]> CyberLeo.Net >> Repos - FreeBSD/releng/10.2.git/blob - sys/netpfil/pf/if_pfsync.c
- Copy stable/10@285827 to releng/10.2 in preparation for 10.2-RC1
[FreeBSD/releng/10.2.git] / sys / netpfil / pf / if_pfsync.c
1 /*-
2  * Copyright (c) 2002 Michael Shalayeff
3  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
19  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
20  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
21  * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
23  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
24  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
25  * THE POSSIBILITY OF SUCH DAMAGE.
26  */
27
28 /*-
29  * Copyright (c) 2009 David Gwynne <dlg@openbsd.org>
30  *
31  * Permission to use, copy, modify, and distribute this software for any
32  * purpose with or without fee is hereby granted, provided that the above
33  * copyright notice and this permission notice appear in all copies.
34  *
35  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
36  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
37  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
38  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
39  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
40  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
41  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
42  */
43
44 /*
45  * $OpenBSD: if_pfsync.c,v 1.110 2009/02/24 05:39:19 dlg Exp $
46  *
47  * Revisions picked from OpenBSD after revision 1.110 import:
48  * 1.119 - don't m_copydata() beyond the len of mbuf in pfsync_input()
49  * 1.118, 1.124, 1.148, 1.149, 1.151, 1.171 - fixes to bulk updates
50  * 1.120, 1.175 - use monotonic time_uptime
51  * 1.122 - reduce number of updates for non-TCP sessions
52  * 1.125, 1.127 - rewrite merge or stale processing
53  * 1.128 - cleanups
54  * 1.146 - bzero() mbuf before sparsely filling it with data
55  * 1.170 - SIOCSIFMTU checks
56  * 1.126, 1.142 - deferred packets processing
57  * 1.173 - correct expire time processing
58  */
59
60 #include <sys/cdefs.h>
61 __FBSDID("$FreeBSD$");
62
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/priv.h>
77 #include <sys/protosw.h>
78 #include <sys/socket.h>
79 #include <sys/sockio.h>
80 #include <sys/sysctl.h>
81
82 #include <net/bpf.h>
83 #include <net/if.h>
84 #include <net/if_clone.h>
85 #include <net/if_types.h>
86 #include <net/pfvar.h>
87 #include <net/if_pfsync.h>
88
89 #include <netinet/if_ether.h>
90 #include <netinet/in.h>
91 #include <netinet/in_var.h>
92 #include <netinet/ip.h>
93 #include <netinet/ip_carp.h>
94 #include <netinet/ip_var.h>
95 #include <netinet/tcp.h>
96 #include <netinet/tcp_fsm.h>
97 #include <netinet/tcp_seq.h>
98
99 #define PFSYNC_MINPKT ( \
100         sizeof(struct ip) + \
101         sizeof(struct pfsync_header) + \
102         sizeof(struct pfsync_subheader) )
103
104 struct pfsync_pkt {
105         struct ip *ip;
106         struct in_addr src;
107         u_int8_t flags;
108 };
109
110 static int      pfsync_upd_tcp(struct pf_state *, struct pfsync_state_peer *,
111                     struct pfsync_state_peer *);
112 static int      pfsync_in_clr(struct pfsync_pkt *, struct mbuf *, int, int);
113 static int      pfsync_in_ins(struct pfsync_pkt *, struct mbuf *, int, int);
114 static int      pfsync_in_iack(struct pfsync_pkt *, struct mbuf *, int, int);
115 static int      pfsync_in_upd(struct pfsync_pkt *, struct mbuf *, int, int);
116 static int      pfsync_in_upd_c(struct pfsync_pkt *, struct mbuf *, int, int);
117 static int      pfsync_in_ureq(struct pfsync_pkt *, struct mbuf *, int, int);
118 static int      pfsync_in_del(struct pfsync_pkt *, struct mbuf *, int, int);
119 static int      pfsync_in_del_c(struct pfsync_pkt *, struct mbuf *, int, int);
120 static int      pfsync_in_bus(struct pfsync_pkt *, struct mbuf *, int, int);
121 static int      pfsync_in_tdb(struct pfsync_pkt *, struct mbuf *, int, int);
122 static int      pfsync_in_eof(struct pfsync_pkt *, struct mbuf *, int, int);
123 static int      pfsync_in_error(struct pfsync_pkt *, struct mbuf *, int, int);
124
125 static int (*pfsync_acts[])(struct pfsync_pkt *, struct mbuf *, int, int) = {
126         pfsync_in_clr,                  /* PFSYNC_ACT_CLR */
127         pfsync_in_ins,                  /* PFSYNC_ACT_INS */
128         pfsync_in_iack,                 /* PFSYNC_ACT_INS_ACK */
129         pfsync_in_upd,                  /* PFSYNC_ACT_UPD */
130         pfsync_in_upd_c,                /* PFSYNC_ACT_UPD_C */
131         pfsync_in_ureq,                 /* PFSYNC_ACT_UPD_REQ */
132         pfsync_in_del,                  /* PFSYNC_ACT_DEL */
133         pfsync_in_del_c,                /* PFSYNC_ACT_DEL_C */
134         pfsync_in_error,                /* PFSYNC_ACT_INS_F */
135         pfsync_in_error,                /* PFSYNC_ACT_DEL_F */
136         pfsync_in_bus,                  /* PFSYNC_ACT_BUS */
137         pfsync_in_tdb,                  /* PFSYNC_ACT_TDB */
138         pfsync_in_eof                   /* PFSYNC_ACT_EOF */
139 };
140
141 struct pfsync_q {
142         void            (*write)(struct pf_state *, void *);
143         size_t          len;
144         u_int8_t        action;
145 };
146
147 /* we have one of these for every PFSYNC_S_ */
148 static void     pfsync_out_state(struct pf_state *, void *);
149 static void     pfsync_out_iack(struct pf_state *, void *);
150 static void     pfsync_out_upd_c(struct pf_state *, void *);
151 static void     pfsync_out_del(struct pf_state *, void *);
152
153 static struct pfsync_q pfsync_qs[] = {
154         { pfsync_out_state, sizeof(struct pfsync_state),   PFSYNC_ACT_INS },
155         { pfsync_out_iack,  sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK },
156         { pfsync_out_state, sizeof(struct pfsync_state),   PFSYNC_ACT_UPD },
157         { pfsync_out_upd_c, sizeof(struct pfsync_upd_c),   PFSYNC_ACT_UPD_C },
158         { pfsync_out_del,   sizeof(struct pfsync_del_c),   PFSYNC_ACT_DEL_C }
159 };
160
161 static void     pfsync_q_ins(struct pf_state *, int);
162 static void     pfsync_q_del(struct pf_state *);
163
164 static void     pfsync_update_state(struct pf_state *);
165
166 struct pfsync_upd_req_item {
167         TAILQ_ENTRY(pfsync_upd_req_item)        ur_entry;
168         struct pfsync_upd_req                   ur_msg;
169 };
170
171 struct pfsync_deferral {
172         struct pfsync_softc             *pd_sc;
173         TAILQ_ENTRY(pfsync_deferral)    pd_entry;
174         u_int                           pd_refs;
175         struct callout                  pd_tmo;
176
177         struct pf_state                 *pd_st;
178         struct mbuf                     *pd_m;
179 };
180
181 struct pfsync_softc {
182         /* Configuration */
183         struct ifnet            *sc_ifp;
184         struct ifnet            *sc_sync_if;
185         struct ip_moptions      sc_imo;
186         struct in_addr          sc_sync_peer;
187         uint32_t                sc_flags;
188 #define PFSYNCF_OK              0x00000001
189 #define PFSYNCF_DEFER           0x00000002
190 #define PFSYNCF_PUSH            0x00000004
191         uint8_t                 sc_maxupdates;
192         struct ip               sc_template;
193         struct callout          sc_tmo;
194         struct mtx              sc_mtx;
195
196         /* Queued data */
197         size_t                  sc_len;
198         TAILQ_HEAD(, pf_state)                  sc_qs[PFSYNC_S_COUNT];
199         TAILQ_HEAD(, pfsync_upd_req_item)       sc_upd_req_list;
200         TAILQ_HEAD(, pfsync_deferral)           sc_deferrals;
201         u_int                   sc_deferred;
202         void                    *sc_plus;
203         size_t                  sc_pluslen;
204
205         /* Bulk update info */
206         struct mtx              sc_bulk_mtx;
207         uint32_t                sc_ureq_sent;
208         int                     sc_bulk_tries;
209         uint32_t                sc_ureq_received;
210         int                     sc_bulk_hashid;
211         uint64_t                sc_bulk_stateid;
212         uint32_t                sc_bulk_creatorid;
213         struct callout          sc_bulk_tmo;
214         struct callout          sc_bulkfail_tmo;
215 };
216
217 #define PFSYNC_LOCK(sc)         mtx_lock(&(sc)->sc_mtx)
218 #define PFSYNC_UNLOCK(sc)       mtx_unlock(&(sc)->sc_mtx)
219 #define PFSYNC_LOCK_ASSERT(sc)  mtx_assert(&(sc)->sc_mtx, MA_OWNED)
220
221 #define PFSYNC_BLOCK(sc)        mtx_lock(&(sc)->sc_bulk_mtx)
222 #define PFSYNC_BUNLOCK(sc)      mtx_unlock(&(sc)->sc_bulk_mtx)
223 #define PFSYNC_BLOCK_ASSERT(sc) mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED)
224
225 static const char pfsyncname[] = "pfsync";
226 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data");
227 static VNET_DEFINE(struct pfsync_softc  *, pfsyncif) = NULL;
228 #define V_pfsyncif              VNET(pfsyncif)
229 static VNET_DEFINE(void *, pfsync_swi_cookie) = NULL;
230 #define V_pfsync_swi_cookie     VNET(pfsync_swi_cookie)
231 static VNET_DEFINE(struct pfsyncstats, pfsyncstats);
232 #define V_pfsyncstats           VNET(pfsyncstats)
233 static VNET_DEFINE(int, pfsync_carp_adj) = CARP_MAXSKEW;
234 #define V_pfsync_carp_adj       VNET(pfsync_carp_adj)
235
236 static void     pfsync_timeout(void *);
237 static void     pfsync_push(struct pfsync_softc *);
238 static void     pfsyncintr(void *);
239 static int      pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *,
240                     void *);
241 static void     pfsync_multicast_cleanup(struct pfsync_softc *);
242 static void     pfsync_pointers_init(void);
243 static void     pfsync_pointers_uninit(void);
244 static int      pfsync_init(void);
245 static void     pfsync_uninit(void);
246
247 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW, 0, "PFSYNC");
248 SYSCTL_VNET_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_RW,
249     &VNET_NAME(pfsyncstats), pfsyncstats,
250     "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)");
251 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_RW,
252     &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment");
253
254 static int      pfsync_clone_create(struct if_clone *, int, caddr_t);
255 static void     pfsync_clone_destroy(struct ifnet *);
256 static int      pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
257                     struct pf_state_peer *);
258 static int      pfsyncoutput(struct ifnet *, struct mbuf *,
259                     const struct sockaddr *, struct route *);
260 static int      pfsyncioctl(struct ifnet *, u_long, caddr_t);
261
262 static int      pfsync_defer(struct pf_state *, struct mbuf *);
263 static void     pfsync_undefer(struct pfsync_deferral *, int);
264 static void     pfsync_undefer_state(struct pf_state *, int);
265 static void     pfsync_defer_tmo(void *);
266
267 static void     pfsync_request_update(u_int32_t, u_int64_t);
268 static void     pfsync_update_state_req(struct pf_state *);
269
270 static void     pfsync_drop(struct pfsync_softc *);
271 static void     pfsync_sendout(int);
272 static void     pfsync_send_plus(void *, size_t);
273
274 static void     pfsync_bulk_start(void);
275 static void     pfsync_bulk_status(u_int8_t);
276 static void     pfsync_bulk_update(void *);
277 static void     pfsync_bulk_fail(void *);
278
279 #ifdef IPSEC
280 static void     pfsync_update_net_tdb(struct pfsync_tdb *);
281 #endif
282
283 #define PFSYNC_MAX_BULKTRIES    12
284
285 VNET_DEFINE(struct if_clone *, pfsync_cloner);
286 #define V_pfsync_cloner VNET(pfsync_cloner)
287
288 static int
289 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param)
290 {
291         struct pfsync_softc *sc;
292         struct ifnet *ifp;
293         int q;
294
295         if (unit != 0)
296                 return (EINVAL);
297
298         sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO);
299         sc->sc_flags |= PFSYNCF_OK;
300
301         for (q = 0; q < PFSYNC_S_COUNT; q++)
302                 TAILQ_INIT(&sc->sc_qs[q]);
303
304         TAILQ_INIT(&sc->sc_upd_req_list);
