		case PFSYNC_S_UPD:
2463 			switch (sc->sc_version) {
2464 				case PFSYNC_MSG_VERSION_1301:
2465 					return PFSYNC_Q_UPD_1301;
2466 				case PFSYNC_MSG_VERSION_1400:
2467 					return PFSYNC_Q_UPD_1400;
2468 			}
2469 			break;
2470 		case PFSYNC_S_UPD_C:
2471 			return PFSYNC_Q_UPD_C;
2472 		case PFSYNC_S_DEL_C:
2473 			return PFSYNC_Q_DEL_C;
2474 		default:
2475 			panic("%s: Unsupported st->sync_state 0x%02x",
2476 			__func__, sync_state);
2477 	}
2478 
2479 	panic("%s: Unsupported pfsync_msg_version %d",
2480 	    __func__, sc->sc_version);
2481 }
2482 
2483 static void
pfsync_q_ins(struct pf_kstate * st,int sync_state,bool ref)2484 pfsync_q_ins(struct pf_kstate *st, int sync_state, bool ref)
2485 {
2486 	enum pfsync_q_id q = pfsync_sstate_to_qid(sync_state);
2487 	struct pfsync_softc *sc = V_pfsyncif;
2488 	size_t nlen = pfsync_qs[q].len;
2489 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2490 
2491 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2492 
2493 	KASSERT(st->sync_state == PFSYNC_S_NONE,
2494 		("%s: st->sync_state %u", __func__, st->sync_state));
2495 	KASSERT(b->b_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
2496 	    b->b_len));
2497 
2498 	if (TAILQ_EMPTY(&b->b_qs[q]))
2499 		nlen += sizeof(struct pfsync_subheader);
2500 
2501 	if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2502 		pfsync_sendout(1, b->b_id);
2503 
2504 		nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
2505 	}
2506 
2507 	b->b_len += nlen;
2508 	st->sync_state = pfsync_qid_sstate[q];
2509 	TAILQ_INSERT_TAIL(&b->b_qs[q], st, sync_list);
2510 	if (ref)
2511 		pf_ref_state(st);
2512 }
2513 
2514 static void
pfsync_q_del(struct pf_kstate * st,bool unref,struct pfsync_bucket * b)2515 pfsync_q_del(struct pf_kstate *st, bool unref, struct pfsync_bucket *b)
2516 {
2517 	enum pfsync_q_id q;
2518 
2519 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2520 	KASSERT(st->sync_state != PFSYNC_S_NONE,
2521 		("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2522 
2523 	q =  pfsync_sstate_to_qid(st->sync_state);
2524 	b->b_len -= pfsync_qs[q].len;
2525 	TAILQ_REMOVE(&b->b_qs[q], st, sync_list);
2526 	st->sync_state = PFSYNC_S_NONE;
2527 	if (unref)
2528 		pf_release_state(st);
2529 
2530 	if (TAILQ_EMPTY(&b->b_qs[q]))
2531 		b->b_len -= sizeof(struct pfsync_subheader);
2532 }
2533 
2534 static void
pfsync_bulk_start(void)2535 pfsync_bulk_start(void)
2536 {
2537 	struct pfsync_softc *sc = V_pfsyncif;
2538 
2539 	if (V_pf_status.debug >= PF_DEBUG_MISC)
2540 		printf("pfsync: received bulk update request\n");
2541 
2542 	PFSYNC_BLOCK(sc);
2543 
2544 	sc->sc_ureq_received = time_uptime;
2545 	sc->sc_bulk_hashid = 0;
2546 	sc->sc_bulk_stateid = 0;
2547 	pfsync_bulk_status(PFSYNC_BUS_START);
2548 	callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2549 	PFSYNC_BUNLOCK(sc);
2550 }
2551 
2552 static void
pfsync_bulk_update(void * arg)2553 pfsync_bulk_update(void *arg)
2554 {
2555 	struct pfsync_softc *sc = arg;
2556 	struct pf_kstate *s;
2557 	int i;
2558 
2559 	PFSYNC_BLOCK_ASSERT(sc);
2560 	CURVNET_SET(sc->sc_ifp->if_vnet);
2561 
2562 	/*
2563 	 * Start with last state from previous invocation.
2564 	 * It may had gone, in this case start from the
2565 	 * hash slot.
