xref: /xnu-8796.141.3/bsd/netinet/ip_output.c (revision 1b191cb58250d0705d8a51287127505aa4bc0789)
1 /*
2  * Copyright (c) 2000-2021 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 /*
29  * Copyright (c) 1982, 1986, 1988, 1990, 1993
30  *	The Regents of the University of California.  All rights reserved.
31  *
32  * Redistribution and use in source and binary forms, with or without
33  * modification, are permitted provided that the following conditions
34  * are met:
35  * 1. Redistributions of source code must retain the above copyright
36  *    notice, this list of conditions and the following disclaimer.
37  * 2. Redistributions in binary form must reproduce the above copyright
38  *    notice, this list of conditions and the following disclaimer in the
39  *    documentation and/or other materials provided with the distribution.
40  * 3. All advertising materials mentioning features or use of this software
41  *    must display the following acknowledgement:
42  *	This product includes software developed by the University of
43  *	California, Berkeley and its contributors.
44  * 4. Neither the name of the University nor the names of its contributors
45  *    may be used to endorse or promote products derived from this software
46  *    without specific prior written permission.
47  *
48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  *
60  *	@(#)ip_output.c	8.3 (Berkeley) 1/21/94
61  */
62 /*
63  * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
64  * support for mandatory and extensible security protections.  This notice
65  * is included in support of clause 2.2 (b) of the Apple Public License,
66  * Version 2.0.
67  */
68 
69 #define _IP_VHL
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/kernel.h>
74 #include <sys/malloc.h>
75 #include <sys/mbuf.h>
76 #include <sys/protosw.h>
77 #include <sys/socket.h>
78 #include <sys/socketvar.h>
79 #include <kern/locks.h>
80 #include <sys/sysctl.h>
81 #include <sys/mcache.h>
82 #include <sys/kdebug.h>
83 
84 #include <machine/endian.h>
85 #include <pexpert/pexpert.h>
86 #include <mach/sdt.h>
87 
88 #include <libkern/OSAtomic.h>
89 #include <libkern/OSByteOrder.h>
90 
91 #include <net/if.h>
92 #include <net/if_dl.h>
93 #include <net/if_types.h>
94 #include <net/route.h>
95 #include <net/ntstat.h>
96 #include <net/net_osdep.h>
97 #include <net/dlil.h>
98 #include <net/net_perf.h>
99 
100 #include <netinet/in.h>
101 #include <netinet/in_systm.h>
102 #include <netinet/ip.h>
103 #include <netinet/in_pcb.h>
104 #include <netinet/in_var.h>
105 #include <netinet/ip_var.h>
106 #include <netinet/kpi_ipfilter_var.h>
107 #include <netinet/in_tclass.h>
108 #include <netinet/udp.h>
109 
110 #include <netinet6/nd6.h>
111 
112 #define DBG_LAYER_BEG           NETDBG_CODE(DBG_NETIP, 1)
113 #define DBG_LAYER_END           NETDBG_CODE(DBG_NETIP, 3)
114 #define DBG_FNC_IP_OUTPUT       NETDBG_CODE(DBG_NETIP, (1 << 8) | 1)
115 #define DBG_FNC_IPSEC4_OUTPUT   NETDBG_CODE(DBG_NETIP, (2 << 8) | 1)
116 
117 #if IPSEC
118 #include <netinet6/ipsec.h>
119 #include <netkey/key.h>
120 #if IPSEC_DEBUG
121 #include <netkey/key_debug.h>
122 #else
123 #define KEYDEBUG(lev, arg)
124 #endif
125 #endif /* IPSEC */
126 
127 #if NECP
128 #include <net/necp.h>
129 #endif /* NECP */
130 
131 
132 #if DUMMYNET
133 #include <netinet/ip_dummynet.h>
134 #endif
135 
136 #if PF
137 #include <net/pfvar.h>
138 #endif /* PF */
139 
140 
141 u_short ip_id;
142 
143 static int sysctl_reset_ip_output_stats SYSCTL_HANDLER_ARGS;
144 static int sysctl_ip_output_measure_bins SYSCTL_HANDLER_ARGS;
145 static int sysctl_ip_output_getperf SYSCTL_HANDLER_ARGS;
146 static void ip_out_cksum_stats(int, u_int32_t);
147 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
148 static int ip_optcopy(struct ip *, struct ip *);
149 static int ip_pcbopts(int, struct mbuf **, struct mbuf *);
150 static void imo_trace(struct ip_moptions *, int);
151 static void ip_mloopback(struct ifnet *, struct ifnet *, struct mbuf *,
152     struct sockaddr_in *, int);
153 static struct ifaddr *in_selectsrcif(struct ip *, struct route *, unsigned int);
154 
155 extern struct ip_linklocal_stat ip_linklocal_stat;
156 
157 /* temporary: for testing */
158 #if IPSEC
159 extern int ipsec_bypass;
160 #endif
161 
162 static int ip_maxchainsent = 0;
163 SYSCTL_INT(_net_inet_ip, OID_AUTO, maxchainsent,
164     CTLFLAG_RW | CTLFLAG_LOCKED, &ip_maxchainsent, 0,
165     "use dlil_output_list");
166 #if DEBUG
167 static int forge_ce = 0;
168 SYSCTL_INT(_net_inet_ip, OID_AUTO, forge_ce,
169     CTLFLAG_RW | CTLFLAG_LOCKED, &forge_ce, 0,
170     "Forge ECN CE");
171 #endif /* DEBUG */
172 
173 static int ip_select_srcif_debug = 0;
174 SYSCTL_INT(_net_inet_ip, OID_AUTO, select_srcif_debug,
175     CTLFLAG_RW | CTLFLAG_LOCKED, &ip_select_srcif_debug, 0,
176     "log source interface selection debug info");
177 
178 static int ip_output_measure = 0;
179 SYSCTL_PROC(_net_inet_ip, OID_AUTO, output_perf,
180     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
181     &ip_output_measure, 0, sysctl_reset_ip_output_stats, "I",
182     "Do time measurement");
183 
184 static uint64_t ip_output_measure_bins = 0;
185 SYSCTL_PROC(_net_inet_ip, OID_AUTO, output_perf_bins,
186     CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED, &ip_output_measure_bins, 0,
187     sysctl_ip_output_measure_bins, "I",
188     "bins for chaining performance data histogram");
189 
190 static net_perf_t net_perf;
191 SYSCTL_PROC(_net_inet_ip, OID_AUTO, output_perf_data,
192     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED,
193     0, 0, sysctl_ip_output_getperf, "S,net_perf",
194     "IP output performance data (struct net_perf, net/net_perf.h)");
195 
196 __private_extern__ int rfc6864 = 1;
197 SYSCTL_INT(_net_inet_ip, OID_AUTO, rfc6864, CTLFLAG_RW | CTLFLAG_LOCKED,
198     &rfc6864, 0, "updated ip id field behavior");
199 
200 #define IMO_TRACE_HIST_SIZE     32      /* size of trace history */
201 
202 /* For gdb */
203 __private_extern__ unsigned int imo_trace_hist_size = IMO_TRACE_HIST_SIZE;
204 
205 struct ip_moptions_dbg {
206 	struct ip_moptions      imo;                    /* ip_moptions */
207 	u_int16_t               imo_refhold_cnt;        /* # of IMO_ADDREF */
208 	u_int16_t               imo_refrele_cnt;        /* # of IMO_REMREF */
209 	/*
210 	 * Alloc and free callers.
211 	 */
212 	ctrace_t                imo_alloc;
213 	ctrace_t                imo_free;
214 	/*
215 	 * Circular lists of IMO_ADDREF and IMO_REMREF callers.
216 	 */
217 	ctrace_t                imo_refhold[IMO_TRACE_HIST_SIZE];
218 	ctrace_t                imo_refrele[IMO_TRACE_HIST_SIZE];
219 };
220 
221 #if DEBUG
222 static unsigned int imo_debug = 1;      /* debugging (enabled) */
223 #else
224 static unsigned int imo_debug;          /* debugging (disabled) */
225 #endif /* !DEBUG */
226 
227 static struct zone *imo_zone;           /* zone for ip_moptions */
228 #define IMO_ZONE_NAME           "ip_moptions"   /* zone name */
229 
230 #if PF
231 __attribute__((noinline))
232 static int
ip_output_pf_dn_hook(struct ifnet * ifp,struct mbuf ** mppn,struct mbuf ** mp,struct pf_rule * dn_pf_rule,struct route * ro,struct sockaddr_in * dst,int flags,struct ip_out_args * ipoa)233 ip_output_pf_dn_hook(struct ifnet *ifp, struct mbuf **mppn, struct mbuf **mp,
234     struct pf_rule *dn_pf_rule, struct route *ro, struct sockaddr_in *dst, int flags,
235     struct ip_out_args *ipoa)
236 {
237 	int rc;
238 	struct ip_fw_args args = {};
239 
240 	args.fwa_pf_rule = dn_pf_rule;
241 	args.fwa_oif = ifp;
242 	args.fwa_ro = ro;
243 	args.fwa_dst = dst;
244 	args.fwa_oflags = flags;
245 	if (flags & IP_OUTARGS) {
246 		args.fwa_ipoa = ipoa;
247 	}
248 	rc = pf_af_hook(ifp, mppn, mp, AF_INET, FALSE, &args);
249 
250 	return rc;
251 }
252 
253 #endif /* PF */
254 
255 
256 /*
257  * IP output.  The packet in mbuf chain m contains a skeletal IP
258  * header (with len, off, ttl, proto, tos, src, dst).
259  * The mbuf chain containing the packet will be freed.
260  * The mbuf opt, if present, will not be freed.