305         TAILQ_INIT(&sc->sc_deferrals);
306
307         sc->sc_len = PFSYNC_MINPKT;
308         sc->sc_maxupdates = 128;
309
310         ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC);
311         if (ifp == NULL) {
312                 free(sc, M_PFSYNC);
313                 return (ENOSPC);
314         }
315         if_initname(ifp, pfsyncname, unit);
316         ifp->if_softc = sc;
317         ifp->if_ioctl = pfsyncioctl;
318         ifp->if_output = pfsyncoutput;
319         ifp->if_type = IFT_PFSYNC;
320         ifp->if_snd.ifq_maxlen = ifqmaxlen;
321         ifp->if_hdrlen = sizeof(struct pfsync_header);
322         ifp->if_mtu = ETHERMTU;
323         mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF);
324         mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF);
325         callout_init(&sc->sc_tmo, CALLOUT_MPSAFE);
326         callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0);
327         callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0);
328
329         if_attach(ifp);
330
331         bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
332
333         V_pfsyncif = sc;
334
335         return (0);
336 }
337
338 static void
339 pfsync_clone_destroy(struct ifnet *ifp)
340 {
341         struct pfsync_softc *sc = ifp->if_softc;
342
343         /*
344          * At this stage, everything should have already been
345          * cleared by pfsync_uninit(), and we have only to
346          * drain callouts.
347          */
348         while (sc->sc_deferred > 0) {
349                 struct pfsync_deferral *pd = TAILQ_FIRST(&sc->sc_deferrals);
350
351                 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
352                 sc->sc_deferred--;
353                 if (callout_stop(&pd->pd_tmo)) {
354                         pf_release_state(pd->pd_st);
355                         m_freem(pd->pd_m);
356                         free(pd, M_PFSYNC);
357                 } else {
358                         pd->pd_refs++;
359                         callout_drain(&pd->pd_tmo);
360                         free(pd, M_PFSYNC);
361                 }
362         }
363
364         callout_drain(&sc->sc_tmo);
365         callout_drain(&sc->sc_bulkfail_tmo);
366         callout_drain(&sc->sc_bulk_tmo);
367
368         if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
369                 (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
370         bpfdetach(ifp);
371         if_detach(ifp);
372
373         pfsync_drop(sc);
374
375         if_free(ifp);
376         if (sc->sc_imo.imo_membership)
377                 pfsync_multicast_cleanup(sc);
378         mtx_destroy(&sc->sc_mtx);
379         mtx_destroy(&sc->sc_bulk_mtx);
380         free(sc, M_PFSYNC);
381
382         V_pfsyncif = NULL;
383 }
384
385 static int
386 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
387     struct pf_state_peer *d)
388 {
389         if (s->scrub.scrub_flag && d->scrub == NULL) {
390                 d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO);
391                 if (d->scrub == NULL)
392                         return (ENOMEM);
393         }
394
395         return (0);
396 }
397
398
399 static int
400 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags)
401 {
402         struct pfsync_softc *sc = V_pfsyncif;
403 #ifndef __NO_STRICT_ALIGNMENT
404         struct pfsync_state_key key[2];
405 #endif
406         struct pfsync_state_key *kw, *ks;
407         struct pf_state *st = NULL;
408         struct pf_state_key *skw = NULL, *sks = NULL;
409         struct pf_rule *r = NULL;
410         struct pfi_kif  *kif;
411         int error;
412
413         PF_RULES_RASSERT();
414
415         if (sp->creatorid == 0) {
416                 if (V_pf_status.debug >= PF_DEBUG_MISC)
417                         printf("%s: invalid creator id: %08x\n", __func__,
418                             ntohl(sp->creatorid));
419                 return (EINVAL);
420         }
421
422         if ((kif = pfi_kif_find(sp->ifname)) == NULL) {
423                 if (V_pf_status.debug >= PF_DEBUG_MISC)
424                         printf("%s: unknown interface: %s\n", __func__,
425                             sp->ifname);
426                 if (flags & PFSYNC_SI_IOCTL)
427                         return (EINVAL);
428                 return (0);     /* skip this state */
429         }
430
431         /*
432          * If the ruleset checksums match or the state is coming from the ioctl,
433          * it's safe to associate the state with the rule of that number.
434          */
435         if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) &&
436             (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) <
437             pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
438                 r = pf_main_ruleset.rules[
439                     PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
440         else
441                 r = &V_pf_default_rule;
442
443         if ((r->max_states &&
444             counter_u64_fetch(r->states_cur) >= r->max_states))
445                 goto cleanup;
446
447         /*
448          * XXXGL: consider M_WAITOK in ioctl path after.
449          */
450         if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL)
451                 goto cleanup;
452
453         if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
454                 goto cleanup;
455
456 #ifndef __NO_STRICT_ALIGNMENT
457         bcopy(&sp->key, key, sizeof(struct pfsync_state_key) * 2);
458         kw = &key[PF_SK_WIRE];
459         ks = &key[PF_SK_STACK];
460 #else
461         kw = &sp->key[PF_SK_WIRE];
462         ks = &sp->key[PF_SK_STACK];
463 #endif
464
465         if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->af) ||
466             PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->af) ||
467             kw->port[0] != ks->port[0] ||
468             kw->port[1] != ks->port[1]) {
469                 sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
470                 if (sks == NULL)
471                         goto cleanup;
472         } else
473                 sks = skw;
474
475         /* allocate memory for scrub info */
476         if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
477             pfsync_alloc_scrub_memory(&sp->dst, &st->dst))
478                 goto cleanup;
479
480         /* Copy to state key(s). */
481         skw->addr[0] = kw->addr[0];
482         skw->addr[1] = kw->addr[1];
483         skw->port[0] = kw->port[0];
484         skw->port[1] = kw->port[1];
485         skw->proto = sp->proto;
486         skw->af = sp->af;
487         if (sks != skw) {
488                 sks->addr[0] = ks->addr[0];
489                 sks->addr[1] = ks->addr[1];
490                 sks->port[0] = ks->port[0];
491                 sks->port[1] = ks->port[1];
492                 sks->proto = sp->proto;
493                 sks->af = sp->af;
494         }
495
496         /* copy to state */
497         bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
498         st->creation = time_uptime - ntohl(sp->creation);
499         st->expire = time_uptime;
500         if (sp->expire) {
501                 uint32_t timeout;
502
503                 timeout = r->timeout[sp->timeout];
504                 if (!timeout)
505                         timeout = V_pf_default_rule.timeout[sp->timeout];
506
507                 /* sp->expire may have been adaptively scaled by export. */
508                 st->expire -= timeout - ntohl(sp->expire);
509         }
510
511         st->direction = sp->direction;
512         st->log = sp->log;
513         st->timeout = sp->timeout;
514         st->state_flags = sp->state_flags;
515
516         st->id = sp->id;
517         st->creatorid = sp->creatorid;
518         pf_state_peer_ntoh(&sp->src, &st->src);
519         pf_state_peer_ntoh(&sp->dst, &st->dst);
520
521         st->rule.ptr = r;
522         st->nat_rule.ptr = NULL;
523         st->anchor.ptr = NULL;
524         st->rt_kif = NULL;
525
526         st->pfsync_time = time_uptime;
527         st->sync_state = PFSYNC_S_NONE;
528
529         if (!(flags & PFSYNC_SI_IOCTL))
530                 st->state_flags |= PFSTATE_NOSYNC;
531
532         if ((error = pf_state_insert(kif, skw, sks, st)) != 0)
533                 goto cleanup_state;
534
535         /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
536         counter_u64_add(r->states_cur, 1);
537         counter_u64_add(r->states_tot, 1);
538
539         if (!(flags & PFSYNC_SI_IOCTL)) {
540                 st->state_flags &= ~PFSTATE_NOSYNC;
541                 if (st->state_flags & PFSTATE_ACK) {
542                         pfsync_q_ins(st, PFSYNC_S_IACK);
543                         pfsync_push(sc);
544                 }
545         }
546         st->state_flags &= ~PFSTATE_ACK;
547         PF_STATE_UNLOCK(st);
548
549         return (0);
550
551 cleanup:
552         error = ENOMEM;
553         if (skw == sks)
554                 sks = NULL;
555         if (skw != NULL)
556                 uma_zfree(V_pf_state_key_z, skw);
557         if (sks != NULL)
558                 uma_zfree(V_pf_state_key_z, sks);
559
560 cleanup_state:  /* pf_state_insert() frees the state keys. */
561         if (st) {
562                 if (st->dst.scrub)
563                         uma_zfree(V_pf_state_scrub_z, st->dst.scrub);
564                 if (st->src.scrub)
565                         uma_zfree(V_pf_state_scrub_z, st->src.scrub);
566                 uma_zfree(V_pf_state_z, st);
567         }
568         return (error);
569 }
570
571 static void
572 pfsync_input(struct mbuf *m, __unused int off)
573 {
574         struct pfsync_softc *sc = V_pfsyncif;
575         struct pfsync_pkt pkt;
576         struct ip *ip = mtod(m, struct ip *);
577         struct pfsync_header *ph;
578         struct pfsync_subheader subh;
579
580         int offset, len;
581         int rv;
582         uint16_t count;
583
584         V_pfsyncstats.pfsyncs_ipackets++;
585
586         /* Verify that we have a sync interface configured. */
587         if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
588             (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
589                 goto done;
590
591         /* verify that the packet came in on the right interface */
592         if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
593                 V_pfsyncstats.pfsyncs_badif++;
594                 goto done;
595         }
596
597         sc->sc_ifp->if_ipackets++;
598         sc->sc_ifp->if_ibytes += m->m_pkthdr.len;
599         /* verify that the IP TTL is 255. */
600         if (ip->ip_ttl != PFSYNC_DFLTTL) {
601                 V_pfsyncstats.pfsyncs_badttl++;
602                 goto done;
603         }
604
605         offset = ip->ip_hl << 2;
606         if (m->m_pkthdr.len < offset + sizeof(*ph)) {
607                 V_pfsyncstats.pfsyncs_hdrops++;
608                 goto done;
609         }
610
611         if (offset + sizeof(*ph) > m->m_len) {
612                 if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
613                         V_pfsyncstats.pfsyncs_hdrops++;
614                         return;
615                 }
616                 ip = mtod(m, struct ip *);
617         }
618         ph = (struct pfsync_header *)((char *)ip + offset);
619
620         /* verify the version */
621         if (ph->version != PFSYNC_VERSION) {
622                 V_pfsyncstats.pfsyncs_badver++;
623                 goto done;
624         }
625
626         len = ntohs(ph->len) + offset;
627         if (m->m_pkthdr.len < len) {
628                 V_pfsyncstats.pfsyncs_badlen++;
629                 goto done;
630         }
631
632         /* Cheaper to grab this now than having to mess with mbufs later */
633         pkt.ip = ip;
634         pkt.src = ip->ip_src;
635         pkt.flags = 0;
636
637         /*
638          * Trusting pf_chksum during packet processing, as well as seeking
639          * in interface name tree, require holding PF_RULES_RLOCK().