2566 	 */
2567 	s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2568 
2569 	if (s != NULL)
2570 		i = PF_IDHASH(s);
2571 	else
2572 		i = sc->sc_bulk_hashid;
2573 
2574 	for (; i <= V_pf_hashmask; i++) {
2575 		struct pf_idhash *ih = &V_pf_idhash[i];
2576 
2577 		if (s != NULL)
2578 			PF_HASHROW_ASSERT(ih);
2579 		else {
2580 			PF_HASHROW_LOCK(ih);
2581 			s = LIST_FIRST(&ih->states);
2582 		}
2583 
2584 		for (; s; s = LIST_NEXT(s, entry)) {
2585 			if (s->sync_state == PFSYNC_S_NONE &&
2586 			    s->timeout < PFTM_MAX &&
2587 			    s->pfsync_time <= sc->sc_ureq_received) {
2588 				if (pfsync_update_state_req(s)) {
2589 					/* We've filled a packet. */
2590 					sc->sc_bulk_hashid = i;
2591 					sc->sc_bulk_stateid = s->id;
2592 					sc->sc_bulk_creatorid = s->creatorid;
2593 					PF_HASHROW_UNLOCK(ih);
2594 					callout_reset(&sc->sc_bulk_tmo, 1,
2595 					    pfsync_bulk_update, sc);
2596 					goto full;
2597 				}
2598 			}
2599 		}
2600 		PF_HASHROW_UNLOCK(ih);
2601 	}
2602 
2603 	/* We're done. */
2604 	pfsync_bulk_status(PFSYNC_BUS_END);
2605 full:
2606 	CURVNET_RESTORE();
2607 }
2608 
2609 static void
pfsync_bulk_status(u_int8_t status)2610 pfsync_bulk_status(u_int8_t status)
2611 {
2612 	struct {
2613 		struct pfsync_subheader subh;
2614 		struct pfsync_bus bus;
2615 	} __packed r;
2616 
2617 	struct pfsync_softc *sc = V_pfsyncif;
2618 
2619 	bzero(&r, sizeof(r));
2620 
2621 	r.subh.action = PFSYNC_ACT_BUS;
2622 	r.subh.count = htons(1);
2623 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2624 
2625 	r.bus.creatorid = V_pf_status.hostid;
2626 	r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2627 	r.bus.status = status;
2628 
2629 	pfsync_send_plus(&r, sizeof(r));
2630 }
2631 
2632 static void
pfsync_bulk_fail(void * arg)2633 pfsync_bulk_fail(void *arg)
2634 {
2635 	struct pfsync_softc *sc = arg;
2636 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2637 
2638 	CURVNET_SET(sc->sc_ifp->if_vnet);
2639 
2640 	PFSYNC_BLOCK_ASSERT(sc);
2641 
2642 	if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2643 		/* Try again */
2644 		callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2645 		    pfsync_bulk_fail, V_pfsyncif);
2646 		PFSYNC_BUCKET_LOCK(b);
2647 		pfsync_request_update(0, 0);
2648 		PFSYNC_BUCKET_UNLOCK(b);
2649 	} else {
2650 		/* Pretend like the transfer was ok. */
2651 		sc->sc_ureq_sent = 0;
2652 		sc->sc_bulk_tries = 0;
2653 		PFSYNC_LOCK(sc);
2654 		if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2655 			(*carp_demote_adj_p)(-V_pfsync_carp_adj,
2656 			    "pfsync bulk fail");
2657 		sc->sc_flags |= PFSYNCF_OK;
2658 		PFSYNC_UNLOCK(sc);
2659 		if (V_pf_status.debug >= PF_DEBUG_MISC)
2660 			printf("pfsync: failed to receive bulk update\n");
2661 	}
2662 
2663 	CURVNET_RESTORE();
2664 }
2665 
2666 static void
pfsync_send_plus(void * plus,size_t pluslen)2667 pfsync_send_plus(void *plus, size_t pluslen)
2668 {
2669 	struct pfsync_softc *sc = V_pfsyncif;
2670 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2671 	uint8_t *newplus;
2672 
2673 	PFSYNC_BUCKET_LOCK(b);
2674 
2675 	if (b->b_len + pluslen > sc->sc_ifp->if_mtu)
2676 		pfsync_sendout(1, b->b_id);
2677 
2678 	newplus = malloc(pluslen + b->b_pluslen, M_PFSYNC, M_NOWAIT);
2679 	if (newplus == NULL)