261  */
262 int
ip_output(struct mbuf * m0,struct mbuf * opt,struct route * ro,int flags,struct ip_moptions * imo,struct ip_out_args * ipoa)263 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro, int flags,
264     struct ip_moptions *imo, struct ip_out_args *ipoa)
265 {
266 	return ip_output_list(m0, 0, opt, ro, flags, imo, ipoa);
267 }
268 
269 /*
270  * IP output.  The packet in mbuf chain m contains a skeletal IP
271  * header (with len, off, ttl, proto, tos, src, dst).
272  * The mbuf chain containing the packet will be freed.
273  * The mbuf opt, if present, will not be freed.
274  *
275  * Route ro MUST be non-NULL; if ro->ro_rt is valid, route lookup would be
276  * skipped and ro->ro_rt would be used.  Otherwise the result of route
277  * lookup is stored in ro->ro_rt.
278  *
279  * In the IP forwarding case, the packet will arrive with options already
280  * inserted, so must have a NULL opt pointer.
281  */
282 int
ip_output_list(struct mbuf * m0,int packetchain,struct mbuf * opt,struct route * ro,int flags,struct ip_moptions * imo,struct ip_out_args * ipoa)283 ip_output_list(struct mbuf *m0, int packetchain, struct mbuf *opt,
284     struct route *ro, int flags, struct ip_moptions *imo,
285     struct ip_out_args *ipoa)
286 {
287 	struct ip *ip;
288 	struct ifnet *ifp = NULL;               /* not refcnt'd */
289 	struct mbuf *m = m0, *prevnxt = NULL, **mppn = &prevnxt;
290 	int hlen = sizeof(struct ip);
291 	int len = 0, error = 0;
292 	struct sockaddr_in *dst = NULL;
293 	struct in_ifaddr *ia = NULL, *src_ia = NULL;
294 	struct in_addr pkt_dst;
295 	struct ipf_pktopts *ippo = NULL;
296 	ipfilter_t inject_filter_ref = NULL;
297 	struct mbuf *packetlist;
298 	uint32_t sw_csum, pktcnt = 0, scnt = 0, bytecnt = 0;
299 	uint32_t packets_processed = 0;
300 	unsigned int ifscope = IFSCOPE_NONE;
301 	struct flowadv *adv = NULL;
302 	struct timeval start_tv;
303 #if IPSEC
304 	struct socket *so = NULL;
305 	struct secpolicy *sp = NULL;
306 #endif /* IPSEC */
307 #if NECP
308 	necp_kernel_policy_result necp_result = 0;
309 	necp_kernel_policy_result_parameter necp_result_parameter;
310 	necp_kernel_policy_id necp_matched_policy_id = 0;
311 #endif /* NECP */
312 #if DUMMYNET
313 	struct m_tag *tag;
314 	struct ip_out_args saved_ipoa;
315 	struct sockaddr_in dst_buf;
316 #endif /* DUMMYNET */
317 	struct {
318 #if IPSEC
319 		struct ipsec_output_state ipsec_state;
320 #endif /* IPSEC */
321 #if NECP
322 		struct route necp_route;
323 #endif /* NECP */
324 #if DUMMYNET
325 		struct route saved_route;
326 #endif /* DUMMYNET */
327 		struct ipf_pktopts ipf_pktopts;
328 	} ipobz;
329 #define ipsec_state     ipobz.ipsec_state
330 #define necp_route      ipobz.necp_route
331 #define sro_fwd         ipobz.sro_fwd
332 #define saved_route     ipobz.saved_route
333 #define ipf_pktopts     ipobz.ipf_pktopts
334 	union {
335 		struct {
336 			boolean_t select_srcif : 1;     /* set once */
337 			boolean_t srcbound : 1;         /* set once */
338 			boolean_t nocell : 1;           /* set once */
339 			boolean_t isbroadcast : 1;
340 			boolean_t didfilter : 1;
341 			boolean_t noexpensive : 1;      /* set once */
342 			boolean_t noconstrained : 1;      /* set once */
343 			boolean_t awdl_unrestricted : 1;        /* set once */
344 			boolean_t management_allowed : 1;        /* set once */
345 		};
346 		uint32_t raw;
347 	} ipobf = { .raw = 0 };
348 
349 	int interface_mtu = 0;
350 	struct pf_rule *dn_pf_rule = NULL;
351 /*
352  * Here we check for restrictions when sending frames.
353  * N.B.: IPv4 over internal co-processor interfaces is not allowed.
354  */
355 #define IP_CHECK_RESTRICTIONS(_ifp, _ipobf)                                 \
356 	(((_ipobf).nocell && IFNET_IS_CELLULAR(_ifp)) ||                    \
357 	 ((_ipobf).noexpensive && IFNET_IS_EXPENSIVE(_ifp)) ||              \
358 	 ((_ipobf).noconstrained && IFNET_IS_CONSTRAINED(_ifp)) ||          \
359 	  (IFNET_IS_INTCOPROC(_ifp)) ||                                     \
360 	 (!(_ipobf).management_allowed && IFNET_IS_MANAGEMENT(_ifp)) ||     \
361 	 (!(_ipobf).awdl_unrestricted && IFNET_IS_AWDL_RESTRICTED(_ifp)))
362 
363 	if (ip_output_measure) {
364 		net_perf_start_time(&net_perf, &start_tv);
365 	}
366 	KERNEL_DEBUG(DBG_FNC_IP_OUTPUT | DBG_FUNC_START, 0, 0, 0, 0, 0);
367 
368 	VERIFY(m0->m_flags & M_PKTHDR);
369 	packetlist = m0;
370 
371 	/* zero out {ipsec_state, args, sro_fwd, saved_route, ipf_pktops} */
372 	bzero(&ipobz, sizeof(ipobz));
373 	ippo = &ipf_pktopts;
374 
375 #if DUMMYNET
376 	if (SLIST_EMPTY(&m0->m_pkthdr.tags)) {
377 		goto ipfw_tags_done;
378 	}
379 
380 	/* Grab info from mtags prepended to the chain */
381 	if ((tag = m_tag_locate(m0, KERNEL_MODULE_TAG_ID,
382 	    KERNEL_TAG_TYPE_DUMMYNET, NULL)) != NULL) {
383 		struct dn_pkt_tag       *dn_tag;
384 
385 		dn_tag = (struct dn_pkt_tag *)(tag + 1);
386 		dn_pf_rule = dn_tag->dn_pf_rule;
387 		opt = NULL;
388 		saved_route = dn_tag->dn_ro;
389 		ro = &saved_route;
390 
391 		imo = NULL;
392 		bcopy(&dn_tag->dn_dst, &dst_buf, sizeof(dst_buf));
393 		dst = &dst_buf;
394 		ifp = dn_tag->dn_ifp;
395 		flags = dn_tag->dn_flags;
396 		if ((dn_tag->dn_flags & IP_OUTARGS)) {
397 			saved_ipoa = dn_tag->dn_ipoa;
398 			ipoa = &saved_ipoa;
399 		}
400 
401 		m_tag_delete(m0, tag);
402 	}
403 ipfw_tags_done:
404 #endif /* DUMMYNET */
405 
406 	m = m0;
407 	m->m_pkthdr.pkt_flags &= ~(PKTF_LOOP | PKTF_IFAINFO);
408 
409 #if IPSEC
410 	if (ipsec_bypass == 0 && !(flags & IP_NOIPSEC)) {
411 		/* If packet is bound to an interface, check bound policies */
412 		if ((flags & IP_OUTARGS) && (ipoa != NULL) &&
413 		    (ipoa->ipoa_flags & IPOAF_BOUND_IF) &&
414 		    ipoa->ipoa_boundif != IFSCOPE_NONE) {
415 			if (ipsec4_getpolicybyinterface(m, IPSEC_DIR_OUTBOUND,
416 			    &flags, ipoa, &sp) != 0) {
417 				goto bad;
418 			}
419 		}
420 	}
421 #endif /* IPSEC */
422 
423 	VERIFY(ro != NULL);
424 
425 	if (flags & IP_OUTARGS) {
426 		/*
427 		 * In the forwarding case, only the ifscope value is used,
428 		 * as source interface selection doesn't take place.