640          */
641         PF_RULES_RLOCK();
642         if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
643                 pkt.flags |= PFSYNC_SI_CKSUM;
644
645         offset += sizeof(*ph);
646         while (offset <= len - sizeof(subh)) {
647                 m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
648                 offset += sizeof(subh);
649
650                 if (subh.action >= PFSYNC_ACT_MAX) {
651                         V_pfsyncstats.pfsyncs_badact++;
652                         PF_RULES_RUNLOCK();
653                         goto done;
654                 }
655
656                 count = ntohs(subh.count);
657                 V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
658                 rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count);
659                 if (rv == -1) {
660                         PF_RULES_RUNLOCK();
661                         return;
662                 }
663
664                 offset += rv;
665         }
666         PF_RULES_RUNLOCK();
667
668 done:
669         m_freem(m);
670 }
671
672 static int
673 pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
674 {
675         struct pfsync_clr *clr;
676         struct mbuf *mp;
677         int len = sizeof(*clr) * count;
678         int i, offp;
679         u_int32_t creatorid;
680
681         mp = m_pulldown(m, offset, len, &offp);
682         if (mp == NULL) {
683                 V_pfsyncstats.pfsyncs_badlen++;
684                 return (-1);
685         }
686         clr = (struct pfsync_clr *)(mp->m_data + offp);
687
688         for (i = 0; i < count; i++) {
689                 creatorid = clr[i].creatorid;
690
691                 if (clr[i].ifname[0] != '\0' &&
692                     pfi_kif_find(clr[i].ifname) == NULL)
693                         continue;
694
695                 for (int i = 0; i <= pf_hashmask; i++) {
696                         struct pf_idhash *ih = &V_pf_idhash[i];
697                         struct pf_state *s;
698 relock:
699                         PF_HASHROW_LOCK(ih);
700                         LIST_FOREACH(s, &ih->states, entry) {
701                                 if (s->creatorid == creatorid) {
702                                         s->state_flags |= PFSTATE_NOSYNC;
703                                         pf_unlink_state(s, PF_ENTER_LOCKED);
704                                         goto relock;
705                                 }
706                         }
707                         PF_HASHROW_UNLOCK(ih);
708                 }
709         }
710
711         return (len);
712 }
713
714 static int
715 pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
716 {
717         struct mbuf *mp;
718         struct pfsync_state *sa, *sp;
719         int len = sizeof(*sp) * count;
720         int i, offp;
721
722         mp = m_pulldown(m, offset, len, &offp);
723         if (mp == NULL) {
724                 V_pfsyncstats.pfsyncs_badlen++;
725                 return (-1);
726         }
727         sa = (struct pfsync_state *)(mp->m_data + offp);
728
729         for (i = 0; i < count; i++) {
730                 sp = &sa[i];
731
732                 /* Check for invalid values. */
733                 if (sp->timeout >= PFTM_MAX ||
734                     sp->src.state > PF_TCPS_PROXY_DST ||
735                     sp->dst.state > PF_TCPS_PROXY_DST ||
736                     sp->direction > PF_OUT ||
737                     (sp->af != AF_INET && sp->af != AF_INET6)) {
738                         if (V_pf_status.debug >= PF_DEBUG_MISC)
739                                 printf("%s: invalid value\n", __func__);
740                         V_pfsyncstats.pfsyncs_badval++;
741                         continue;
742                 }
743
744                 if (pfsync_state_import(sp, pkt->flags) == ENOMEM)
745                         /* Drop out, but process the rest of the actions. */
746                         break;
747         }
748
749         return (len);
750 }
751
752 static int
753 pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
754 {
755         struct pfsync_ins_ack *ia, *iaa;
756         struct pf_state *st;
757
758         struct mbuf *mp;
759         int len = count * sizeof(*ia);
760         int offp, i;
761
762         mp = m_pulldown(m, offset, len, &offp);
763         if (mp == NULL) {
764                 V_pfsyncstats.pfsyncs_badlen++;
765                 return (-1);
766         }
767         iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
768
769         for (i = 0; i < count; i++) {
770                 ia = &iaa[i];
771
772                 st = pf_find_state_byid(ia->id, ia->creatorid);
773                 if (st == NULL)
774                         continue;
775
776                 if (st->state_flags & PFSTATE_ACK) {
777                         PFSYNC_LOCK(V_pfsyncif);
778                         pfsync_undefer_state(st, 0);
779                         PFSYNC_UNLOCK(V_pfsyncif);
780                 }
781                 PF_STATE_UNLOCK(st);
782         }
783         /*
784          * XXX this is not yet implemented, but we know the size of the
785          * message so we can skip it.
786          */
787
788         return (count * sizeof(struct pfsync_ins_ack));
789 }
790
791 static int
792 pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src,
793     struct pfsync_state_peer *dst)
794 {
795         int sync = 0;
796
797         PF_STATE_LOCK_ASSERT(st);
798
799         /*
800          * The state should never go backwards except
801          * for syn-proxy states.  Neither should the
802          * sequence window slide backwards.