2680 		goto out;
2681 
2682 	if (b->b_plus != NULL) {
2683 		memcpy(newplus, b->b_plus, b->b_pluslen);
2684 		free(b->b_plus, M_PFSYNC);
2685 	} else {
2686 		MPASS(b->b_pluslen == 0);
2687 	}
2688 	memcpy(newplus + b->b_pluslen, plus, pluslen);
2689 
2690 	b->b_plus = newplus;
2691 	b->b_pluslen += pluslen;
2692 	b->b_len += pluslen;
2693 
2694 	pfsync_sendout(1, b->b_id);
2695 
2696 out:
2697 	PFSYNC_BUCKET_UNLOCK(b);
2698 }
2699 
2700 static void
pfsync_timeout(void * arg)2701 pfsync_timeout(void *arg)
2702 {
2703 	struct pfsync_bucket *b = arg;
2704 
2705 	CURVNET_SET(b->b_sc->sc_ifp->if_vnet);
2706 	PFSYNC_BUCKET_LOCK(b);
2707 	pfsync_push(b);
2708 	PFSYNC_BUCKET_UNLOCK(b);
2709 	CURVNET_RESTORE();
2710 }
2711 
2712 static void
pfsync_push(struct pfsync_bucket * b)2713 pfsync_push(struct pfsync_bucket *b)
2714 {
2715 
2716 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2717 
2718 	b->b_flags |= PFSYNCF_BUCKET_PUSH;
2719 	swi_sched(V_pfsync_swi_cookie, 0);
2720 }
2721 
2722 static void
pfsync_push_all(struct pfsync_softc * sc)2723 pfsync_push_all(struct pfsync_softc *sc)
2724 {
2725 	int c;
2726 	struct pfsync_bucket *b;
2727 
2728 	for (c = 0; c < pfsync_buckets; c++) {
2729 		b = &sc->sc_buckets[c];
2730 
2731 		PFSYNC_BUCKET_LOCK(b);
2732 		pfsync_push(b);
2733 		PFSYNC_BUCKET_UNLOCK(b);
2734 	}
2735 }
2736 
2737 static void
pfsync_tx(struct pfsync_softc * sc,struct mbuf * m)2738 pfsync_tx(struct pfsync_softc *sc, struct mbuf *m)
2739 {
2740 	struct ip *ip;
2741 	int af, error = 0;
2742 
2743 	ip = mtod(m, struct ip *);
2744 	MPASS(ip->ip_v == IPVERSION || ip->ip_v == (IPV6_VERSION >> 4));
2745 
2746 	af = ip->ip_v == IPVERSION ? AF_INET : AF_INET6;
2747 
2748 	/*
2749 	 * We distinguish between a deferral packet and our
2750 	 * own pfsync packet based on M_SKIP_FIREWALL
2751 	 * flag. This is XXX.
2752 	 */
2753 	switch (af) {
2754 #ifdef INET
2755 	case AF_INET:
2756 		if (m->m_flags & M_SKIP_FIREWALL) {
2757 			error = ip_output(m, NULL, NULL, 0,
2758 			    NULL, NULL);
2759 		} else {
2760 			error = ip_output(m, NULL, NULL,
2761 			    IP_RAWOUTPUT, &sc->sc_imo, NULL);
2762 		}
2763 		break;
2764 #endif
2765 #ifdef INET6
2766 	case AF_INET6:
2767 		if (m->m_flags & M_SKIP_FIREWALL) {
2768 			error = ip6_output(m, NULL, NULL, 0,
2769 			    NULL, NULL, NULL);
2770 		} else {
2771 			error = ip6_output(m, NULL, NULL, 0,
2772 				&sc->sc_im6o, NULL, NULL);
2773 		}
2774 		break;
2775 #endif
2776 	}
2777 
2778 	if (error == 0)
2779 		V_pfsyncstats.pfsyncs_opackets++;
2780 	else
2781 		V_pfsyncstats.pfsyncs_oerrors++;
2782 
2783 }
2784 
2785 static void
pfsyncintr(void * arg)2786 pfsyncintr(void *arg)
2787 {
2788 	struct epoch_tracker et;
2789 	struct pfsync_softc *sc = arg;
2790 	struct pfsync_bucket *b;
2791 	struct mbuf *m, *n;
2792 	int c;
2793 
2794 	NET_EPOCH_ENTER(et);
2795 	CURVNET_SET(sc->sc_ifp->if_vnet);
2796 
2797 	for (c = 0; c < pfsync_buckets; c++) {
2798 		b = &sc->sc_buckets[c];
2799 
2800 		PFSYNC_BUCKET_LOCK(b);
2801 		if ((b->b_flags & PFSYNCF_BUCKET_PUSH) && b->b_len > PFSYNC_MINPKT) {
2802 			pfsync_sendout(0, b->b_id);
2803 			b->b_flags &= ~PFSYNCF_BUCKET_PUSH;
2804 		}
2805 		_IF_DEQUEUE_ALL(&b->b_snd, m);
2806 		PFSYNC_BUCKET_UNLOCK(b);
2807 