429 		 */
430 		if ((ipobf.select_srcif = (!(flags & IP_FORWARDING) &&
431 		    (ipoa->ipoa_flags & IPOAF_SELECT_SRCIF)))) {
432 			ipf_pktopts.ippo_flags |= IPPOF_SELECT_SRCIF;
433 		}
434 
435 		if ((ipoa->ipoa_flags & IPOAF_BOUND_IF) &&
436 		    ipoa->ipoa_boundif != IFSCOPE_NONE) {
437 			ifscope = ipoa->ipoa_boundif;
438 			ipf_pktopts.ippo_flags |=
439 			    (IPPOF_BOUND_IF | (ifscope << IPPOF_SHIFT_IFSCOPE));
440 		}
441 
442 		/* double negation needed for bool bit field */
443 		ipobf.srcbound = !!(ipoa->ipoa_flags & IPOAF_BOUND_SRCADDR);
444 		if (ipobf.srcbound) {
445 			ipf_pktopts.ippo_flags |= IPPOF_BOUND_SRCADDR;
446 		}
447 	} else {
448 		ipobf.select_srcif = FALSE;
449 		ipobf.srcbound = FALSE;
450 		ifscope = IFSCOPE_NONE;
451 		if (flags & IP_OUTARGS) {
452 			ipoa->ipoa_boundif = IFSCOPE_NONE;
453 			ipoa->ipoa_flags &= ~(IPOAF_SELECT_SRCIF |
454 			    IPOAF_BOUND_IF | IPOAF_BOUND_SRCADDR);
455 		}
456 	}
457 
458 	if (flags & IP_OUTARGS) {
459 		if (ipoa->ipoa_flags & IPOAF_NO_CELLULAR) {
460 			ipobf.nocell = TRUE;
461 			ipf_pktopts.ippo_flags |= IPPOF_NO_IFT_CELLULAR;
462 		}
463 		if (ipoa->ipoa_flags & IPOAF_NO_EXPENSIVE) {
464 			ipobf.noexpensive = TRUE;
465 			ipf_pktopts.ippo_flags |= IPPOF_NO_IFF_EXPENSIVE;
466 		}
467 		if (ipoa->ipoa_flags & IPOAF_NO_CONSTRAINED) {
468 			ipobf.noconstrained = TRUE;
469 			ipf_pktopts.ippo_flags |= IPPOF_NO_IFF_CONSTRAINED;
470 		}
471 		if (ipoa->ipoa_flags & IPOAF_AWDL_UNRESTRICTED) {
472 			ipobf.awdl_unrestricted = TRUE;
473 		}
474 		if (ipoa->ipoa_flags & IPOAF_MANAGEMENT_ALLOWED) {
475 			ipobf.management_allowed = true;
476 		}
477 		adv = &ipoa->ipoa_flowadv;
478 		adv->code = FADV_SUCCESS;
479 		ipoa->ipoa_flags &= ~IPOAF_RET_MASK;
480 	}
481 
482 #if IPSEC
483 	if (ipsec_bypass == 0 && !(flags & IP_NOIPSEC)) {
484 		so = ipsec_getsocket(m);
485 		if (so != NULL) {
486 			(void) ipsec_setsocket(m, NULL);
487 		}
488 	}
489 #endif /* IPSEC */
490 
491 #if DUMMYNET
492 	if (dn_pf_rule != NULL) {
493 		/* dummynet already saw us */
494 		ip = mtod(m, struct ip *);
495 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
496 		pkt_dst = ip->ip_dst;
497 		if (ro->ro_rt != NULL) {
498 			RT_LOCK_SPIN(ro->ro_rt);
499 			ia = (struct in_ifaddr *)ro->ro_rt->rt_ifa;
500 			if (ia) {
501 				/* Become a regular mutex */
502 				RT_CONVERT_LOCK(ro->ro_rt);
503 				IFA_ADDREF(&ia->ia_ifa);
504 			}
505 			RT_UNLOCK(ro->ro_rt);
506 		}
507 
508 		goto sendit;
509 	}
510 #endif /* DUMMYNET */
511 
512 loopit:
513 	packets_processed++;
514 	ipobf.isbroadcast = FALSE;
515 	ipobf.didfilter = FALSE;
516 
517 	VERIFY(m->m_flags & M_PKTHDR);
518 	/*
519 	 * No need to proccess packet twice if we've already seen it.
520 	 */
521 	if (!SLIST_EMPTY(&m->m_pkthdr.tags)) {
522 		inject_filter_ref = ipf_get_inject_filter(m);
523 	} else {
524 		inject_filter_ref = NULL;
525 	}
526 
527 	if (opt) {
528 		m = ip_insertoptions(m, opt, &len);
529 		hlen = len;
530 		/* Update the chain */
531 		if (m != m0) {
532 			if (m0 == packetlist) {
533 				packetlist = m;
534 			}
535 			m0 = m;
536 		}
537 	}
538 	ip = mtod(m, struct ip *);
539 
540 	pkt_dst = ip->ip_dst;
541 
542 	/*
543 	 * We must not send if the packet is destined to network zero.
544 	 * RFC1122 3.2.1.3 (a) and (b).
545 	 */
546 	if (IN_ZERONET(ntohl(pkt_dst.s_addr))) {
547 		error = EHOSTUNREACH;
548 		goto bad;
549 	}
550 
551 	/*
552 	 * Fill in IP header.
553 	 */
554 	if (!(flags & (IP_FORWARDING | IP_RAWOUTPUT))) {
555 		ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
556 		ip->ip_off &= IP_DF;
557 		if (rfc6864 && IP_OFF_IS_ATOMIC(ip->ip_off)) {
558 			// Per RFC6864, value of ip_id is undefined for atomic ip packets
559 			ip->ip_id = 0;
560 		} else {
561 			ip->ip_id = ip_randomid((uint64_t)m);
562 		}
563 		OSAddAtomic(1, &ipstat.ips_localout);
564 	} else {
565 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
566 	}
567 
568 #if DEBUG
569 	/* For debugging, we let the stack forge congestion */
570 	if (forge_ce != 0 &&
571 	    ((ip->ip_tos & IPTOS_ECN_MASK) == IPTOS_ECN_ECT1 ||
572 	    (ip->ip_tos & IPTOS_ECN_MASK) == IPTOS_ECN_ECT0)) {
573 		ip->ip_tos = (ip->ip_tos & ~IPTOS_ECN_MASK) | IPTOS_ECN_CE;
574 		forge_ce--;
575 	}
576 #endif /* DEBUG */
577 
578 	if ((ip->ip_tos & IPTOS_ECN_MASK) == IPTOS_ECN_ECT1) {
579 		m->m_pkthdr.pkt_ext_flags |= PKTF_EXT_L4S;
580 	}
581 
582 	KERNEL_DEBUG(DBG_LAYER_BEG, ip->ip_dst.s_addr, ip->ip_src.s_addr,
583 	    ip->ip_p, ip->ip_off, ip->ip_len);
584 
585 	dst = SIN(&ro->ro_dst);
586 
587 	/*
588 	 * If there is a cached route,
589 	 * check that it is to the same destination
590 	 * and is still up.  If not, free it and try again.
591 	 * The address family should also be checked in case of sharing the
592 	 * cache with IPv6.
593 	 */
594 
595 	if (ro->ro_rt != NULL) {
596 		if (ROUTE_UNUSABLE(ro) && ip->ip_src.s_addr != INADDR_ANY &&
597 		    !(flags & (IP_ROUTETOIF | IP_FORWARDING))) {
598 			src_ia = ifa_foraddr(ip->ip_src.s_addr);
599 			if (src_ia == NULL) {
600 				error = EADDRNOTAVAIL;
601 				goto bad;
602 			}
603 			IFA_REMREF(&src_ia->ia_ifa);
604 			src_ia = NULL;
605 		}
606 		/*
607 		 * Test rt_flags without holding rt_lock for performance
608 		 * reasons; if the route is down it will hopefully be
609 		 * caught by the layer below (since it uses this route
610 		 * as a hint) or during the next transmit.
611 		 */
612 		if (ROUTE_UNUSABLE(ro) || dst->sin_family != AF_INET ||
613 		    dst->sin_addr.s_addr != pkt_dst.s_addr) {
614 			ROUTE_RELEASE(ro);
615 		}
616 
617 		/*
618 		 * If we're doing source interface selection, we may not
619 		 * want to use this route; only synch up the generation
620 		 * count otherwise.
621 		 */
622 		if (!ipobf.select_srcif && ro->ro_rt != NULL &&
623 		    RT_GENID_OUTOFSYNC(ro->ro_rt)) {
624 			RT_GENID_SYNC(ro->ro_rt);
625 		}
626 	}
627 	if (ro->ro_rt == NULL) {
628 		bzero(dst, sizeof(*dst));
629 		dst->sin_family = AF_INET;
630 		dst->sin_len = sizeof(*dst);
631 		dst->sin_addr = pkt_dst;
632 	}
633 	/*
634 	 * If routing to interface only,
635 	 * short circuit routing lookup.
636 	 */
637 	if (flags & IP_ROUTETOIF) {
638 		if (ia != NULL) {
639 			IFA_REMREF(&ia->ia_ifa);
640 		}
641 		if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL) {
642 			ia = ifatoia(ifa_ifwithnet(sintosa(dst)));
643 			if (ia == NULL) {
644 				OSAddAtomic(1, &ipstat.ips_noroute);
645 				error = ENETUNREACH;
646 				/* XXX IPv6 APN fallback notification?? */
647 				goto bad;
648 			}
649 		}
650 		ifp = ia->ia_ifp;
651 		ip->ip_ttl = 1;
652 		ipobf.isbroadcast = in_broadcast(dst->sin_addr, ifp);
653 		/*
654 		 * For consistency with other cases below.  Loopback
655 		 * multicast case is handled separately by ip_mloopback().
656 		 */
657 		if ((ifp->if_flags & IFF_LOOPBACK) &&
658 		    !IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
659 			m->m_pkthdr.rcvif = ifp;
660 			ip_setsrcifaddr_info(m, ifp->if_index, NULL);
661 			ip_setdstifaddr_info(m, ifp->if_index, NULL);
662 		}
663 	} else if (IN_MULTICAST(ntohl(pkt_dst.s_addr)) &&
664 	    imo != NULL && (ifp = imo->imo_multicast_ifp) != NULL) {
665 		/*
666 		 * Bypass the normal routing lookup for multicast
667 		 * packets if the interface is specified.
668 		 */
669 		ipobf.isbroadcast = FALSE;
670 		if (ia != NULL) {
671 			IFA_REMREF(&ia->ia_ifa);
672 		}
673 
674 		/* Macro takes reference on ia */
675 		IFP_TO_IA(ifp, ia);
676 	} else {
677 		struct ifaddr *ia0 = NULL;
678 		boolean_t cloneok = FALSE;
679 		/*
680 		 * Perform source interface selection; the source IP address
681 		 * must belong to one of the addresses of the interface used
682 		 * by the route.  For performance reasons, do this only if
683 		 * there is no route, or if the routing table has changed,
684 		 * or if we haven't done source interface selection on this
685 		 * route (for this PCB instance) before.
686 		 */
687 		if (ipobf.select_srcif &&
688 		    ip->ip_src.s_addr != INADDR_ANY && (ROUTE_UNUSABLE(ro) ||
689 		    !(ro->ro_flags & ROF_SRCIF_SELECTED))) {
690 			/* Find the source interface */
691 			ia0 = in_selectsrcif(ip, ro, ifscope);
692 
693 			/*
694 			 * If the source address belongs to a restricted
695 			 * interface and the caller forbids our using
696 			 * interfaces of such type, pretend that there is no
697 			 * route.