803          */
804         if ((st->src.state > src->state &&
805             (st->src.state < PF_TCPS_PROXY_SRC ||
806             src->state >= PF_TCPS_PROXY_SRC)) ||
807
808             (st->src.state == src->state &&
809             SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
810                 sync++;
811         else
812                 pf_state_peer_ntoh(src, &st->src);
813
814         if ((st->dst.state > dst->state) ||
815
816             (st->dst.state >= TCPS_SYN_SENT &&
817             SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
818                 sync++;
819         else
820                 pf_state_peer_ntoh(dst, &st->dst);
821
822         return (sync);
823 }
824
825 static int
826 pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
827 {
828         struct pfsync_softc *sc = V_pfsyncif;
829         struct pfsync_state *sa, *sp;
830         struct pf_state *st;
831         int sync;
832
833         struct mbuf *mp;
834         int len = count * sizeof(*sp);
835         int offp, i;
836
837         mp = m_pulldown(m, offset, len, &offp);
838         if (mp == NULL) {
839                 V_pfsyncstats.pfsyncs_badlen++;
840                 return (-1);
841         }
842         sa = (struct pfsync_state *)(mp->m_data + offp);
843
844         for (i = 0; i < count; i++) {
845                 sp = &sa[i];
846
847                 /* check for invalid values */
848                 if (sp->timeout >= PFTM_MAX ||
849                     sp->src.state > PF_TCPS_PROXY_DST ||
850                     sp->dst.state > PF_TCPS_PROXY_DST) {
851                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
852                                 printf("pfsync_input: PFSYNC_ACT_UPD: "
853                                     "invalid value\n");
854                         }
855                         V_pfsyncstats.pfsyncs_badval++;
856                         continue;
857                 }
858
859                 st = pf_find_state_byid(sp->id, sp->creatorid);
860                 if (st == NULL) {
861                         /* insert the update */
862                         if (pfsync_state_import(sp, 0))
863                                 V_pfsyncstats.pfsyncs_badstate++;
864                         continue;
865                 }
866
867                 if (st->state_flags & PFSTATE_ACK) {
868                         PFSYNC_LOCK(sc);
869                         pfsync_undefer_state(st, 1);
870                         PFSYNC_UNLOCK(sc);
871                 }
872
873                 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
874                         sync = pfsync_upd_tcp(st, &sp->src, &sp->dst);
875                 else {
876                         sync = 0;
877
878                         /*
879                          * Non-TCP protocol state machine always go
880                          * forwards
881                          */
882                         if (st->src.state > sp->src.state)
883                                 sync++;
884                         else
885                                 pf_state_peer_ntoh(&sp->src, &st->src);
886                         if (st->dst.state > sp->dst.state)
887                                 sync++;
888                         else
889                                 pf_state_peer_ntoh(&sp->dst, &st->dst);
890                 }
891                 if (sync < 2) {
892                         pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
893                         pf_state_peer_ntoh(&sp->dst, &st->dst);
894                         st->expire = time_uptime;
895                         st->timeout = sp->timeout;
896                 }
897                 st->pfsync_time = time_uptime;
898
899                 if (sync) {
900                         V_pfsyncstats.pfsyncs_stale++;
901
902                         pfsync_update_state(st);
903                         PF_STATE_UNLOCK(st);
904                         PFSYNC_LOCK(sc);
905                         pfsync_push(sc);
906                         PFSYNC_UNLOCK(sc);
907                         continue;
908                 }
909                 PF_STATE_UNLOCK(st);
910         }
911
912         return (len);
913 }
914
915 static int
916 pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
917 {
918         struct pfsync_softc *sc = V_pfsyncif;
919         struct pfsync_upd_c *ua, *up;
920         struct pf_state *st;
921         int len = count * sizeof(*up);
922         int sync;
923         struct mbuf *mp;
924         int offp, i;
925
926         mp = m_pulldown(m, offset, len, &offp);
927         if (mp == NULL) {
928                 V_pfsyncstats.pfsyncs_badlen++;
929                 return (-1);
930         }
931         ua = (struct pfsync_upd_c *)(mp->m_data + offp);
932
933         for (i = 0; i < count; i++) {
934                 up = &ua[i];
935
936                 /* check for invalid values */
937                 if (up->timeout >= PFTM_MAX ||
938                     up->src.state > PF_TCPS_PROXY_DST ||
939                     up->dst.state > PF_TCPS_PROXY_DST) {
940                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
941                                 printf("pfsync_input: "
942                                     "PFSYNC_ACT_UPD_C: "
943                                     "invalid value\n");
944                         }
945                         V_pfsyncstats.pfsyncs_badval++;
946                         continue;
947                 }
948
949                 st = pf_find_state_byid(up->id, up->creatorid);
950                 if (st == NULL) {
951                         /* We don't have this state. Ask for it. */
952                         PFSYNC_LOCK(sc);
953                         pfsync_request_update(up->creatorid, up->id);
954                         PFSYNC_UNLOCK(sc);
955                         continue;
956                 }
957
958                 if (st->state_flags & PFSTATE_ACK) {
959                         PFSYNC_LOCK(sc);
960                         pfsync_undefer_state(st, 1);
961                         PFSYNC_UNLOCK(sc);
962                 }
963
964                 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
965                         sync = pfsync_upd_tcp(st, &up->src, &up->dst);
966                 else {
967                         sync = 0;
968
969                         /*
970                          * Non-TCP protocol state machine always go
971                          * forwards
972                          */
973                         if (st->src.state > up->src.state)
974                                 sync++;
975                         else
976                                 pf_state_peer_ntoh(&up->src, &st->src);
977                         if (st->dst.state > up->dst.state)
978                                 sync++;
979                         else
980                                 pf_state_peer_ntoh(&up->dst, &st->dst);
981                 }
982                 if (sync < 2) {
983                         pfsync_alloc_scrub_memory(&up->dst, &st->dst);
984                         pf_state_peer_ntoh(&up->dst, &st->dst);
985                         st->expire = time_uptime;
986                         st->timeout = up->timeout;
987                 }
988                 st->pfsync_time = time_uptime;
989
990                 if (sync) {
991                         V_pfsyncstats.pfsyncs_stale++;
992
993                         pfsync_update_state(st);
994                         PF_STATE_UNLOCK(st);
995                         PFSYNC_LOCK(sc);
996                         pfsync_push(sc);
997                         PFSYNC_UNLOCK(sc);
998                         continue;
999                 }
1000                 PF_STATE_UNLOCK(st);
1001         }
1002
1003         return (len);
1004 }
1005
1006 static int
1007 pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1008 {
1009         struct pfsync_upd_req *ur, *ura;
1010         struct mbuf *mp;
1011         int len = count * sizeof(*ur);
1012         int i, offp;
1013
1014         struct pf_state *st;
1015
1016         mp = m_pulldown(m, offset, len, &offp);
1017         if (mp == NULL) {
1018                 V_pfsyncstats.pfsyncs_badlen++;
1019                 return (-1);
1020         }
1021         ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1022
1023         for (i = 0; i < count; i++) {
1024                 ur = &ura[i];
1025
1026                 if (ur->id == 0 && ur->creatorid == 0)
1027                         pfsync_bulk_start();
1028                 else {
1029                         st = pf_find_state_byid(ur->id, ur->creatorid);
1030                         if (st == NULL) {
1031                                 V_pfsyncstats.pfsyncs_badstate++;
1032                                 continue;
1033                         }
1034                         if (st->state_flags & PFSTATE_NOSYNC) {
1035                                 PF_STATE_UNLOCK(st);
1036                                 continue;
1037                         }
1038
1039                         pfsync_update_state_req(st);
1040                         PF_STATE_UNLOCK(st);
1041                 }
1042         }
1043
1044         return (len);
1045 }
1046
1047 static int
1048 pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1049 {
1050         struct mbuf *mp;
1051         struct pfsync_state *sa, *sp;
1052         struct pf_state *st;
1053         int len = count * sizeof(*sp);
1054         int offp, i;
1055
1056         mp = m_pulldown(m, offset, len, &offp);
1057         if (mp == NULL) {
1058                 V_pfsyncstats.pfsyncs_badlen++;
1059                 return (-1);
1060         }
1061         sa = (struct pfsync_state *)(mp->m_data + offp);
1062
1063         for (i = 0; i < count; i++) {
1064                 sp = &sa[i];
1065
1066                 st = pf_find_state_byid(sp->id, sp->creatorid);
1067                 if (st == NULL) {
1068                         V_pfsyncstats.pfsyncs_badstate++;
1069                         continue;
1070                 }
1071                 st->state_flags |= PFSTATE_NOSYNC;
1072                 pf_unlink_state(st, PF_ENTER_LOCKED);
1073         }
1074
1075         return (len);
1076 }
1077
1078 static int
1079 pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1080 {
1081         struct mbuf *mp;
1082         struct pfsync_del_c *sa, *sp;
1083         struct pf_state *st;
1084         int len = count * sizeof(*sp);
1085         int offp, i;
1086
1087         mp = m_pulldown(m, offset, len, &offp);
1088         if (mp == NULL) {
1089                 V_pfsyncstats.pfsyncs_badlen++;
1090                 return (-1);
1091         }
1092         sa = (struct pfsync_del_c *)(mp->m_data + offp);
1093
1094         for (i = 0; i < count; i++) {
1095                 sp = &sa[i];
1096
1097                 st = pf_find_state_byid(sp->id, sp->creatorid);
1098                 if (st == NULL) {
1099                         V_pfsyncstats.pfsyncs_badstate++;
1100                         continue;
1101                 }
1102
1103                 st->state_flags |= PFSTATE_NOSYNC;
1104                 pf_unlink_state(st, PF_ENTER_LOCKED);
1105         }
1106
1107         return (len);
1108 }
1109
1110 static int
1111 pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1112 {
1113         struct pfsync_softc *sc = V_pfsyncif;
1114         struct pfsync_bus *bus;
1115         struct mbuf *mp;
1116         int len = count * sizeof(*bus);
1117         int offp;
1118
1119         PFSYNC_BLOCK(sc);
1120
1121         /* If we're not waiting for a bulk update, who cares. */
1122         if (sc->sc_ureq_sent == 0) {
1123                 PFSYNC_BUNLOCK(sc);
1124                 return (len);
1125         }
1126
1127         mp = m_pulldown(m, offset, len, &offp);
1128         if (mp == NULL) {
1129                 PFSYNC_BUNLOCK(sc);
1130                 V_pfsyncstats.pfsyncs_badlen++;
1131                 return (-1);
1132         }
1133         bus = (struct pfsync_bus *)(mp->m_data + offp);
1134
1135         switch (bus->status) {
1136         case PFSYNC_BUS_START:
1137                 callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1138                     V_pf_limits[PF_LIMIT_STATES].limit /
1139                     ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1140                     sizeof(struct pfsync_state)),
1141                     pfsync_bulk_fail, sc);
1142                 if (V_pf_status.debug >= PF_DEBUG_MISC)
1143                         printf("pfsync: received bulk update start\n");
1144                 break;
1145
1146         case PFSYNC_BUS_END:
1147                 if (time_uptime - ntohl(bus->endtime) >=
1148                     sc->sc_ureq_sent) {