2808 		for (; m != NULL; m = n) {
2809 			n = m->m_nextpkt;
2810 			m->m_nextpkt = NULL;
2811 
2812 			pfsync_tx(sc, m);
2813 		}
2814 	}
2815 	CURVNET_RESTORE();
2816 	NET_EPOCH_EXIT(et);
2817 }
2818 
2819 static int
pfsync_multicast_setup(struct pfsync_softc * sc,struct ifnet * ifp,struct in_mfilter * imf,struct in6_mfilter * im6f)2820 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp,
2821     struct in_mfilter* imf, struct in6_mfilter* im6f)
2822 {
2823 #ifdef  INET
2824 	struct ip_moptions *imo = &sc->sc_imo;
2825 #endif
2826 #ifdef INET6
2827 	struct ip6_moptions *im6o = &sc->sc_im6o;
2828 	struct sockaddr_in6 *syncpeer_sa6 = NULL;
2829 #endif
2830 
2831 	if (!(ifp->if_flags & IFF_MULTICAST))
2832 		return (EADDRNOTAVAIL);
2833 
2834 	switch (sc->sc_sync_peer.ss_family) {
2835 #ifdef INET
2836 	case AF_INET:
2837 	{
2838 		int error;
2839 
2840 		ip_mfilter_init(&imo->imo_head);
2841 		imo->imo_multicast_vif = -1;
2842 		if ((error = in_joingroup(ifp,
2843 		    &((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr, NULL,
2844 		    &imf->imf_inm)) != 0)
2845 			return (error);
2846 
2847 		ip_mfilter_insert(&imo->imo_head, imf);
2848 		imo->imo_multicast_ifp = ifp;
2849 		imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2850 		imo->imo_multicast_loop = 0;
2851 		break;
2852 	}
2853 #endif
2854 #ifdef INET6
2855 	case AF_INET6:
2856 	{
2857 		int error;
2858 
2859 		syncpeer_sa6 = (struct sockaddr_in6 *)&sc->sc_sync_peer;
2860 		if ((error = in6_setscope(&syncpeer_sa6->sin6_addr, ifp, NULL)))
2861 			return (error);
2862 
2863 		ip6_mfilter_init(&im6o->im6o_head);
2864 		if ((error = in6_joingroup(ifp, &syncpeer_sa6->sin6_addr, NULL,
2865 		    &(im6f->im6f_in6m), 0)) != 0)
2866 			return (error);
2867 
2868 		ip6_mfilter_insert(&im6o->im6o_head, im6f);
2869 		im6o->im6o_multicast_ifp = ifp;
2870 		im6o->im6o_multicast_hlim = PFSYNC_DFLTTL;
2871 		im6o->im6o_multicast_loop = 0;
2872 		break;
2873 	}
2874 #endif
2875 	}
2876 
2877 	return (0);
2878 }
2879 
2880 static void
pfsync_multicast_cleanup(struct pfsync_softc * sc)2881 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2882 {
2883 #ifdef INET
2884 	struct ip_moptions *imo = &sc->sc_imo;
2885 	struct in_mfilter *imf;
2886 
2887 	while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
2888 		ip_mfilter_remove(&imo->imo_head, imf);
2889 		in_leavegroup(imf->imf_inm, NULL);
2890 		ip_mfilter_free(imf);
2891 	}
2892 	imo->imo_multicast_ifp = NULL;
2893 #endif
2894 
2895 #ifdef INET6
2896 	struct ip6_moptions *im6o = &sc->sc_im6o;
2897 	struct in6_mfilter *im6f;
2898 
2899 	while ((im6f = ip6_mfilter_first(&im6o->im6o_head)) != NULL) {
2900 		ip6_mfilter_remove(&im6o->im6o_head, im6f);
2901 		in6_leavegroup(im6f->im6f_in6m, NULL);
2902 		ip6_mfilter_free(im6f);
2903 	}
2904 	im6o->im6o_multicast_ifp = NULL;
2905 #endif
2906 }
2907 
2908 void
pfsync_detach_ifnet(struct ifnet * ifp)2909 pfsync_detach_ifnet(struct ifnet *ifp)
2910 {
2911 	struct pfsync_softc *sc = V_pfsyncif;
2912 
2913 	if (sc == NULL)
2914 		return;
2915 
2916 	PFSYNC_LOCK(sc);
2917 
2918 	if (sc->sc_sync_if == ifp) {
2919 		/* We don't need mutlicast cleanup here, because the interface
2920 		 * is going away. We do need to ensure we don't try to do
2921 		 * cleanup later.