698 			 */
699 			if (ia0 != NULL &&
700 			    IP_CHECK_RESTRICTIONS(ia0->ifa_ifp, ipobf)) {
701 				IFA_REMREF(ia0);
702 				ia0 = NULL;
703 				error = EHOSTUNREACH;
704 				if (flags & IP_OUTARGS) {
705 					ipoa->ipoa_flags |= IPOAF_R_IFDENIED;
706 				}
707 				goto bad;
708 			}
709 
710 			/*
711 			 * If the source address is spoofed (in the case of
712 			 * IP_RAWOUTPUT on an unbounded socket), or if this
713 			 * is destined for local/loopback, just let it go out
714 			 * using the interface of the route.  Otherwise,
715 			 * there's no interface having such an address,
716 			 * so bail out.
717 			 */
718 			if (ia0 == NULL && (!(flags & IP_RAWOUTPUT) ||
719 			    ipobf.srcbound) && ifscope != lo_ifp->if_index) {
720 				error = EADDRNOTAVAIL;
721 				goto bad;
722 			}
723 
724 			/*
725 			 * If the caller didn't explicitly specify the scope,
726 			 * pick it up from the source interface.  If the cached
727 			 * route was wrong and was blown away as part of source
728 			 * interface selection, don't mask out RTF_PRCLONING
729 			 * since that route may have been allocated by the ULP,
730 			 * unless the IP header was created by the caller or
731 			 * the destination is IPv4 LLA.  The check for the
732 			 * latter is needed because IPv4 LLAs are never scoped
733 			 * in the current implementation, and we don't want to
734 			 * replace the resolved IPv4 LLA route with one whose
735 			 * gateway points to that of the default gateway on
736 			 * the primary interface of the system.
737 			 */
738 			if (ia0 != NULL) {
739 				if (ifscope == IFSCOPE_NONE) {
740 					ifscope = ia0->ifa_ifp->if_index;
741 				}
742 				cloneok = (!(flags & IP_RAWOUTPUT) &&
743 				    !(IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))));
744 			}
745 		}
746 
747 		/*
748 		 * If this is the case, we probably don't want to allocate
749 		 * a protocol-cloned route since we didn't get one from the
750 		 * ULP.  This lets TCP do its thing, while not burdening
751 		 * forwarding or ICMP with the overhead of cloning a route.
752 		 * Of course, we still want to do any cloning requested by
753 		 * the link layer, as this is probably required in all cases
754 		 * for correct operation (as it is for ARP).
755 		 */
756 		if (ro->ro_rt == NULL) {
757 			uint32_t ign = RTF_PRCLONING;
758 			/*
759 			 * We make an exception here: if the destination
760 			 * address is INADDR_BROADCAST, allocate a protocol-
761 			 * cloned host route so that we end up with a route
762 			 * marked with the RTF_BROADCAST flag.  Otherwise,
763 			 * we would end up referring to the default route,
764 			 * instead of creating a cloned host route entry.
765 			 * That would introduce inconsistencies between ULPs
766 			 * that allocate a route and those that don't.  The
767 			 * RTF_BROADCAST route is important since we'd want
768 			 * to send out undirected IP broadcast packets using
769 			 * link-level broadcast address. Another exception
770 			 * is for ULP-created routes that got blown away by
771 			 * source interface selection (see above).
772 			 *
773 			 * These exceptions will no longer be necessary when
774 			 * the RTF_PRCLONING scheme is no longer present.
775 			 */
776 			if (cloneok || dst->sin_addr.s_addr == INADDR_BROADCAST) {
777 				ign &= ~RTF_PRCLONING;
778 			}
779 
780 			/*
781 			 * Loosen the route lookup criteria if the ifscope
782 			 * corresponds to the loopback interface; this is
783 			 * needed to support Application Layer Gateways
784 			 * listening on loopback, in conjunction with packet
785 			 * filter redirection rules.  The final source IP
786 			 * address will be rewritten by the packet filter
787 			 * prior to the RFC1122 loopback check below.
788 			 */
789 			if (ifscope == lo_ifp->if_index) {
790 				rtalloc_ign(ro, ign);
791 			} else {
792 				rtalloc_scoped_ign(ro, ign, ifscope);
793 			}
794 
795 			/*
796 			 * If the route points to a cellular/expensive interface
797 			 * and the caller forbids our using interfaces of such type,
798 			 * pretend that there is no route.
799 			 */
800 			if (ro->ro_rt != NULL) {
801 				RT_LOCK_SPIN(ro->ro_rt);
802 				if (IP_CHECK_RESTRICTIONS(ro->ro_rt->rt_ifp,
803 				    ipobf)) {
804 					RT_UNLOCK(ro->ro_rt);
805 					ROUTE_RELEASE(ro);
806 					if (flags & IP_OUTARGS) {
807 						ipoa->ipoa_flags |=
808 						    IPOAF_R_IFDENIED;
809 					}
810 				} else {
811 					RT_UNLOCK(ro->ro_rt);
812 				}
813 			}
814 		}
815 
816 		if (ro->ro_rt == NULL) {
817 			OSAddAtomic(1, &ipstat.ips_noroute);
818 			error = EHOSTUNREACH;
819 			if (ia0 != NULL) {
820 				IFA_REMREF(ia0);
821 				ia0 = NULL;
822 			}
823 			goto bad;
824 		}
825 
826 		if (ia != NULL) {
827 			IFA_REMREF(&ia->ia_ifa);
828 		}
829 		RT_LOCK_SPIN(ro->ro_rt);
830 		ia = ifatoia(ro->ro_rt->rt_ifa);
831 		if (ia != NULL) {
832 			/* Become a regular mutex */
833 			RT_CONVERT_LOCK(ro->ro_rt);
834 			IFA_ADDREF(&ia->ia_ifa);
835 		}
836 		/*
837 		 * Note: ia_ifp may not be the same as rt_ifp; the latter
838 		 * is what we use for determining outbound i/f, mtu, etc.
839 		 */
840 		ifp = ro->ro_rt->rt_ifp;
841 		ro->ro_rt->rt_use++;
842 		if (ro->ro_rt->rt_flags & RTF_GATEWAY) {
843 			dst = SIN(ro->ro_rt->rt_gateway);
844 		}
845 		if (ro->ro_rt->rt_flags & RTF_HOST) {
846 			/* double negation needed for bool bit field */
847 			ipobf.isbroadcast =
848 			    !!(ro->ro_rt->rt_flags & RTF_BROADCAST);
849 		} else {
850 			/* Become a regular mutex */
851 			RT_CONVERT_LOCK(ro->ro_rt);
852 			ipobf.isbroadcast = in_broadcast(dst->sin_addr, ifp);
853 		}
854 		/*
855 		 * For consistency with IPv6, as well as to ensure that
856 		 * IP_RECVIF is set correctly for packets that are sent
857 		 * to one of the local addresses.  ia (rt_ifa) would have
858 		 * been fixed up by rt_setif for local routes.  This
859 		 * would make it appear as if the packet arrives on the
860 		 * interface which owns the local address.  Loopback
861 		 * multicast case is handled separately by ip_mloopback().
862 		 */
863 		if (ia != NULL && (ifp->if_flags & IFF_LOOPBACK) &&
864 		    !IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
865 			uint16_t srcidx;
866 
867 			m->m_pkthdr.rcvif = ia->ia_ifa.ifa_ifp;
868 
869 			if (ia0 != NULL) {
870 				srcidx = ia0->ifa_ifp->if_index;
871 			} else if ((ro->ro_flags & ROF_SRCIF_SELECTED) &&
872 			    ro->ro_srcia != NULL) {
873 				srcidx = ro->ro_srcia->ifa_ifp->if_index;
874 			} else {
875 				srcidx = 0;
876 			}
877 
878 			ip_setsrcifaddr_info(m, srcidx, NULL);
879 			ip_setdstifaddr_info(m, 0, ia);
880 		}
881 		RT_UNLOCK(ro->ro_rt);
882 		if (ia0 != NULL) {
883 			IFA_REMREF(ia0);
884 			ia0 = NULL;
885 		}
886 	}
887 
888 	if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
889 		struct ifnet *srcifp = NULL;
890 		struct in_multi *inm;
891 		u_int32_t vif = 0;
892 		u_int8_t ttl = IP_DEFAULT_MULTICAST_TTL;
893 		u_int8_t loop = IP_DEFAULT_MULTICAST_LOOP;
894 
895 		m->m_flags |= M_MCAST;
896 		/*
897 		 * IP destination address is multicast.  Make sure "dst"
898 		 * still points to the address in "ro".  (It may have been
899 		 * changed to point to a gateway address, above.)
900 		 */
901 		dst = SIN(&ro->ro_dst);
902 		/*
903 		 * See if the caller provided any multicast options
904 		 */
905 		if (imo != NULL) {
906 			IMO_LOCK(imo);
907 			vif = imo->imo_multicast_vif;
908 			ttl = imo->imo_multicast_ttl;
909 			loop = imo->imo_multicast_loop;
910 			if (!(flags & IP_RAWOUTPUT)) {
911 				ip->ip_ttl = ttl;
912 			}
913 			if (imo->imo_multicast_ifp != NULL) {
914 				ifp = imo->imo_multicast_ifp;
915 			}
916 			IMO_UNLOCK(imo);
917 		} else if (!(flags & IP_RAWOUTPUT)) {
918 			vif = -1;
919 			ip->ip_ttl = ttl;
920 		}
921 		/*
922 		 * Confirm that the outgoing interface supports multicast.
923 		 */
924 		if (imo == NULL || vif == -1) {
925 			if (!(ifp->if_flags & IFF_MULTICAST)) {
926 				OSAddAtomic(1, &ipstat.ips_noroute);
927 				error = ENETUNREACH;
928 				goto bad;
929 			}
930 		}
931 		/*
932 		 * If source address not specified yet, use address
933 		 * of outgoing interface.