1149                         /* that's it, we're happy */
1150                         sc->sc_ureq_sent = 0;
1151                         sc->sc_bulk_tries = 0;
1152                         callout_stop(&sc->sc_bulkfail_tmo);
1153                         if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1154                                 (*carp_demote_adj_p)(-V_pfsync_carp_adj,
1155                                     "pfsync bulk done");
1156                         sc->sc_flags |= PFSYNCF_OK;
1157                         if (V_pf_status.debug >= PF_DEBUG_MISC)
1158                                 printf("pfsync: received valid "
1159                                     "bulk update end\n");
1160                 } else {
1161                         if (V_pf_status.debug >= PF_DEBUG_MISC)
1162                                 printf("pfsync: received invalid "
1163                                     "bulk update end: bad timestamp\n");
1164                 }
1165                 break;
1166         }
1167         PFSYNC_BUNLOCK(sc);
1168
1169         return (len);
1170 }
1171
1172 static int
1173 pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1174 {
1175         int len = count * sizeof(struct pfsync_tdb);
1176
1177 #if defined(IPSEC)
1178         struct pfsync_tdb *tp;
1179         struct mbuf *mp;
1180         int offp;
1181         int i;
1182         int s;
1183
1184         mp = m_pulldown(m, offset, len, &offp);
1185         if (mp == NULL) {
1186                 V_pfsyncstats.pfsyncs_badlen++;
1187                 return (-1);
1188         }
1189         tp = (struct pfsync_tdb *)(mp->m_data + offp);
1190
1191         for (i = 0; i < count; i++)
1192                 pfsync_update_net_tdb(&tp[i]);
1193 #endif
1194
1195         return (len);
1196 }
1197
1198 #if defined(IPSEC)
1199 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1200 static void
1201 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1202 {
1203         struct tdb              *tdb;
1204         int                      s;
1205
1206         /* check for invalid values */
1207         if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1208             (pt->dst.sa.sa_family != AF_INET &&
1209             pt->dst.sa.sa_family != AF_INET6))
1210                 goto bad;
1211
1212         tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1213         if (tdb) {
1214                 pt->rpl = ntohl(pt->rpl);
1215                 pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1216
1217                 /* Neither replay nor byte counter should ever decrease. */
1218                 if (pt->rpl < tdb->tdb_rpl ||
1219                     pt->cur_bytes < tdb->tdb_cur_bytes) {
1220                         goto bad;
1221                 }
1222
1223                 tdb->tdb_rpl = pt->rpl;
1224                 tdb->tdb_cur_bytes = pt->cur_bytes;
1225         }
1226         return;
1227
1228 bad:
1229         if (V_pf_status.debug >= PF_DEBUG_MISC)
1230                 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1231                     "invalid value\n");
1232         V_pfsyncstats.pfsyncs_badstate++;
1233         return;
1234 }
1235 #endif
1236
1237
1238 static int
1239 pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1240 {
1241         /* check if we are at the right place in the packet */
1242         if (offset != m->m_pkthdr.len)
1243                 V_pfsyncstats.pfsyncs_badlen++;
1244
1245         /* we're done. free and let the caller return */
1246         m_freem(m);
1247         return (-1);
1248 }
1249
1250 static int
1251 pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1252 {
1253         V_pfsyncstats.pfsyncs_badact++;
1254
1255         m_freem(m);
1256         return (-1);
1257 }
1258
1259 static int
1260 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1261         struct route *rt)
1262 {
1263         m_freem(m);
1264         return (0);
1265 }
1266
1267 /* ARGSUSED */
1268 static int
1269 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1270 {
1271         struct pfsync_softc *sc = ifp->if_softc;
1272         struct ifreq *ifr = (struct ifreq *)data;
1273         struct pfsyncreq pfsyncr;
1274         int error;
1275
1276         switch (cmd) {
1277         case SIOCSIFFLAGS:
1278                 PFSYNC_LOCK(sc);
1279                 if (ifp->if_flags & IFF_UP) {
1280                         ifp->if_drv_flags |= IFF_DRV_RUNNING;
1281                         PFSYNC_UNLOCK(sc);
1282                         pfsync_pointers_init();
1283                 } else {
1284                         ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1285                         PFSYNC_UNLOCK(sc);
1286                         pfsync_pointers_uninit();
1287                 }
1288                 break;
1289         case SIOCSIFMTU:
1290                 if (!sc->sc_sync_if ||
1291                     ifr->ifr_mtu <= PFSYNC_MINPKT ||
1292                     ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1293                         return (EINVAL);
1294                 if (ifr->ifr_mtu < ifp->if_mtu) {
1295                         PFSYNC_LOCK(sc);
1296                         if (sc->sc_len > PFSYNC_MINPKT)
1297                                 pfsync_sendout(1);
1298                         PFSYNC_UNLOCK(sc);
1299                 }
1300                 ifp->if_mtu = ifr->ifr_mtu;
1301                 break;
1302         case SIOCGETPFSYNC:
1303                 bzero(&pfsyncr, sizeof(pfsyncr));
1304                 PFSYNC_LOCK(sc);
1305                 if (sc->sc_sync_if) {
1306                         strlcpy(pfsyncr.pfsyncr_syncdev,
1307                             sc->sc_sync_if->if_xname, IFNAMSIZ);
1308                 }
1309                 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer;
1310                 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1311                 pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER ==
1312                     (sc->sc_flags & PFSYNCF_DEFER));
1313                 PFSYNC_UNLOCK(sc);
1314                 return (copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr)));
1315
1316         case SIOCSETPFSYNC:
1317             {
1318                 struct ip_moptions *imo = &sc->sc_imo;
1319                 struct ifnet *sifp;
1320                 struct ip *ip;
1321                 void *mship = NULL;
1322
1323                 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1324                         return (error);
1325                 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr))))
1326                         return (error);
1327
1328                 if (pfsyncr.pfsyncr_maxupdates > 255)
1329                         return (EINVAL);
1330
1331                 if (pfsyncr.pfsyncr_syncdev[0] == 0)
1332                         sifp = NULL;
1333                 else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL)
1334                         return (EINVAL);
1335
1336                 if (sifp != NULL && (
1337                     pfsyncr.pfsyncr_syncpeer.s_addr == 0 ||
1338                     pfsyncr.pfsyncr_syncpeer.s_addr ==
1339                     htonl(INADDR_PFSYNC_GROUP)))
1340                         mship = malloc((sizeof(struct in_multi *) *
1341                             IP_MIN_MEMBERSHIPS), M_PFSYNC, M_WAITOK | M_ZERO);
1342
1343                 PFSYNC_LOCK(sc);
1344                 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0)
1345                         sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP);
1346                 else
1347                         sc->sc_sync_peer.s_addr =
1348                             pfsyncr.pfsyncr_syncpeer.s_addr;
1349
1350                 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates;
1351                 if (pfsyncr.pfsyncr_defer) {
1352                         sc->sc_flags |= PFSYNCF_DEFER;
1353                         pfsync_defer_ptr = pfsync_defer;
1354                 } else {
1355                         sc->sc_flags &= ~PFSYNCF_DEFER;
1356                         pfsync_defer_ptr = NULL;
1357                 }
1358
1359                 if (sifp == NULL) {
1360                         if (sc->sc_sync_if)
1361                                 if_rele(sc->sc_sync_if);
1362                         sc->sc_sync_if = NULL;
1363                         if (imo->imo_membership)
1364                                 pfsync_multicast_cleanup(sc);
1365                         PFSYNC_UNLOCK(sc);
1366                         break;
1367                 }
1368
1369                 if (sc->sc_len > PFSYNC_MINPKT &&
1370                     (sifp->if_mtu < sc->sc_ifp->if_mtu ||
1371                     (sc->sc_sync_if != NULL &&
1372                     sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
1373                     sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
1374                         pfsync_sendout(1);
1375
1376                 if (imo->imo_membership)
1377                         pfsync_multicast_cleanup(sc);
1378
1379                 if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) {
1380                         error = pfsync_multicast_setup(sc, sifp, mship);
1381                         if (error) {
1382                                 if_rele(sifp);
1383                                 free(mship, M_PFSYNC);
1384                                 return (error);
1385                         }
1386                 }
1387                 if (sc->sc_sync_if)
1388                         if_rele(sc->sc_sync_if);
1389                 sc->sc_sync_if = sifp;
1390
1391                 ip = &sc->sc_template;
1392                 bzero(ip, sizeof(*ip));
1393                 ip->ip_v = IPVERSION;
1394                 ip->ip_hl = sizeof(sc->sc_template) >> 2;
1395                 ip->ip_tos = IPTOS_LOWDELAY;
1396                 /* len and id are set later. */
1397                 ip->ip_off = htons(IP_DF);
1398                 ip->ip_ttl = PFSYNC_DFLTTL;
1399                 ip->ip_p = IPPROTO_PFSYNC;
1400                 ip->ip_src.s_addr = INADDR_ANY;
1401                 ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr;
1402
1403                 /* Request a full state table update. */
1404                 if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1405                         (*carp_demote_adj_p)(V_pfsync_carp_adj,
1406                             "pfsync bulk start");
1407                 sc->sc_flags &= ~PFSYNCF_OK;
1408                 if (V_pf_status.debug >= PF_DEBUG_MISC)
1409                         printf("pfsync: requesting bulk update\n");
1410                 pfsync_request_update(0, 0);
1411                 PFSYNC_UNLOCK(sc);
1412                 PFSYNC_BLOCK(sc);
1413                 sc->sc_ureq_sent = time_uptime;
1414                 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail,
1415                     sc);
1416                 PFSYNC_BUNLOCK(sc);
1417
1418                 break;
1419             }
1420         default:
1421                 return (ENOTTY);
1422         }
1423
1424         return (0);
1425 }
1426
1427 static void
1428 pfsync_out_state(struct pf_state *st, void *buf)
1429 {
1430         struct pfsync_state *sp = buf;
1431
1432         pfsync_state_export(sp, st);
1433 }
1434
1435 static void
1436 pfsync_out_iack(struct pf_state *st, void *buf)
1437 {
1438         struct pfsync_ins_ack *iack = buf;
1439
1440         iack->id = st->id;
1441         iack->creatorid = st->creatorid;
1442 }
1443
1444 static void
1445 pfsync_out_upd_c(struct pf_state *st, void *buf)
1446 {
1447         struct pfsync_upd_c *up = buf;
1448
1449         bzero(up, sizeof(*up));
1450         up->id = st->id;
1451         pf_state_peer_hton(&st->src, &up->src);
1452         pf_state_peer_hton(&st->dst, &up->dst);
1453         up->creatorid = st->creatorid;
1454         up->timeout = st->timeout;
1455 }
1456
1457 static void
1458 pfsync_out_del(struct pf_state *st, void *buf)
1459 {
1460         struct pfsync_del_c *dp = buf;
1461
1462         dp->id = st->id;
1463         dp->creatorid = st->creatorid;
1464         st->state_flags |= PFSTATE_NOSYNC;
1465 }
1466
1467 static void
1468 pfsync_drop(struct pfsync_softc *sc)
1469 {
1470         struct pf_state *st, *next;
1471         struct pfsync_upd_req_item *ur;
1472         int q;
1473
1474         for (q = 0; q < PFSYNC_S_COUNT; q++) {
1475                 if (TAILQ_EMPTY(&sc->sc_qs[q]))
1476                         continue;
1477
1478                 TAILQ_FOREACH_SAFE(st, &sc->sc_qs[q], sync_list, next) {
1479                         KASSERT(st->sync_state == q,
1480                                 ("%s: st->sync_state == q",
1481                                         __func__));