2922 		 */
2923 		ip_mfilter_init(&sc->sc_imo.imo_head);
2924 		sc->sc_imo.imo_multicast_ifp = NULL;
2925 		sc->sc_im6o.im6o_multicast_ifp = NULL;
2926 		sc->sc_sync_if = NULL;
2927 	}
2928 
2929 	PFSYNC_UNLOCK(sc);
2930 }
2931 
2932 static int
pfsync_pfsyncreq_to_kstatus(struct pfsyncreq * pfsyncr,struct pfsync_kstatus * status)2933 pfsync_pfsyncreq_to_kstatus(struct pfsyncreq *pfsyncr, struct pfsync_kstatus *status)
2934 {
2935 	struct sockaddr_storage sa;
2936 	status->maxupdates = pfsyncr->pfsyncr_maxupdates;
2937 	status->flags = pfsyncr->pfsyncr_defer;
2938 
2939 	strlcpy(status->syncdev, pfsyncr->pfsyncr_syncdev, IFNAMSIZ);
2940 
2941 	memset(&sa, 0, sizeof(sa));
2942 	if (pfsyncr->pfsyncr_syncpeer.s_addr != 0) {
2943 		struct sockaddr_in *in = (struct sockaddr_in *)&sa;
2944 		in->sin_family = AF_INET;
2945 		in->sin_len = sizeof(*in);
2946 		in->sin_addr.s_addr = pfsyncr->pfsyncr_syncpeer.s_addr;
2947 	}
2948 	status->syncpeer = sa;
2949 
2950 	return 0;
2951 }
2952 
2953 static int
pfsync_kstatus_to_softc(struct pfsync_kstatus * status,struct pfsync_softc * sc)2954 pfsync_kstatus_to_softc(struct pfsync_kstatus *status, struct pfsync_softc *sc)
2955 {
2956 	struct ifnet *sifp;
2957 	struct in_mfilter *imf = NULL;
2958 	struct in6_mfilter *im6f = NULL;
2959 	int error;
2960 	int c;
2961 
2962 	if ((status->maxupdates < 0) || (status->maxupdates > 255))
2963 		return (EINVAL);
2964 
2965 	if (status->syncdev[0] == '\0')
2966 		sifp = NULL;
2967 	else if ((sifp = ifunit_ref(status->syncdev)) == NULL)
2968 		return (EINVAL);
2969 
2970 	switch (status->syncpeer.ss_family) {
2971 #ifdef INET
2972 	case AF_UNSPEC:
2973 	case AF_INET: {
2974 		struct sockaddr_in *status_sin;
2975 		status_sin = (struct sockaddr_in *)&(status->syncpeer);
2976 		if (sifp != NULL) {
2977 			if (status_sin->sin_addr.s_addr == 0 ||
2978 			    status_sin->sin_addr.s_addr ==
2979 			    htonl(INADDR_PFSYNC_GROUP)) {
2980 				status_sin->sin_family = AF_INET;
2981 				status_sin->sin_len = sizeof(*status_sin);
2982 				status_sin->sin_addr.s_addr =
2983 				    htonl(INADDR_PFSYNC_GROUP);
2984 			}
2985 
2986 			if (IN_MULTICAST(ntohl(status_sin->sin_addr.s_addr))) {
2987 				imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
2988 			}
2989 		}
2990 		break;
2991 	}
2992 #endif
2993 #ifdef INET6
2994 	case AF_INET6: {
2995 		struct sockaddr_in6 *status_sin6;
2996 		status_sin6 = (struct sockaddr_in6*)&(status->syncpeer);
2997 		if (sifp != NULL) {
2998 			if (IN6_IS_ADDR_UNSPECIFIED(&status_sin6->sin6_addr) ||
2999 			    IN6_ARE_ADDR_EQUAL(&status_sin6->sin6_addr,
3000 				&in6addr_linklocal_pfsync_group)) {
3001 				status_sin6->sin6_family = AF_INET6;
3002 				status_sin6->sin6_len = sizeof(*status_sin6);
3003 				status_sin6->sin6_addr =
3004 				    in6addr_linklocal_pfsync_group;
3005 			}
3006 
3007 			if (IN6_IS_ADDR_MULTICAST(&status_sin6->sin6_addr)) {