934 		 */
935 		if (ip->ip_src.s_addr == INADDR_ANY) {
936 			struct in_ifaddr *ia1;
937 			lck_rw_lock_shared(&in_ifaddr_rwlock);
938 			TAILQ_FOREACH(ia1, &in_ifaddrhead, ia_link) {
939 				IFA_LOCK_SPIN(&ia1->ia_ifa);
940 				if (ia1->ia_ifp == ifp) {
941 					ip->ip_src = IA_SIN(ia1)->sin_addr;
942 					srcifp = ifp;
943 					IFA_UNLOCK(&ia1->ia_ifa);
944 					break;
945 				}
946 				IFA_UNLOCK(&ia1->ia_ifa);
947 			}
948 			lck_rw_done(&in_ifaddr_rwlock);
949 			if (ip->ip_src.s_addr == INADDR_ANY) {
950 				error = ENETUNREACH;
951 				goto bad;
952 			}
953 		}
954 
955 		in_multihead_lock_shared();
956 		IN_LOOKUP_MULTI(&pkt_dst, ifp, inm);
957 		in_multihead_lock_done();
958 		if (inm != NULL && (imo == NULL || loop)) {
959 			/*
960 			 * If we belong to the destination multicast group
961 			 * on the outgoing interface, and the caller did not
962 			 * forbid loopback, loop back a copy.
963 			 */
964 			if (!TAILQ_EMPTY(&ipv4_filters)
965 #if NECP
966 			    && !necp_packet_should_skip_filters(m)
967 #endif // NECP
968 			    ) {
969 				struct ipfilter *filter;
970 				int seen = (inject_filter_ref == NULL);
971 
972 				if (imo != NULL) {
973 					ipf_pktopts.ippo_flags |=
974 					    IPPOF_MCAST_OPTS;
975 					ipf_pktopts.ippo_mcast_ifnet = ifp;
976 					ipf_pktopts.ippo_mcast_ttl = ttl;
977 					ipf_pktopts.ippo_mcast_loop = loop;
978 				}
979 
980 				ipf_ref();
981 
982 				/*
983 				 * 4135317 - always pass network byte
984 				 * order to filter
985 				 */
986 #if BYTE_ORDER != BIG_ENDIAN
987 				HTONS(ip->ip_len);
988 				HTONS(ip->ip_off);
989 #endif
990 				TAILQ_FOREACH(filter, &ipv4_filters, ipf_link) {
991 					if (seen == 0) {
992 						if ((struct ipfilter *)
993 						    inject_filter_ref == filter) {
994 							seen = 1;
995 						}
996 					} else if (filter->ipf_filter.
997 					    ipf_output != NULL) {
998 						errno_t result;
999 						result = filter->ipf_filter.
1000 						    ipf_output(filter->
1001 						    ipf_filter.cookie,
1002 						    (mbuf_t *)&m, ippo);
1003 						if (result == EJUSTRETURN) {
1004 							ipf_unref();
1005 							INM_REMREF(inm);
1006 							goto done;
1007 						}
1008 						if (result != 0) {
1009 							ipf_unref();
1010 							INM_REMREF(inm);
1011 							goto bad;
1012 						}
1013 					}
1014 				}
1015 
1016 				/* set back to host byte order */
1017 				ip = mtod(m, struct ip *);
1018 #if BYTE_ORDER != BIG_ENDIAN
1019 				NTOHS(ip->ip_len);
1020 				NTOHS(ip->ip_off);
1021 #endif
1022 				ipf_unref();
1023 				ipobf.didfilter = TRUE;
1024 			}
1025 			ip_mloopback(srcifp, ifp, m, dst, hlen);
1026 		}
1027 		if (inm != NULL) {
1028 			INM_REMREF(inm);
1029 		}
1030 		/*
1031 		 * Multicasts with a time-to-live of zero may be looped-
1032 		 * back, above, but must not be transmitted on a network.
1033 		 * Also, multicasts addressed to the loopback interface
1034 		 * are not sent -- the above call to ip_mloopback() will
1035 		 * loop back a copy if this host actually belongs to the
1036 		 * destination group on the loopback interface.
1037 		 */
1038 		if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
1039 			m_freem(m);
1040 			goto done;
1041 		}
1042 
1043 		goto sendit;
1044 	}
1045 	/*
1046 	 * If source address not specified yet, use address
1047 	 * of outgoing interface.
1048 	 */
1049 	if (ip->ip_src.s_addr == INADDR_ANY) {
1050 		IFA_LOCK_SPIN(&ia->ia_ifa);
1051 		ip->ip_src = IA_SIN(ia)->sin_addr;
1052 		IFA_UNLOCK(&ia->ia_ifa);
1053 	}
1054 
1055 	/*
1056 	 * Look for broadcast address and
1057 	 * and verify user is allowed to send
1058 	 * such a packet.
1059 	 */
1060 	if (ipobf.isbroadcast) {
1061 		if (!(ifp->if_flags & IFF_BROADCAST)) {
1062 			error = EADDRNOTAVAIL;
1063 			goto bad;
1064 		}
1065 		if (!(flags & IP_ALLOWBROADCAST)) {
1066 			error = EACCES;
1067 			goto bad;
1068 		}
1069 		/* don't allow broadcast messages to be fragmented */
1070 		if ((u_short)ip->ip_len > ifp->if_mtu) {
1071 			error = EMSGSIZE;
1072 			goto bad;
1073 		}
1074 		m->m_flags |= M_BCAST;
1075 	} else {
1076 		m->m_flags &= ~M_BCAST;
1077 	}
1078 
1079 sendit:
1080 #if PF
1081 	/* Invoke outbound packet filter */
1082 	if (PF_IS_ENABLED) {
1083 		int rc;
1084 
1085 		m0 = m; /* Save for later */
1086 #if DUMMYNET
1087 		rc = ip_output_pf_dn_hook(ifp, mppn, &m, dn_pf_rule, ro, dst, flags, ipoa);
1088 #else /* DUMMYNET */
1089 		rc = pf_af_hook(ifp, mppn, &m, AF_INET, FALSE, NULL);
1090 #endif /* DUMMYNET */
1091 		if (rc != 0 || m == NULL) {
1092 			/* Move to the next packet */
1093 			m = *mppn;
1094 
1095 			/* Skip ahead if first packet in list got dropped */
1096 			if (packetlist == m0) {
1097 				packetlist = m;
1098 			}
1099 
1100 			if (m != NULL) {
1101 				m0 = m;
1102 				/* Next packet in the chain */
1103 				goto loopit;
1104 			} else if (packetlist != NULL) {
1105 				/* No more packet; send down the chain */
1106 				goto sendchain;
1107 			}
1108 			/* Nothing left; we're done */
1109 			goto done;
1110 		}
1111 		m0 = m;
1112 		ip = mtod(m, struct ip *);
1113 		pkt_dst = ip->ip_dst;
1114 		hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1115 	}
1116 #endif /* PF */
1117 	/*
1118 	 * Force IP TTL to 255 following draft-ietf-zeroconf-ipv4-linklocal.txt
1119 	 */
1120 	if (IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)) ||
1121 	    IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) {
1122 		ip_linklocal_stat.iplls_out_total++;
1123 		if (ip->ip_ttl != MAXTTL) {
1124 			ip_linklocal_stat.iplls_out_badttl++;
1125 			ip->ip_ttl = MAXTTL;
1126 		}
1127 	}
1128 
1129 	if (!ipobf.didfilter &&
1130 	    !TAILQ_EMPTY(&ipv4_filters)
1131 #if NECP
1132 	    && !necp_packet_should_skip_filters(m)
1133 #endif // NECP
1134 	    ) {
1135 		struct ipfilter *filter;
1136 		int seen = (inject_filter_ref == NULL);
1137 		ipf_pktopts.ippo_flags &= ~IPPOF_MCAST_OPTS;
1138 
1139 		/*
1140 		 * Check that a TSO frame isn't passed to a filter.
1141 		 * This could happen if a filter is inserted while
1142 		 * TCP is sending the TSO packet.
1143 		 */
1144 		if (m->m_pkthdr.csum_flags & CSUM_TSO_IPV4) {
1145 			error = EMSGSIZE;
1146 			goto bad;
1147 		}
1148 
1149 		ipf_ref();
1150 
1151 		/* 4135317 - always pass network byte order to filter */
1152 #if BYTE_ORDER != BIG_ENDIAN
1153 		HTONS(ip->ip_len);
1154 		HTONS(ip->ip_off);
1155 #endif
1156 		TAILQ_FOREACH(filter, &ipv4_filters, ipf_link) {
1157 			if (seen == 0) {
1158 				if ((struct ipfilter *)inject_filter_ref ==
1159 				    filter) {
1160 					seen = 1;
1161 				}
1162 			} else if (filter->ipf_filter.ipf_output) {
1163 				errno_t result;
1164 				result = filter->ipf_filter.