1482                         st->sync_state = PFSYNC_S_NONE;
1483                         pf_release_state(st);
1484                 }
1485                 TAILQ_INIT(&sc->sc_qs[q]);
1486         }
1487
1488         while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) {
1489                 TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry);
1490                 free(ur, M_PFSYNC);
1491         }
1492
1493         sc->sc_plus = NULL;
1494         sc->sc_len = PFSYNC_MINPKT;
1495 }
1496
1497 static void
1498 pfsync_sendout(int schedswi)
1499 {
1500         struct pfsync_softc *sc = V_pfsyncif;
1501         struct ifnet *ifp = sc->sc_ifp;
1502         struct mbuf *m;
1503         struct ip *ip;
1504         struct pfsync_header *ph;
1505         struct pfsync_subheader *subh;
1506         struct pf_state *st;
1507         struct pfsync_upd_req_item *ur;
1508         int offset;
1509         int q, count = 0;
1510
1511         KASSERT(sc != NULL, ("%s: null sc", __func__));
1512         KASSERT(sc->sc_len > PFSYNC_MINPKT,
1513             ("%s: sc_len %zu", __func__, sc->sc_len));
1514         PFSYNC_LOCK_ASSERT(sc);
1515
1516         if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) {
1517                 pfsync_drop(sc);
1518                 return;
1519         }
1520
1521         m = m_get2(max_linkhdr + sc->sc_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1522         if (m == NULL) {
1523                 sc->sc_ifp->if_oerrors++;
1524                 V_pfsyncstats.pfsyncs_onomem++;
1525                 return;
1526         }
1527         m->m_data += max_linkhdr;
1528         m->m_len = m->m_pkthdr.len = sc->sc_len;
1529
1530         /* build the ip header */
1531         ip = (struct ip *)m->m_data;
1532         bcopy(&sc->sc_template, ip, sizeof(*ip));
1533         offset = sizeof(*ip);
1534
1535         ip->ip_len = htons(m->m_pkthdr.len);
1536         ip->ip_id = htons(ip_randomid());
1537
1538         /* build the pfsync header */
1539         ph = (struct pfsync_header *)(m->m_data + offset);
1540         bzero(ph, sizeof(*ph));
1541         offset += sizeof(*ph);
1542
1543         ph->version = PFSYNC_VERSION;
1544         ph->len = htons(sc->sc_len - sizeof(*ip));
1545         bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1546
1547         /* walk the queues */
1548         for (q = 0; q < PFSYNC_S_COUNT; q++) {
1549                 if (TAILQ_EMPTY(&sc->sc_qs[q]))
1550                         continue;
1551
1552                 subh = (struct pfsync_subheader *)(m->m_data + offset);
1553                 offset += sizeof(*subh);
1554
1555                 count = 0;
1556                 TAILQ_FOREACH(st, &sc->sc_qs[q], sync_list) {
1557                         KASSERT(st->sync_state == q,
1558                                 ("%s: st->sync_state == q",
1559                                         __func__));
1560                         /*
1561                          * XXXGL: some of write methods do unlocked reads
1562                          * of state data :(
1563                          */
1564                         pfsync_qs[q].write(st, m->m_data + offset);
1565                         offset += pfsync_qs[q].len;
1566                         st->sync_state = PFSYNC_S_NONE;
1567                         pf_release_state(st);
1568                         count++;
1569                 }
1570                 TAILQ_INIT(&sc->sc_qs[q]);
1571
1572                 bzero(subh, sizeof(*subh));
1573                 subh->action = pfsync_qs[q].action;
1574                 subh->count = htons(count);
1575                 V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1576         }
1577
1578         if (!TAILQ_EMPTY(&sc->sc_upd_req_list)) {
1579                 subh = (struct pfsync_subheader *)(m->m_data + offset);
1580                 offset += sizeof(*subh);
1581
1582                 count = 0;
1583                 while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) {
1584                         TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry);
1585
1586                         bcopy(&ur->ur_msg, m->m_data + offset,
1587                             sizeof(ur->ur_msg));
1588                         offset += sizeof(ur->ur_msg);
1589                         free(ur, M_PFSYNC);
1590                         count++;
1591                 }
1592
1593                 bzero(subh, sizeof(*subh));
1594                 subh->action = PFSYNC_ACT_UPD_REQ;
1595                 subh->count = htons(count);
1596                 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1597         }
1598
1599         /* has someone built a custom region for us to add? */
1600         if (sc->sc_plus != NULL) {
1601                 bcopy(sc->sc_plus, m->m_data + offset, sc->sc_pluslen);
1602                 offset += sc->sc_pluslen;
1603
1604                 sc->sc_plus = NULL;
1605         }
1606
1607         subh = (struct pfsync_subheader *)(m->m_data + offset);
1608         offset += sizeof(*subh);
1609
1610         bzero(subh, sizeof(*subh));
1611         subh->action = PFSYNC_ACT_EOF;
1612         subh->count = htons(1);
1613         V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
1614
1615         /* we're done, let's put it on the wire */
1616         if (ifp->if_bpf) {
1617                 m->m_data += sizeof(*ip);
1618                 m->m_len = m->m_pkthdr.len = sc->sc_len - sizeof(*ip);
1619                 BPF_MTAP(ifp, m);
1620                 m->m_data -= sizeof(*ip);
1621                 m->m_len = m->m_pkthdr.len = sc->sc_len;
1622         }
1623
1624         if (sc->sc_sync_if == NULL) {
1625                 sc->sc_len = PFSYNC_MINPKT;
1626                 m_freem(m);
1627                 return;
1628         }
1629
1630         sc->sc_ifp->if_opackets++;
1631         sc->sc_ifp->if_obytes += m->m_pkthdr.len;
1632         sc->sc_len = PFSYNC_MINPKT;
1633
1634         if (!_IF_QFULL(&sc->sc_ifp->if_snd))
1635                 _IF_ENQUEUE(&sc->sc_ifp->if_snd, m);
1636         else {
1637                 m_freem(m);
1638                 sc->sc_ifp->if_snd.ifq_drops++;
1639         }
1640         if (schedswi)
1641                 swi_sched(V_pfsync_swi_cookie, 0);
1642 }
1643
1644 static void
1645 pfsync_insert_state(struct pf_state *st)
1646 {
1647         struct pfsync_softc *sc = V_pfsyncif;
1648
1649         if (st->state_flags & PFSTATE_NOSYNC)
1650                 return;
1651
1652         if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) ||
1653             st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
1654                 st->state_flags |= PFSTATE_NOSYNC;
1655                 return;
1656         }
1657
1658         KASSERT(st->sync_state == PFSYNC_S_NONE,
1659                 ("%s: st->sync_state %u", __func__, st->sync_state));
1660
1661         PFSYNC_LOCK(sc);
1662         if (sc->sc_len == PFSYNC_MINPKT)
1663                 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1664
1665         pfsync_q_ins(st, PFSYNC_S_INS);
1666         PFSYNC_UNLOCK(sc);
1667
1668         st->sync_updates = 0;
1669 }
1670
1671 static int
1672 pfsync_defer(struct pf_state *st, struct mbuf *m)
1673 {
1674         struct pfsync_softc *sc = V_pfsyncif;
1675         struct pfsync_deferral *pd;
1676
1677         if (m->m_flags & (M_BCAST|M_MCAST))
1678                 return (0);
1679
1680         PFSYNC_LOCK(sc);
1681
1682         if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) ||
1683             !(sc->sc_flags & PFSYNCF_DEFER)) {
1684                 PFSYNC_UNLOCK(sc);
1685                 return (0);
1686         }
1687
1688          if (sc->sc_deferred >= 128)
1689                 pfsync_undefer(TAILQ_FIRST(&sc->sc_deferrals), 0);
1690
1691         pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
1692         if (pd == NULL)
1693                 return (0);
1694         sc->sc_deferred++;
1695
1696         m->m_flags |= M_SKIP_FIREWALL;
1697         st->state_flags |= PFSTATE_ACK;
1698
1699         pd->pd_sc = sc;
1700         pd->pd_refs = 0;
1701         pd->pd_st = st;
1702         pf_ref_state(st);
1703         pd->pd_m = m;
1704
1705         TAILQ_INSERT_TAIL(&sc->sc_deferrals, pd, pd_entry);
1706         callout_init_mtx(&pd->pd_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1707         callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd);
1708
1709         pfsync_push(sc);
1710
1711         return (1);
1712 }
1713
1714 static void
1715 pfsync_undefer(struct pfsync_deferral *pd, int drop)
1716 {
1717         struct pfsync_softc *sc = pd->pd_sc;
1718         struct mbuf *m = pd->pd_m;
1719         struct pf_state *st = pd->pd_st;
1720
1721         PFSYNC_LOCK_ASSERT(sc);
1722
1723         TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
1724         sc->sc_deferred--;
1725         pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */
1726         free(pd, M_PFSYNC);
1727         pf_release_state(st);
1728
1729         if (drop)
1730                 m_freem(m);
1731         else {
1732                 _IF_ENQUEUE(&sc->sc_ifp->if_snd, m);
1733                 pfsync_push(sc);
1734         }
1735 }
1736
1737 static void
1738 pfsync_defer_tmo(void *arg)
1739 {
1740         struct pfsync_deferral *pd = arg;
1741         struct pfsync_softc *sc = pd->pd_sc;
1742         struct mbuf *m = pd->pd_m;
1743         struct pf_state *st = pd->pd_st;
1744
1745         PFSYNC_LOCK_ASSERT(sc);
1746
1747         CURVNET_SET(m->m_pkthdr.rcvif->if_vnet);
1748
1749         TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
1750         sc->sc_deferred--;
1751         pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */
1752         if (pd->pd_refs == 0)
1753                 free(pd, M_PFSYNC);
1754         PFSYNC_UNLOCK(sc);
1755
1756         ip_output(m, NULL, NULL, 0, NULL, NULL);
1757
1758         pf_release_state(st);
1759
1760         CURVNET_RESTORE();
1761 }
1762
1763 static void
1764 pfsync_undefer_state(struct pf_state *st, int drop)
1765 {
1766         struct pfsync_softc *sc = V_pfsyncif;
1767         struct pfsync_deferral *pd;
1768
1769         PFSYNC_LOCK_ASSERT(sc);
1770
1771         TAILQ_FOREACH(pd, &sc->sc_deferrals, pd_entry) {
1772                  if (pd->pd_st == st) {
1773                         if (callout_stop(&pd->pd_tmo))
1774                                 pfsync_undefer(pd, drop);
1775                         return;
1776                 }
1777         }
1778
1779         panic("%s: unable to find deferred state", __func__);
1780 }
1781
1782 static void
1783 pfsync_update_state(struct pf_state *st)
1784 {
1785         struct pfsync_softc *sc = V_pfsyncif;
1786         int sync = 0;
1787
1788         PF_STATE_LOCK_ASSERT(st);
1789         PFSYNC_LOCK(sc);
1790
1791         if (st->state_flags & PFSTATE_ACK)
1792                 pfsync_undefer_state(st, 0);
1793         if (st->state_flags & PFSTATE_NOSYNC) {
1794                 if (st->sync_state != PFSYNC_S_NONE)
1795                         pfsync_q_del(st);
1796                 PFSYNC_UNLOCK(sc);
1797                 return;
1798         }
1799
1800         if (sc->sc_len == PFSYNC_MINPKT)
1801                 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1802
1803         switch (st->sync_state) {
1804         case PFSYNC_S_UPD_C:
1805         case PFSYNC_S_UPD:
1806         case PFSYNC_S_INS:
1807                 /* we're already handling it */
1808
1809                 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
1810                         st->sync_updates++;
1811                         if (st->sync_updates >= sc->sc_maxupdates)
1812                                 sync = 1;
1813                 }
1814                 break;
1815
1816         case PFSYNC_S_IACK:
1817                 pfsync_q_del(st);
1818         case PFSYNC_S_NONE:
1819                 pfsync_q_ins(st, PFSYNC_S_UPD_C);
1820                 st->sync_updates = 0;
1821                 break;
1822
1823         default:
1824                 panic("%s: unexpected sync state %d", __func__, st->sync_state);
1825         }
1826
1827         if (sync || (time_uptime - st->pfsync_time) < 2)
1828                 pfsync_push(sc);
1829
1830         PFSYNC_UNLOCK(sc);
1831 }
1832
1833 static void
1834 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
1835 {
1836         struct pfsync_softc *sc = V_pfsyncif;
1837         struct pfsync_upd_req_item *item;
1838         size_t nlen = sizeof(struct pfsync_upd_req);
1839
1840         PFSYNC_LOCK_ASSERT(sc);
1841
1842         /*
1843          * This code does a bit to prevent multiple update requests for the
1844          * same state being generated. It searches current subheader queue,
1845          * but it doesn't lookup into queue of already packed datagrams.