3008 				im6f = ip6_mfilter_alloc(M_WAITOK, 0, 0);
3009 			}
3010 		}
3011 		break;
3012 	}
3013 #endif
3014 	}
3015 
3016 	PFSYNC_LOCK(sc);
3017 
3018 	switch (status->version) {
3019 		case PFSYNC_MSG_VERSION_UNSPECIFIED:
3020 			sc->sc_version = PFSYNC_MSG_VERSION_DEFAULT;
3021 			break;
3022 		case PFSYNC_MSG_VERSION_1301:
3023 		case PFSYNC_MSG_VERSION_1400:
3024 			sc->sc_version = status->version;
3025 			break;
3026 		default:
3027 			PFSYNC_UNLOCK(sc);
3028 			return (EINVAL);
3029 	}
3030 
3031 	switch (status->syncpeer.ss_family) {
3032 	case AF_INET: {
3033 		struct sockaddr_in *status_sin = (struct sockaddr_in *)&(status->syncpeer);
3034 		struct sockaddr_in *sc_sin = (struct sockaddr_in *)&sc->sc_sync_peer;
3035 		sc_sin->sin_family = AF_INET;
3036 		sc_sin->sin_len = sizeof(*sc_sin);
3037 		if (status_sin->sin_addr.s_addr == 0) {
3038 			sc_sin->sin_addr.s_addr = htonl(INADDR_PFSYNC_GROUP);
3039 		} else {
3040 			sc_sin->sin_addr.s_addr = status_sin->sin_addr.s_addr;
3041 		}
3042 		break;
3043 	}
3044 	case AF_INET6: {
3045 		struct sockaddr_in6 *status_sin = (struct sockaddr_in6 *)&(status->syncpeer);
3046 		struct sockaddr_in6 *sc_sin = (struct sockaddr_in6 *)&sc->sc_sync_peer;
3047 		sc_sin->sin6_family = AF_INET6;
3048 		sc_sin->sin6_len = sizeof(*sc_sin);
3049 		if(IN6_IS_ADDR_UNSPECIFIED(&status_sin->sin6_addr)) {
3050 			sc_sin->sin6_addr = in6addr_linklocal_pfsync_group;
3051 		} else {
3052 			sc_sin->sin6_addr = status_sin->sin6_addr;
3053 		}
3054 		break;
3055 	}
3056 	}
3057 
3058 	sc->sc_maxupdates = status->maxupdates;
3059 	if (status->flags & PFSYNCF_DEFER) {
3060 		sc->sc_flags |= PFSYNCF_DEFER;
3061 		V_pfsync_defer_ptr = pfsync_defer;
3062 	} else {
3063 		sc->sc_flags &= ~PFSYNCF_DEFER;
3064 		V_pfsync_defer_ptr = NULL;
3065 	}
3066 
3067 	if (sifp == NULL) {
3068 		if (sc->sc_sync_if)
3069 			if_rele(sc->sc_sync_if);
3070 		sc->sc_sync_if = NULL;
3071 		pfsync_multicast_cleanup(sc);
3072 		PFSYNC_UNLOCK(sc);
3073 		return (0);
3074 	}
3075 
3076 	for (c = 0; c < pfsync_buckets; c++) {
3077 		PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
3078 		if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT &&
3079 		    (sifp->if_mtu < sc->sc_ifp->if_mtu ||
3080 			(sc->sc_sync_if != NULL &&
3081 			    sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
3082 			sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
3083 			pfsync_sendout(1, c);
3084 		PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
3085 	}
3086 
3087 	pfsync_multicast_cleanup(sc);
3088 
3089 	if (((sc->sc_sync_peer.ss_family == AF_INET) &&
3090 	    IN_MULTICAST(ntohl(((struct sockaddr_in *)
3091 	        &sc->sc_sync_peer)->sin_addr.s_addr))) ||
3092 	    ((sc->sc_sync_peer.ss_family == AF_INET6) &&
3093 	    IN6_IS_ADDR_MULTICAST(&((struct sockaddr_in6*)
3094 	        &sc->sc_sync_peer)->sin6_addr))) {
3095 		error = pfsync_multicast_setup(sc, sifp, imf, im6f);