1165 				    ipf_output(filter->ipf_filter.cookie,
1166 				    (mbuf_t *)&m, ippo);
1167 				if (result == EJUSTRETURN) {
1168 					ipf_unref();
1169 					goto done;
1170 				}
1171 				if (result != 0) {
1172 					ipf_unref();
1173 					goto bad;
1174 				}
1175 			}
1176 		}
1177 		/* set back to host byte order */
1178 		ip = mtod(m, struct ip *);
1179 #if BYTE_ORDER != BIG_ENDIAN
1180 		NTOHS(ip->ip_len);
1181 		NTOHS(ip->ip_off);
1182 #endif
1183 		ipf_unref();
1184 	}
1185 
1186 #if NECP
1187 	/* Process Network Extension Policy. Will Pass, Drop, or Rebind packet. */
1188 	necp_matched_policy_id = necp_ip_output_find_policy_match(m,
1189 	    flags, (flags & IP_OUTARGS) ? ipoa : NULL, ro ? ro->ro_rt : NULL, &necp_result, &necp_result_parameter);
1190 	if (necp_matched_policy_id) {
1191 		necp_mark_packet_from_ip(m, necp_matched_policy_id);
1192 		switch (necp_result) {
1193 		case NECP_KERNEL_POLICY_RESULT_PASS:
1194 			if (necp_result_parameter.pass_flags & NECP_KERNEL_POLICY_PASS_NO_SKIP_IPSEC) {
1195 				break;
1196 			}
1197 			/* Check if the interface is allowed */
1198 			if (!necp_packet_is_allowed_over_interface(m, ifp)) {
1199 				error = EHOSTUNREACH;
1200 				OSAddAtomic(1, &ipstat.ips_necp_policy_drop);
1201 				goto bad;
1202 			}
1203 			goto skip_ipsec;
1204 		case NECP_KERNEL_POLICY_RESULT_DROP:
1205 		case NECP_KERNEL_POLICY_RESULT_SOCKET_DIVERT:
1206 			/* Flow divert packets should be blocked at the IP layer */
1207 			error = EHOSTUNREACH;
1208 			OSAddAtomic(1, &ipstat.ips_necp_policy_drop);
1209 			goto bad;
1210 		case NECP_KERNEL_POLICY_RESULT_IP_TUNNEL: {
1211 			/* Verify that the packet is being routed to the tunnel */
1212 			struct ifnet *policy_ifp = necp_get_ifnet_from_result_parameter(&necp_result_parameter);
1213 			if (policy_ifp == ifp) {
1214 				/* Check if the interface is allowed */
1215 				if (!necp_packet_is_allowed_over_interface(m, ifp)) {
1216 					error = EHOSTUNREACH;
1217 					OSAddAtomic(1, &ipstat.ips_necp_policy_drop);
1218 					goto bad;
1219 				}
1220 				goto skip_ipsec;
1221 			} else {
1222 				if (necp_packet_can_rebind_to_ifnet(m, policy_ifp, &necp_route, AF_INET)) {
1223 					/* Check if the interface is allowed */
1224 					if (!necp_packet_is_allowed_over_interface(m, policy_ifp)) {
1225 						error = EHOSTUNREACH;
1226 						OSAddAtomic(1, &ipstat.ips_necp_policy_drop);
1227 						goto bad;
1228 					}
1229 
1230 					/*
1231 					 * Update the QOS marking policy if
1232 					 * 1. up layer asks it to do so
1233 					 * 2. net_qos_policy_restricted is not set
1234 					 * 3. qos_marking_gencount doesn't match necp_kernel_socket_policies_gencount (checked in necp_lookup_current_qos_marking)
1235 					 */
1236 					if (ipoa != NULL &&
1237 					    (ipoa->ipoa_flags & IPOAF_REDO_QOSMARKING_POLICY) &&
1238 					    net_qos_policy_restricted != 0) {
1239 						bool qos_marking = (ipoa->ipoa_flags & IPOAF_QOSMARKING_ALLOWED) ? TRUE : FALSE;
1240 						qos_marking = necp_lookup_current_qos_marking(&ipoa->qos_marking_gencount, NULL, policy_ifp, necp_result_parameter.route_rule_id, qos_marking);
1241 						if (qos_marking) {
1242 							ipoa->ipoa_flags |= IPOAF_QOSMARKING_ALLOWED;
1243 						} else {
1244 							ipoa->ipoa_flags &= ~IPOAF_QOSMARKING_ALLOWED;
1245 						}
1246 					}
1247 
1248 					/* Set ifp to the tunnel interface, since it is compatible with the packet */
1249 					ifp = policy_ifp;
1250 					ro = &necp_route;
1251 					goto skip_ipsec;
1252 				} else {
1253 					error = ENETUNREACH;
1254 					OSAddAtomic(1, &ipstat.ips_necp_policy_drop);
1255 					goto bad;
1256 				}
1257 			}
1258 		}
1259 		default:
1260 			break;
1261 		}
1262 	}
1263 	/* Catch-all to check if the interface is allowed */
1264 	if (!necp_packet_is_allowed_over_interface(m, ifp)) {
1265 		error = EHOSTUNREACH;
1266 		OSAddAtomic(1, &ipstat.ips_necp_policy_drop);
1267 		goto bad;
1268 	}
1269 #endif /* NECP */
1270 
1271 #if IPSEC
1272 	if (ipsec_bypass != 0 || (flags & IP_NOIPSEC)) {
1273 		goto skip_ipsec;
1274 	}
1275 
1276 	KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_START, 0, 0, 0, 0, 0);
1277 
1278 	if (sp == NULL) {
1279 		/* get SP for this packet */
1280 		if (so != NULL) {
1281 			sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND,
1282 			    so, &error);
1283 		} else {
1284 			sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND,
1285 			    flags, &error);
1286 		}
1287 		if (sp == NULL) {
1288 			IPSEC_STAT_INCREMENT(ipsecstat.out_inval);
1289 			KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END,
1290 			    0, 0, 0, 0, 0);
1291 			goto bad;
1292 		}
1293 	}
1294 
1295 	error = 0;
1296 
1297 	/* check policy */
1298 	switch (sp->policy) {
1299 	case IPSEC_POLICY_DISCARD:
1300 	case IPSEC_POLICY_GENERATE:
1301 		/*
1302 		 * This packet is just discarded.
1303 		 */
1304 		IPSEC_STAT_INCREMENT(ipsecstat.out_polvio);
1305 		KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END,
1306 		    1, 0, 0, 0, 0);
1307 		goto bad;
1308 
1309 	case IPSEC_POLICY_BYPASS:
1310 	case IPSEC_POLICY_NONE:
1311 		/* no need to do IPsec. */
1312 		KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END,
1313 		    2, 0, 0, 0, 0);
1314 		goto skip_ipsec;
1315 
1316 	case IPSEC_POLICY_IPSEC:
1317 		if (sp->req == NULL) {
1318 			/* acquire a policy */
1319 			error = key_spdacquire(sp);
1320 			KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END,
1321 			    3, 0, 0, 0, 0);
1322 			goto bad;
1323 		}
1324 		if (sp->ipsec_if) {
1325 			/* Verify the redirect to ipsec interface */
1326 			if (sp->ipsec_if == ifp) {
1327 				goto skip_ipsec;
1328 			}
1329 			goto bad;
1330 		}
1331 		break;
1332 
1333 	case IPSEC_POLICY_ENTRUST:
1334 	default:
1335 		printf("ip_output: Invalid policy found. %d\n", sp->policy);
1336 	}
1337 	{
1338 		ipsec_state.m = m;
1339 		if (flags & IP_ROUTETOIF) {
1340 			bzero(&ipsec_state.ro, sizeof(ipsec_state.ro));
1341 		} else {
1342 			route_copyout((struct route *)&ipsec_state.ro, ro, sizeof(struct route));
1343 		}
1344 		ipsec_state.dst = SA(dst);
1345 
1346 		ip->ip_sum = 0;
1347 
1348 		/*
1349 		 * XXX
1350 		 * delayed checksums are not currently compatible with IPsec
1351 		 */
1352 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
1353 			in_delayed_cksum(m);
1354 		}
1355 
1356 #if BYTE_ORDER != BIG_ENDIAN
1357 		HTONS(ip->ip_len);
1358 		HTONS(ip->ip_off);
1359 #endif
1360 
1361 		DTRACE_IP6(send, struct mbuf *, m, struct inpcb *, NULL,
1362 		    struct ip *, ip, struct ifnet *, ifp,
1363 		    struct ip *, ip, struct ip6_hdr *, NULL);
1364 
1365 		error = ipsec4_output(&ipsec_state, sp, flags);
1366 		if (ipsec_state.tunneled == 6) {
1367 			m0 = m = NULL;
1368 			error = 0;
1369 			goto bad;
1370 		}
1371 
1372 		m0 = m = ipsec_state.m;
1373 
1374 #if DUMMYNET
1375 		/*
1376 		 * If we're about to use the route in ipsec_state
1377 		 * and this came from dummynet, cleaup now.
1378 		 */
1379 		if (ro == &saved_route &&
1380 		    (!(flags & IP_ROUTETOIF) || ipsec_state.tunneled)) {
1381 			ROUTE_RELEASE(ro);
1382 		}
1383 #endif /* DUMMYNET */
1384 
1385 		if (flags & IP_ROUTETOIF) {
1386 			/*
1387 			 * if we have tunnel mode SA, we may need to ignore
1388 			 * IP_ROUTETOIF.