1846          */
1847         TAILQ_FOREACH(item, &sc->sc_upd_req_list, ur_entry)
1848                 if (item->ur_msg.id == id &&
1849                     item->ur_msg.creatorid == creatorid)
1850                         return;
1851
1852         item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
1853         if (item == NULL)
1854                 return; /* XXX stats */
1855
1856         item->ur_msg.id = id;
1857         item->ur_msg.creatorid = creatorid;
1858
1859         if (TAILQ_EMPTY(&sc->sc_upd_req_list))
1860                 nlen += sizeof(struct pfsync_subheader);
1861
1862         if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) {
1863                 pfsync_sendout(1);
1864
1865                 nlen = sizeof(struct pfsync_subheader) +
1866                     sizeof(struct pfsync_upd_req);
1867         }
1868
1869         TAILQ_INSERT_TAIL(&sc->sc_upd_req_list, item, ur_entry);
1870         sc->sc_len += nlen;
1871 }
1872
1873 static void
1874 pfsync_update_state_req(struct pf_state *st)
1875 {
1876         struct pfsync_softc *sc = V_pfsyncif;
1877
1878         PF_STATE_LOCK_ASSERT(st);
1879         PFSYNC_LOCK(sc);
1880
1881         if (st->state_flags & PFSTATE_NOSYNC) {
1882                 if (st->sync_state != PFSYNC_S_NONE)
1883                         pfsync_q_del(st);
1884                 PFSYNC_UNLOCK(sc);
1885                 return;
1886         }
1887
1888         switch (st->sync_state) {
1889         case PFSYNC_S_UPD_C:
1890         case PFSYNC_S_IACK:
1891                 pfsync_q_del(st);
1892         case PFSYNC_S_NONE:
1893                 pfsync_q_ins(st, PFSYNC_S_UPD);
1894                 pfsync_push(sc);
1895                 break;
1896
1897         case PFSYNC_S_INS:
1898         case PFSYNC_S_UPD:
1899         case PFSYNC_S_DEL:
1900                 /* we're already handling it */
1901                 break;
1902
1903         default:
1904                 panic("%s: unexpected sync state %d", __func__, st->sync_state);
1905         }
1906
1907         PFSYNC_UNLOCK(sc);
1908 }
1909
1910 static void
1911 pfsync_delete_state(struct pf_state *st)
1912 {
1913         struct pfsync_softc *sc = V_pfsyncif;
1914
1915         PFSYNC_LOCK(sc);
1916         if (st->state_flags & PFSTATE_ACK)
1917                 pfsync_undefer_state(st, 1);
1918         if (st->state_flags & PFSTATE_NOSYNC) {
1919                 if (st->sync_state != PFSYNC_S_NONE)
1920                         pfsync_q_del(st);
1921                 PFSYNC_UNLOCK(sc);
1922                 return;
1923         }
1924
1925         if (sc->sc_len == PFSYNC_MINPKT)
1926                 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1927
1928         switch (st->sync_state) {
1929         case PFSYNC_S_INS:
1930                 /* We never got to tell the world so just forget about it. */
1931                 pfsync_q_del(st);
1932                 break;
1933
1934         case PFSYNC_S_UPD_C:
1935         case PFSYNC_S_UPD:
1936         case PFSYNC_S_IACK:
1937                 pfsync_q_del(st);
1938                 /* FALLTHROUGH to putting it on the del list */
1939
1940         case PFSYNC_S_NONE:
1941                 pfsync_q_ins(st, PFSYNC_S_DEL);
1942                 break;
1943
1944         default:
1945                 panic("%s: unexpected sync state %d", __func__, st->sync_state);
1946         }
1947         PFSYNC_UNLOCK(sc);
1948 }
1949
1950 static void
1951 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
1952 {
1953         struct pfsync_softc *sc = V_pfsyncif;
1954         struct {
1955                 struct pfsync_subheader subh;
1956                 struct pfsync_clr clr;
1957         } __packed r;
1958
1959         bzero(&r, sizeof(r));
1960
1961         r.subh.action = PFSYNC_ACT_CLR;
1962         r.subh.count = htons(1);
1963         V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
1964
1965         strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
1966         r.clr.creatorid = creatorid;
1967
1968         PFSYNC_LOCK(sc);
1969         pfsync_send_plus(&r, sizeof(r));
1970         PFSYNC_UNLOCK(sc);
1971 }
1972
1973 static void
1974 pfsync_q_ins(struct pf_state *st, int q)
1975 {
1976         struct pfsync_softc *sc = V_pfsyncif;
1977         size_t nlen = pfsync_qs[q].len;
1978
1979         PFSYNC_LOCK_ASSERT(sc);
1980
1981         KASSERT(st->sync_state == PFSYNC_S_NONE,
1982                 ("%s: st->sync_state %u", __func__, st->sync_state));
1983         KASSERT(sc->sc_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
1984             sc->sc_len));
1985
1986         if (TAILQ_EMPTY(&sc->sc_qs[q]))
1987                 nlen += sizeof(struct pfsync_subheader);
1988
1989         if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) {
1990                 pfsync_sendout(1);
1991
1992                 nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
1993         }
1994
1995         sc->sc_len += nlen;
1996         TAILQ_INSERT_TAIL(&sc->sc_qs[q], st, sync_list);
1997         st->sync_state = q;
1998         pf_ref_state(st);
1999 }
2000
2001 static void
2002 pfsync_q_del(struct pf_state *st)
2003 {
2004         struct pfsync_softc *sc = V_pfsyncif;
2005         int q = st->sync_state;
2006
2007         PFSYNC_LOCK_ASSERT(sc);
2008         KASSERT(st->sync_state != PFSYNC_S_NONE,
2009                 ("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2010
2011         sc->sc_len -= pfsync_qs[q].len;
2012         TAILQ_REMOVE(&sc->sc_qs[q], st, sync_list);
2013         st->sync_state = PFSYNC_S_NONE;
2014         pf_release_state(st);
2015
2016         if (TAILQ_EMPTY(&sc->sc_qs[q]))
2017                 sc->sc_len -= sizeof(struct pfsync_subheader);
2018 }
2019
2020 static void
2021 pfsync_bulk_start(void)
2022 {
2023         struct pfsync_softc *sc = V_pfsyncif;
2024
2025         if (V_pf_status.debug >= PF_DEBUG_MISC)
2026                 printf("pfsync: received bulk update request\n");
2027
2028         PFSYNC_BLOCK(sc);
2029
2030         sc->sc_ureq_received = time_uptime;
2031         sc->sc_bulk_hashid = 0;
2032         sc->sc_bulk_stateid = 0;
2033         pfsync_bulk_status(PFSYNC_BUS_START);
2034         callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2035         PFSYNC_BUNLOCK(sc);
2036 }
2037
2038 static void
2039 pfsync_bulk_update(void *arg)
2040 {
2041         struct pfsync_softc *sc = arg;
2042         struct pf_state *s;
2043         int i, sent = 0;
2044
2045         PFSYNC_BLOCK_ASSERT(sc);
2046         CURVNET_SET(sc->sc_ifp->if_vnet);
2047
2048         /*
2049          * Start with last state from previous invocation.
2050          * It may had gone, in this case start from the
2051          * hash slot.