3096 		if (error) {
3097 			if_rele(sifp);
3098 			PFSYNC_UNLOCK(sc);
3099 #ifdef INET
3100 			if (imf != NULL)
3101 				ip_mfilter_free(imf);
3102 #endif
3103 #ifdef INET6
3104 			if (im6f != NULL)
3105 				ip6_mfilter_free(im6f);
3106 #endif
3107 			return (error);
3108 		}
3109 	}
3110 	if (sc->sc_sync_if)
3111 		if_rele(sc->sc_sync_if);
3112 	sc->sc_sync_if = sifp;
3113 
3114 	switch (sc->sc_sync_peer.ss_family) {
3115 #ifdef INET
3116 	case AF_INET: {
3117 		struct ip *ip;
3118 		ip = &sc->sc_template.ipv4;
3119 		bzero(ip, sizeof(*ip));
3120 		ip->ip_v = IPVERSION;
3121 		ip->ip_hl = sizeof(sc->sc_template.ipv4) >> 2;
3122 		ip->ip_tos = IPTOS_LOWDELAY;
3123 		/* len and id are set later. */
3124 		ip->ip_off = htons(IP_DF);
3125 		ip->ip_ttl = PFSYNC_DFLTTL;
3126 		ip->ip_p = IPPROTO_PFSYNC;
3127 		ip->ip_src.s_addr = INADDR_ANY;
3128 		ip->ip_dst = ((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr;
3129 		break;
3130 	}
3131 #endif
3132 #ifdef INET6
3133 	case AF_INET6: {
3134 		struct ip6_hdr *ip6;
3135 		ip6 = &sc->sc_template.ipv6;
3136 		bzero(ip6, sizeof(*ip6));
3137 		ip6->ip6_vfc = IPV6_VERSION;
3138 		ip6->ip6_hlim = PFSYNC_DFLTTL;
3139 		ip6->ip6_nxt = IPPROTO_PFSYNC;
3140 		ip6->ip6_dst = ((struct sockaddr_in6 *)&sc->sc_sync_peer)->sin6_addr;
3141 
3142 		struct epoch_tracker et;
3143 		NET_EPOCH_ENTER(et);
3144 		in6_selectsrc_addr(if_getfib(sc->sc_sync_if), &ip6->ip6_dst, 0,
3145 		    sc->sc_sync_if, &ip6->ip6_src, NULL);
3146 		NET_EPOCH_EXIT(et);
3147 		break;
3148 	}
3149 #endif
3150 	}
3151 
3152 	/* Request a full state table update. */
3153 	if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
3154 		(*carp_demote_adj_p)(V_pfsync_carp_adj,
3155 		    "pfsync bulk start");
3156 	sc->sc_flags &= ~PFSYNCF_OK;
3157 	if (V_pf_status.debug >= PF_DEBUG_MISC)
3158 		printf("pfsync: requesting bulk update\n");
3159 	PFSYNC_UNLOCK(sc);
3160 	PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
3161 	pfsync_request_update(0, 0);
3162 	PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
3163 	PFSYNC_BLOCK(sc);
3164 	sc->sc_ureq_sent = time_uptime;
3165 	callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail, sc);
3166 	PFSYNC_BUNLOCK(sc);
3167 	return (0);
3168 }
3169 
3170 static void
pfsync_pointers_init(void)3171 pfsync_pointers_init(void)
3172 {
3173 
3174 	PF_RULES_WLOCK();
3175 	V_pfsync_state_import_ptr = pfsync_state_import;
3176 	V_pfsync_insert_state_ptr = pfsync_insert_state;
3177 	V_pfsync_update_state_ptr = pfsync_update_state;
3178 	V_pfsync_delete_state_ptr = pfsync_delete_state;
3179 	V_pfsync_clear_states_ptr = pfsync_clear_states;
3180 	V_pfsync_defer_ptr = pfsync_defer;
3181 	PF_RULES_WUNLOCK();
3182 }
3183 
3184 static void
pfsync_pointers_uninit(void)3185 pfsync_pointers_uninit(void)
3186 {
3187 
3188 	PF_RULES_WLOCK();
3189 	V_pfsync_state_import_ptr = NULL;
3190 	V_pfsync_insert_state_ptr = NULL;