1389 			 */
1390 			if (ipsec_state.tunneled) {
1391 				flags &= ~IP_ROUTETOIF;
1392 				ro = (struct route *)&ipsec_state.ro;
1393 			}
1394 		} else {
1395 			ro = (struct route *)&ipsec_state.ro;
1396 		}
1397 		dst = SIN(ipsec_state.dst);
1398 		if (error) {
1399 			/* mbuf is already reclaimed in ipsec4_output. */
1400 			m0 = NULL;
1401 			switch (error) {
1402 			case EHOSTUNREACH:
1403 			case ENETUNREACH:
1404 			case EMSGSIZE:
1405 			case ENOBUFS:
1406 			case ENOMEM:
1407 				break;
1408 			default:
1409 				printf("ip4_output (ipsec): error code %d\n", error);
1410 				OS_FALLTHROUGH;
1411 			case ENOENT:
1412 				/* don't show these error codes to the user */
1413 				error = 0;
1414 				break;
1415 			}
1416 			KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END,
1417 			    4, 0, 0, 0, 0);
1418 			goto bad;
1419 		}
1420 	}
1421 
1422 	/* be sure to update variables that are affected by ipsec4_output() */
1423 	ip = mtod(m, struct ip *);
1424 
1425 #ifdef _IP_VHL
1426 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1427 #else /* !_IP_VHL */
1428 	hlen = ip->ip_hl << 2;
1429 #endif /* !_IP_VHL */
1430 	/* Check that there wasn't a route change and src is still valid */
1431 	if (ROUTE_UNUSABLE(ro)) {
1432 		ROUTE_RELEASE(ro);
1433 		VERIFY(src_ia == NULL);
1434 		if (ip->ip_src.s_addr != INADDR_ANY &&
1435 		    !(flags & (IP_ROUTETOIF | IP_FORWARDING)) &&
1436 		    (src_ia = ifa_foraddr(ip->ip_src.s_addr)) == NULL) {
1437 			error = EADDRNOTAVAIL;
1438 			KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END,
1439 			    5, 0, 0, 0, 0);
1440 			goto bad;
1441 		}
1442 		if (src_ia != NULL) {
1443 			IFA_REMREF(&src_ia->ia_ifa);
1444 			src_ia = NULL;
1445 		}
1446 	}
1447 
1448 	if (ro->ro_rt == NULL) {
1449 		if (!(flags & IP_ROUTETOIF)) {
1450 			printf("%s: can't update route after "
1451 			    "IPsec processing\n", __func__);
1452 			error = EHOSTUNREACH;   /* XXX */
1453 			KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END,
1454 			    6, 0, 0, 0, 0);
1455 			goto bad;
1456 		}
1457 	} else {
1458 		if (ia != NULL) {
1459 			IFA_REMREF(&ia->ia_ifa);
1460 		}
1461 		RT_LOCK_SPIN(ro->ro_rt);
1462 		ia = ifatoia(ro->ro_rt->rt_ifa);
1463 		if (ia != NULL) {
1464 			/* Become a regular mutex */
1465 			RT_CONVERT_LOCK(ro->ro_rt);
1466 			IFA_ADDREF(&ia->ia_ifa);
1467 		}
1468 		ifp = ro->ro_rt->rt_ifp;
1469 		RT_UNLOCK(ro->ro_rt);
1470 	}
1471 
1472 	/* make it flipped, again. */
1473 #if BYTE_ORDER != BIG_ENDIAN
1474 	NTOHS(ip->ip_len);
1475 	NTOHS(ip->ip_off);
1476 #endif
1477 	KERNEL_DEBUG(DBG_FNC_IPSEC4_OUTPUT | DBG_FUNC_END,
1478 	    7, 0xff, 0xff, 0xff, 0xff);
1479 
1480 	/* Pass to filters again */
1481 	if (!TAILQ_EMPTY(&ipv4_filters)
1482 #if NECP
1483 	    && !necp_packet_should_skip_filters(m)
1484 #endif // NECP
1485 	    ) {
1486 		struct ipfilter *filter;
1487 
1488 		ipf_pktopts.ippo_flags &= ~IPPOF_MCAST_OPTS;
1489 
1490 		/*
1491 		 * Check that a TSO frame isn't passed to a filter.
1492 		 * This could happen if a filter is inserted while
1493 		 * TCP is sending the TSO packet.
1494 		 */
1495 		if (m->m_pkthdr.csum_flags & CSUM_TSO_IPV4) {
1496 			error = EMSGSIZE;
1497 			goto bad;
1498 		}
1499 
1500 		ipf_ref();
1501 
1502 		/* 4135317 - always pass network byte order to filter */
1503 #if BYTE_ORDER != BIG_ENDIAN
1504 		HTONS(ip->ip_len);
1505 		HTONS(ip->ip_off);
1506 #endif
1507 		TAILQ_FOREACH(filter, &ipv4_filters, ipf_link) {
1508 			if (filter->ipf_filter.ipf_output) {
1509 				errno_t result;
1510 				result = filter->ipf_filter.
1511 				    ipf_output(filter->ipf_filter.cookie,
1512 				    (mbuf_t *)&m, ippo);
1513 				if (result == EJUSTRETURN) {
1514 					ipf_unref();
1515 					goto done;
1516 				}
1517 				if (result != 0) {
1518 					ipf_unref();
1519 					goto bad;
1520 				}
1521 			}
1522 		}
1523 		/* set back to host byte order */
1524 		ip = mtod(m, struct ip *);
1525 #if BYTE_ORDER != BIG_ENDIAN
1526 		NTOHS(ip->ip_len);
1527 		NTOHS(ip->ip_off);
1528 #endif
1529 		ipf_unref();
1530 	}
1531 skip_ipsec:
1532 #endif /* IPSEC */
1533 
1534 
1535 	/* 127/8 must not appear on wire - RFC1122 */
1536 	if (!(ifp->if_flags & IFF_LOOPBACK) &&
1537 	    ((ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
1538 	    (ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)) {
1539 		OSAddAtomic(1, &ipstat.ips_badaddr);
1540 		error = EADDRNOTAVAIL;
1541 		goto bad;
1542 	}
1543 
1544 	if (ipoa != NULL) {
1545 		u_int8_t dscp = ip->ip_tos >> IPTOS_DSCP_SHIFT;
1546 
1547 		error = set_packet_qos(m, ifp,
1548 		    ipoa->ipoa_flags & IPOAF_QOSMARKING_ALLOWED ? TRUE : FALSE,
1549 		    ipoa->ipoa_sotc, ipoa->ipoa_netsvctype, &dscp);
1550 		if (error == 0) {
1551 			ip->ip_tos &= IPTOS_ECN_MASK;
1552 			ip->ip_tos |= dscp << IPTOS_DSCP_SHIFT;
1553 		} else {
1554 			printf("%s if_dscp_for_mbuf() error %d\n", __func__, error);
1555 			error = 0;
1556 		}
1557 	}
1558 
1559 	ip_output_checksum(ifp, m, (IP_VHL_HL(ip->ip_vhl) << 2),
1560 	    ip->ip_len, &sw_csum);
1561 
1562 	interface_mtu = ifp->if_mtu;
1563 
1564 	if (INTF_ADJUST_MTU_FOR_CLAT46(ifp)) {
1565 		interface_mtu = IN6_LINKMTU(ifp);
1566 		/* Further adjust the size for CLAT46 expansion */
1567 		interface_mtu -= CLAT46_HDR_EXPANSION_OVERHD;
1568 	}
1569 
1570 	/*
1571 	 * If small enough for interface, or the interface will take
1572 	 * care of the fragmentation for us, can just send directly.
1573 	 */
1574 	if ((u_short)ip->ip_len <= interface_mtu || TSO_IPV4_OK(ifp, m) ||
1575 	    (!(ip->ip_off & IP_DF) && (ifp->if_hwassist & CSUM_FRAGMENT))) {
1576 #if BYTE_ORDER != BIG_ENDIAN
1577 		HTONS(ip->ip_len);
1578 		HTONS(ip->ip_off);
1579 #endif
1580 
1581 		ip->ip_sum = 0;
1582 		if (sw_csum & CSUM_DELAY_IP) {
1583 			ip->ip_sum = ip_cksum_hdr_out(m, hlen);
1584 			sw_csum &= ~CSUM_DELAY_IP;
1585 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_IP;
1586 		}
1587 
1588 #if IPSEC
1589 		/* clean ipsec history once it goes out of the node */
1590 		if (ipsec_bypass == 0 && !(flags & IP_NOIPSEC)) {
1591 			ipsec_delaux(m);
1592 		}
1593 #endif /* IPSEC */
1594 		if ((m->m_pkthdr.csum_flags & CSUM_TSO_IPV4) &&
1595 		    (m->m_pkthdr.tso_segsz > 0)) {
1596 			scnt += m->m_pkthdr.len / m->m_pkthdr.tso_segsz;
1597 		} else {
1598 			scnt++;
1599 		}
1600 
1601 		if (packetchain == 0) {
1602 			if (ro->ro_rt != NULL && nstat_collect) {
1603 				nstat_route_tx(ro->ro_rt, scnt,
1604 				    m->m_pkthdr.len, 0);
1605 			}
1606 
1607 			error = dlil_output(ifp, PF_INET, m, ro->ro_rt,
1608 			    SA(dst), 0, adv);
1609 			if (dlil_verbose && error) {
1610 				printf("dlil_output error on interface %s: %d\n",
1611 				    ifp->if_xname, error);
1612 			}
1613 			scnt = 0;
1614 			goto done;
1615 		} else {
1616 			/*
1617 			 * packet chaining allows us to reuse the
1618 			 * route for all packets
1619 			 */
1620 			bytecnt += m->m_pkthdr.len;
1621 			mppn = &m->m_nextpkt;
1622 			m = m->m_nextpkt;
1623 			if (m == NULL) {
1624 #if PF
1625 sendchain:
1626 #endif /* PF */
1627 				if (pktcnt > ip_maxchainsent) {
1628 					ip_maxchainsent = pktcnt;
1629 				}
1630 				if (ro->ro_rt != NULL && nstat_collect) {
1631 					nstat_route_tx(ro->ro_rt, scnt,
1632 					    bytecnt, 0);
1633 				}
1634 
1635 				error = dlil_output(ifp, PF_INET, packetlist,
1636 				    ro->ro_rt, SA(dst), 0, adv);
1637 				if (dlil_verbose && error) {
1638 					printf("dlil_output error on interface %s: %d\n",
1639 					    ifp->if_xname, error);
1640 				}
1641 				pktcnt = 0;
1642 				scnt = 0;
1643 				bytecnt = 0;
1644 				goto done;
1645 			}
1646 			m0 = m;
1647 			pktcnt++;
1648 			goto loopit;
1649 		}
1650 	}
1651 
1652 	VERIFY(interface_mtu != 0);
1653 	/*
1654 	 * Too large for interface; fragment if possible.
1655 	 * Must be able to put at least 8 bytes per fragment.
1656 	 * Balk when DF bit is set or the interface didn't support TSO.
1657 	 */
1658 	if ((ip->ip_off & IP_DF) || pktcnt > 0 ||
1659 	    (m->m_pkthdr.csum_flags & CSUM_TSO_IPV4)) {
1660 		error = EMSGSIZE;
1661 		/*
1662 		 * This case can happen if the user changed the MTU
1663 		 * of an interface after enabling IP on it.  Because
1664 		 * most netifs don't keep track of routes pointing to
1665 		 * them, there is no way for one to update all its
1666 		 * routes when the MTU is changed.