2052          */
2053         s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2054
2055         if (s != NULL)
2056                 i = PF_IDHASH(s);
2057         else
2058                 i = sc->sc_bulk_hashid;
2059
2060         for (; i <= pf_hashmask; i++) {
2061                 struct pf_idhash *ih = &V_pf_idhash[i];
2062
2063                 if (s != NULL)
2064                         PF_HASHROW_ASSERT(ih);
2065                 else {
2066                         PF_HASHROW_LOCK(ih);
2067                         s = LIST_FIRST(&ih->states);
2068                 }
2069
2070                 for (; s; s = LIST_NEXT(s, entry)) {
2071
2072                         if (sent > 1 && (sc->sc_ifp->if_mtu - sc->sc_len) <
2073                             sizeof(struct pfsync_state)) {
2074                                 /* We've filled a packet. */
2075                                 sc->sc_bulk_hashid = i;
2076                                 sc->sc_bulk_stateid = s->id;
2077                                 sc->sc_bulk_creatorid = s->creatorid;
2078                                 PF_HASHROW_UNLOCK(ih);
2079                                 callout_reset(&sc->sc_bulk_tmo, 1,
2080                                     pfsync_bulk_update, sc);
2081                                 goto full;
2082                         }
2083
2084                         if (s->sync_state == PFSYNC_S_NONE &&
2085                             s->timeout < PFTM_MAX &&
2086                             s->pfsync_time <= sc->sc_ureq_received) {
2087                                 pfsync_update_state_req(s);
2088                                 sent++;
2089                         }
2090                 }
2091                 PF_HASHROW_UNLOCK(ih);
2092         }
2093
2094         /* We're done. */
2095         pfsync_bulk_status(PFSYNC_BUS_END);
2096
2097 full:
2098         CURVNET_RESTORE();
2099 }
2100
2101 static void
2102 pfsync_bulk_status(u_int8_t status)
2103 {
2104         struct {
2105                 struct pfsync_subheader subh;
2106                 struct pfsync_bus bus;
2107         } __packed r;
2108
2109         struct pfsync_softc *sc = V_pfsyncif;
2110
2111         bzero(&r, sizeof(r));
2112
2113         r.subh.action = PFSYNC_ACT_BUS;
2114         r.subh.count = htons(1);
2115         V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2116
2117         r.bus.creatorid = V_pf_status.hostid;
2118         r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2119         r.bus.status = status;
2120
2121         PFSYNC_LOCK(sc);
2122         pfsync_send_plus(&r, sizeof(r));
2123         PFSYNC_UNLOCK(sc);
2124 }
2125
2126 static void
2127 pfsync_bulk_fail(void *arg)
2128 {
2129         struct pfsync_softc *sc = arg;
2130
2131         CURVNET_SET(sc->sc_ifp->if_vnet);
2132
2133         PFSYNC_BLOCK_ASSERT(sc);
2134
2135         if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2136                 /* Try again */
2137                 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2138                     pfsync_bulk_fail, V_pfsyncif);
2139                 PFSYNC_LOCK(sc);
2140                 pfsync_request_update(0, 0);
2141                 PFSYNC_UNLOCK(sc);
2142         } else {
2143                 /* Pretend like the transfer was ok. */
2144                 sc->sc_ureq_sent = 0;
2145                 sc->sc_bulk_tries = 0;
2146                 PFSYNC_LOCK(sc);
2147                 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2148                         (*carp_demote_adj_p)(-V_pfsync_carp_adj,
2149                             "pfsync bulk fail");
2150                 sc->sc_flags |= PFSYNCF_OK;
2151                 PFSYNC_UNLOCK(sc);
2152                 if (V_pf_status.debug >= PF_DEBUG_MISC)
2153                         printf("pfsync: failed to receive bulk update\n");
2154         }
2155
2156         CURVNET_RESTORE();
2157 }
2158
2159 static void
2160 pfsync_send_plus(void *plus, size_t pluslen)
2161 {
2162         struct pfsync_softc *sc = V_pfsyncif;
2163
2164         PFSYNC_LOCK_ASSERT(sc);
2165
2166         if (sc->sc_len + pluslen > sc->sc_ifp->if_mtu)
2167                 pfsync_sendout(1);
2168
2169         sc->sc_plus = plus;
2170         sc->sc_len += (sc->sc_pluslen = pluslen);
2171
2172         pfsync_sendout(1);
2173 }
2174
2175 static void
2176 pfsync_timeout(void *arg)
2177 {
2178         struct pfsync_softc *sc = arg;
2179
2180         CURVNET_SET(sc->sc_ifp->if_vnet);
2181         PFSYNC_LOCK(sc);
2182         pfsync_push(sc);
2183         PFSYNC_UNLOCK(sc);
2184         CURVNET_RESTORE();
2185 }
2186
2187 static void
2188 pfsync_push(struct pfsync_softc *sc)
2189 {
2190
2191         PFSYNC_LOCK_ASSERT(sc);
2192
2193         sc->sc_flags |= PFSYNCF_PUSH;
2194         swi_sched(V_pfsync_swi_cookie, 0);
2195 }
2196
2197 static void
2198 pfsyncintr(void *arg)
2199 {
2200         struct pfsync_softc *sc = arg;
2201         struct mbuf *m, *n;
2202
2203         CURVNET_SET(sc->sc_ifp->if_vnet);
2204
2205         PFSYNC_LOCK(sc);
2206         if ((sc->sc_flags & PFSYNCF_PUSH) && sc->sc_len > PFSYNC_MINPKT) {
2207                 pfsync_sendout(0);
2208                 sc->sc_flags &= ~PFSYNCF_PUSH;
2209         }
2210         _IF_DEQUEUE_ALL(&sc->sc_ifp->if_snd, m);
2211         PFSYNC_UNLOCK(sc);
2212
2213         for (; m != NULL; m = n) {
2214
2215                 n = m->m_nextpkt;
2216                 m->m_nextpkt = NULL;
2217
2218                 /*
2219                  * We distinguish between a deferral packet and our
2220                  * own pfsync packet based on M_SKIP_FIREWALL
2221                  * flag. This is XXX.
2222                  */
2223                 if (m->m_flags & M_SKIP_FIREWALL)
2224                         ip_output(m, NULL, NULL, 0, NULL, NULL);
2225                 else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo,
2226                     NULL) == 0)
2227                         V_pfsyncstats.pfsyncs_opackets++;
2228                 else
2229                         V_pfsyncstats.pfsyncs_oerrors++;
2230         }
2231         CURVNET_RESTORE();
2232 }
2233
2234 static int
2235 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp, void *mship)
2236 {
2237         struct ip_moptions *imo = &sc->sc_imo;
2238         int error;
2239
2240         if (!(ifp->if_flags & IFF_MULTICAST))
2241                 return (EADDRNOTAVAIL);
2242
2243         imo->imo_membership = (struct in_multi **)mship;
2244         imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
2245         imo->imo_multicast_vif = -1;
2246
2247         if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL,
2248             &imo->imo_membership[0])) != 0) {
2249                 imo->imo_membership = NULL;
2250                 return (error);
2251         }
2252         imo->imo_num_memberships++;
2253         imo->imo_multicast_ifp = ifp;
2254         imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2255         imo->imo_multicast_loop = 0;
2256
2257         return (0);
2258 }
2259
2260 static void
2261 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2262 {
2263         struct ip_moptions *imo = &sc->sc_imo;
2264
2265         in_leavegroup(imo->imo_membership[0], NULL);
2266         free(imo->imo_membership, M_PFSYNC);
2267         imo->imo_membership = NULL;
2268         imo->imo_multicast_ifp = NULL;
2269 }
2270
2271 #ifdef INET
2272 extern  struct domain inetdomain;
2273 static struct protosw in_pfsync_protosw = {
2274         .pr_type =              SOCK_RAW,
2275         .pr_domain =            &inetdomain,
2276         .pr_protocol =          IPPROTO_PFSYNC,
2277         .pr_flags =             PR_ATOMIC|PR_ADDR,
2278         .pr_input =             pfsync_input,
2279         .pr_output =            (pr_output_t *)rip_output,
2280         .pr_ctloutput =         rip_ctloutput,
2281         .pr_usrreqs =           &rip_usrreqs
2282 };
2283 #endif
2284
2285 static void
2286 pfsync_pointers_init()
2287 {
2288
2289         PF_RULES_WLOCK();
2290         pfsync_state_import_ptr = pfsync_state_import;
2291         pfsync_insert_state_ptr = pfsync_insert_state;
2292         pfsync_update_state_ptr = pfsync_update_state;
2293         pfsync_delete_state_ptr = pfsync_delete_state;
2294         pfsync_clear_states_ptr = pfsync_clear_states;
2295         pfsync_defer_ptr = pfsync_defer;
2296         PF_RULES_WUNLOCK();
2297 }
2298
2299 static void
2300 pfsync_pointers_uninit()
2301 {
2302
2303         PF_RULES_WLOCK();
2304         pfsync_state_import_ptr = NULL;
2305         pfsync_insert_state_ptr = NULL;
2306         pfsync_update_state_ptr = NULL;
2307         pfsync_delete_state_ptr = NULL;
2308         pfsync_clear_states_ptr = NULL;
2309         pfsync_defer_ptr = NULL;
2310         PF_RULES_WUNLOCK();
2311 }
2312
2313 static int
2314 pfsync_init()
2315 {
2316         VNET_ITERATOR_DECL(vnet_iter);
2317         int error = 0;
2318
2319         VNET_LIST_RLOCK();
2320         VNET_FOREACH(vnet_iter) {
2321                 CURVNET_SET(vnet_iter);
2322                 V_pfsync_cloner = if_clone_simple(pfsyncname,
2323                     pfsync_clone_create, pfsync_clone_destroy, 1);
2324                 error = swi_add(NULL, pfsyncname, pfsyncintr, V_pfsyncif,
2325                     SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
2326                 CURVNET_RESTORE();
2327                 if (error)
2328                         goto fail_locked;
2329         }
2330         VNET_LIST_RUNLOCK();
2331 #ifdef INET
2332         error = pf_proto_register(PF_INET, &in_pfsync_protosw);
2333         if (error)
2334                 goto fail;
2335         error = ipproto_register(IPPROTO_PFSYNC);
2336         if (error) {
2337                 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2338                 goto fail;
2339         }
2340 #endif
2341         pfsync_pointers_init();
2342
2343         return (0);
2344
2345 fail:
2346         VNET_LIST_RLOCK();
2347 fail_locked:
2348         VNET_FOREACH(vnet_iter) {
2349                 CURVNET_SET(vnet_iter);
2350                 if (V_pfsync_swi_cookie) {
2351                         swi_remove(V_pfsync_swi_cookie);
2352                         if_clone_detach(V_pfsync_cloner);
2353                 }
2354                 CURVNET_RESTORE();
2355         }
2356         VNET_LIST_RUNLOCK();
2357
2358         return (error);
2359 }
2360
2361 static void
2362 pfsync_uninit()
2363 {
2364         VNET_ITERATOR_DECL(vnet_iter);
2365
2366         pfsync_pointers_uninit();
2367
2368         ipproto_unregister(IPPROTO_PFSYNC);
2369         pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2370         VNET_LIST_RLOCK();
2371         VNET_FOREACH(vnet_iter) {
2372                 CURVNET_SET(vnet_iter);
2373                 if_clone_detach(V_pfsync_cloner);
2374                 swi_remove(V_pfsync_swi_cookie);
2375                 CURVNET_RESTORE();
2376         }
2377         VNET_LIST_RUNLOCK();
2378 }
2379
2380 static int
2381 pfsync_modevent(module_t mod, int type, void *data)
2382 {
2383         int error = 0;
2384
2385         switch (type) {
2386         case MOD_LOAD:
2387                 error = pfsync_init();
2388                 break;
2389         case MOD_QUIESCE:
2390                 /*
2391                  * Module should not be unloaded due to race conditions.
2392                  */
2393                 error = EBUSY;
2394                 break;
2395         case MOD_UNLOAD:
2396                 pfsync_uninit();
2397                 break;
2398         default:
2399                 error = EINVAL;
2400                 break;
2401         }
2402
2403         return (error);
2404 }
2405
2406 static moduledata_t pfsync_mod = {
2407         pfsyncname,
2408         pfsync_modevent,
2409         0
2410 };
2411
2412 #define PFSYNC_MODVER 1
2413
2414 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY);
2415 MODULE_VERSION(pfsync, PFSYNC_MODVER);
2416 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);