3191 	V_pfsync_update_state_ptr = NULL;
3192 	V_pfsync_delete_state_ptr = NULL;
3193 	V_pfsync_clear_states_ptr = NULL;
3194 	V_pfsync_defer_ptr = NULL;
3195 	PF_RULES_WUNLOCK();
3196 }
3197 
3198 static void
vnet_pfsync_init(const void * unused __unused)3199 vnet_pfsync_init(const void *unused __unused)
3200 {
3201 	int error;
3202 
3203 	V_pfsync_cloner = if_clone_simple(pfsyncname,
3204 	    pfsync_clone_create, pfsync_clone_destroy, 1);
3205 	error = swi_add(&V_pfsync_swi_ie, pfsyncname, pfsyncintr, V_pfsyncif,
3206 	    SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
3207 	if (error) {
3208 		if_clone_detach(V_pfsync_cloner);
3209 		log(LOG_INFO, "swi_add() failed in %s\n", __func__);
3210 	}
3211 
3212 	pfsync_pointers_init();
3213 }
3214 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY,
3215     vnet_pfsync_init, NULL);
3216 
3217 static void
vnet_pfsync_uninit(const void * unused __unused)3218 vnet_pfsync_uninit(const void *unused __unused)
3219 {
3220 	int ret __diagused;
3221 
3222 	pfsync_pointers_uninit();
3223 
3224 	if_clone_detach(V_pfsync_cloner);
3225 	ret = swi_remove(V_pfsync_swi_cookie);
3226 	MPASS(ret == 0);
3227 	ret = intr_event_destroy(V_pfsync_swi_ie);
3228 	MPASS(ret == 0);
3229 }
3230 
3231 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_FOURTH,
3232     vnet_pfsync_uninit, NULL);
3233 
3234 static int
pfsync_init(void)3235 pfsync_init(void)
3236 {
3237 	int error;
3238 
3239 	pfsync_detach_ifnet_ptr = pfsync_detach_ifnet;
3240 
3241 #ifdef INET
3242 	error = ipproto_register(IPPROTO_PFSYNC, pfsync_input, NULL);
3243 	if (error)
3244 		return (error);
3245 #endif
3246 #ifdef INET6
3247 	error = ip6proto_register(IPPROTO_PFSYNC, pfsync6_input, NULL);
3248 	if (error) {
3249 		ipproto_unregister(IPPROTO_PFSYNC);
3250 		return (error);
3251 	}
3252 #endif
3253 
3254 	return (0);
3255 }
3256 
3257 static void
pfsync_uninit(void)3258 pfsync_uninit(void)
3259 {
3260 	pfsync_detach_ifnet_ptr = NULL;
3261 
3262 #ifdef INET
3263 	ipproto_unregister(IPPROTO_PFSYNC);
3264 #endif
3265 #ifdef INET6
3266 	ip6proto_unregister(IPPROTO_PFSYNC);
3267 #endif
3268 }
3269 
3270 static int
pfsync_modevent(module_t mod,int type,void * data)3271 pfsync_modevent(module_t mod, int type, void *data)
3272 {
3273 	int error = 0;
3274 
3275 	switch (type) {
3276 	case MOD_LOAD:
3277 		error = pfsync_init();
3278 		break;
3279 	case MOD_UNLOAD:
3280 		pfsync_uninit();
3281 		break;
3282 	default:
3283 		error = EINVAL;
3284 		break;
3285 	}
3286 
3287 	return (error);
3288 }
3289 
3290 static moduledata_t pfsync_mod = {
3291 	pfsyncname,
3292 	pfsync_modevent,
3293 	0
3294 };
3295 
3296 #define PFSYNC_MODVER 1
3297 
3298 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */
3299 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
3300 MODULE_VERSION(pfsync, PFSYNC_MODVER);
3301 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);
3302