1667 		 */
1668 		if (ro->ro_rt) {
1669 			RT_LOCK_SPIN(ro->ro_rt);
1670 			if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1671 			    !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1672 			    (ro->ro_rt->rt_rmx.rmx_mtu > interface_mtu)) {
1673 				ro->ro_rt->rt_rmx.rmx_mtu = interface_mtu;
1674 			}
1675 			RT_UNLOCK(ro->ro_rt);
1676 		}
1677 		if (pktcnt > 0) {
1678 			m0 = packetlist;
1679 		}
1680 		OSAddAtomic(1, &ipstat.ips_cantfrag);
1681 		goto bad;
1682 	}
1683 
1684 	/*
1685 	 * XXX Only TCP seems to be passing a list of packets here.
1686 	 * The following issue is limited to UDP datagrams with 0 checksum.
1687 	 * For now limit it to the case when single packet is passed down.
1688 	 */
1689 	if (packetchain == 0 && IS_INTF_CLAT46(ifp)) {
1690 		/*
1691 		 * If it is a UDP packet that has checksum set to 0
1692 		 * and is also not being offloaded, compute a full checksum
1693 		 * and update the UDP checksum.
1694 		 */
1695 		if (ip->ip_p == IPPROTO_UDP &&
1696 		    !(m->m_pkthdr.csum_flags & (CSUM_UDP | CSUM_PARTIAL))) {
1697 			struct udphdr *uh = NULL;
1698 
1699 			if (m->m_len < hlen + sizeof(struct udphdr)) {
1700 				m = m_pullup(m, hlen + sizeof(struct udphdr));
1701 				if (m == NULL) {
1702 					error = ENOBUFS;
1703 					m0 = m;
1704 					goto bad;
1705 				}
1706 				m0 = m;
1707 				ip = mtod(m, struct ip *);
1708 			}
1709 			/*
1710 			 * Get UDP header and if checksum is 0, then compute the full
1711 			 * checksum.
1712 			 */
1713 			uh = (struct udphdr *)(void *)((caddr_t)ip + hlen);
1714 			if (uh->uh_sum == 0) {
1715 				uh->uh_sum = inet_cksum(m, IPPROTO_UDP, hlen,
1716 				    ip->ip_len - hlen);
1717 				if (uh->uh_sum == 0) {
1718 					uh->uh_sum = 0xffff;
1719 				}
1720 			}
1721 		}
1722 	}
1723 
1724 	error = ip_fragment(m, ifp, interface_mtu, sw_csum);
1725 	if (error != 0) {
1726 		m0 = m = NULL;
1727 		goto bad;
1728 	}
1729 
1730 	KERNEL_DEBUG(DBG_LAYER_END, ip->ip_dst.s_addr,
1731 	    ip->ip_src.s_addr, ip->ip_p, ip->ip_off, ip->ip_len);
1732 
1733 	for (m = m0; m; m = m0) {
1734 		m0 = m->m_nextpkt;
1735 		m->m_nextpkt = 0;
1736 #if IPSEC
1737 		/* clean ipsec history once it goes out of the node */
1738 		if (ipsec_bypass == 0 && !(flags & IP_NOIPSEC)) {
1739 			ipsec_delaux(m);
1740 		}
1741 #endif /* IPSEC */
1742 		if (error == 0) {
1743 			if ((packetchain != 0) && (pktcnt > 0)) {
1744 				panic("%s: mix of packet in packetlist is "
1745 				    "wrong=%p", __func__, packetlist);
1746 				/* NOTREACHED */
1747 			}
1748 			if (ro->ro_rt != NULL && nstat_collect) {
1749 				nstat_route_tx(ro->ro_rt, 1,
1750 				    m->m_pkthdr.len, 0);
1751 			}
1752 			error = dlil_output(ifp, PF_INET, m, ro->ro_rt,
1753 			    SA(dst), 0, adv);
1754 			if (dlil_verbose && error) {
1755 				printf("dlil_output error on interface %s: %d\n",
1756 				    ifp->if_xname, error);
1757 			}
1758 		} else {
1759 			m_freem(m);
1760 		}
1761 	}
1762 
1763 	if (error == 0) {
1764 		OSAddAtomic(1, &ipstat.ips_fragmented);
1765 	}
1766 
1767 done:
1768 	if (ia != NULL) {
1769 		IFA_REMREF(&ia->ia_ifa);
1770 		ia = NULL;
1771 	}
1772 #if IPSEC
1773 	ROUTE_RELEASE(&ipsec_state.ro);
1774 	if (sp != NULL) {
1775 		KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1776 		    printf("DP ip_output call free SP:%x\n", sp));
1777 		key_freesp(sp, KEY_SADB_UNLOCKED);
1778 	}
1779 #endif /* IPSEC */
1780 #if NECP
1781 	ROUTE_RELEASE(&necp_route);
1782 #endif /* NECP */
1783 #if DUMMYNET
1784 	ROUTE_RELEASE(&saved_route);
1785 #endif /* DUMMYNET */
1786 
1787 	KERNEL_DEBUG(DBG_FNC_IP_OUTPUT | DBG_FUNC_END, error, 0, 0, 0, 0);
1788 	if (ip_output_measure) {
1789 		net_perf_measure_time(&net_perf, &start_tv, packets_processed);
1790 		net_perf_histogram(&net_perf, packets_processed);
1791 	}
1792 	return error;
1793 bad:
1794 	if (pktcnt > 0) {
1795 		m0 = packetlist;
1796 	}
1797 	m_freem_list(m0);
1798 	goto done;
1799 
1800 #undef ipsec_state
1801 #undef args
1802 #undef sro_fwd
1803 #undef saved_route
1804 #undef ipf_pktopts
1805 #undef IP_CHECK_RESTRICTIONS
1806 }
1807 
1808 int
ip_fragment(struct mbuf * m,struct ifnet * ifp,uint32_t mtu,int sw_csum)1809 ip_fragment(struct mbuf *m, struct ifnet *ifp, uint32_t mtu, int sw_csum)
1810 {
1811 	struct ip *ip, *mhip;
1812 	int len, hlen, mhlen, firstlen, off, error = 0;
1813 	struct mbuf **mnext = &m->m_nextpkt, *m0;
1814 	int nfrags = 1;
1815 
1816 	ip = mtod(m, struct ip *);
1817 #ifdef _IP_VHL
1818 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1819 #else /* !_IP_VHL */
1820 	hlen = ip->ip_hl << 2;
1821 #endif /* !_IP_VHL */
1822 
1823 	/*
1824 	 * We need to adjust the fragment sizes to account
1825 	 * for IPv6 fragment header if it needs to be translated
1826 	 * from IPv4 to IPv6.
1827 	 */
1828 	if (IS_INTF_CLAT46(ifp)) {
1829 		mtu -= sizeof(struct ip6_frag);
1830 	}
1831 
1832 	firstlen = len = (mtu - hlen) & ~7;
1833 	if (len < 8) {
1834 		m_freem(m);
1835 		return EMSGSIZE;
1836 	}
1837 
1838 	/*
1839 	 * if the interface will not calculate checksums on
1840 	 * fragmented packets, then do it here.
1841 	 */
1842 	if ((m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1843 	    !(ifp->if_hwassist & CSUM_IP_FRAGS)) {
1844 		in_delayed_cksum(m);
1845 	}
1846 
1847 	/*
1848 	 * Loop through length of segment after first fragment,
1849 	 * make new header and copy data of each part and link onto chain.
1850 	 */
1851 	m0 = m;
1852 	mhlen = sizeof(struct ip);
1853 	for (off = hlen + len; off < (u_short)ip->ip_len; off += len) {
1854 		MGETHDR(m, M_DONTWAIT, MT_HEADER);      /* MAC-OK */
1855 		if (m == NULL) {
1856 			error = ENOBUFS;
1857 			OSAddAtomic(1, &ipstat.ips_odropped);
1858 			goto sendorfree;
1859 		}
1860 		m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1861 		m->m_data += max_linkhdr;
1862 		mhip = mtod(m, struct ip *);
1863 		*mhip = *ip;
1864 		if (hlen > sizeof(struct ip)) {
1865 			mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1866 			mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1867 		}
1868 		m->m_len = mhlen;
1869 		mhip->ip_off = (u_short)(((off - hlen) >> 3) + (ip->ip_off & ~IP_MF));
1870 		if (ip->ip_off & IP_MF) {
1871 			mhip->ip_off |= IP_MF;
1872 		}
1873 		if (off + len >= (u_short)ip->ip_len) {
1874 			len = (u_short)ip->ip_len - off;
1875 		} else {
1876 			mhip->ip_off |= IP_MF;
1877 		}
1878 		mhip->ip_len = htons((u_short)(len + mhlen));
1879 		m->m_next = m_copy(m0, off, len);
1880 		if (m->m_next == NULL) {
1881 			(void) m_free(m);
1882 			error = ENOBUFS;        /* ??? */
1883 			OSAddAtomic(1, &ipstat.ips_odropped);
1884 			goto sendorfree;
1885 		}
1886 		m->m_pkthdr.len = mhlen + len;
1887 		m->m_pkthdr.rcvif = NULL;
1888 		m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1889 
1890 		M_COPY_CLASSIFIER(m, m0);
1891 		M_COPY_PFTAG(m, m0);
1892 		M_COPY_NECPTAG(m, m0);
1893 
1894 #if BYTE_ORDER != BIG_ENDIAN
1895 		HTONS(mhip->ip_off);
1896 #endif
1897 
1898 		mhip->ip_sum = 0;
1899 		if (sw_csum & CSUM_DELAY_IP) {
1900 			mhip->ip_sum = ip_cksum_hdr_out(m, mhlen);
1901 			m->m_pkthdr.csum_flags &= ~CSUM_DELAY_IP;
1902 		}
1903 		*mnext = m;
1904 		mnext = &m->m_nextpkt;
1905 		nfrags++;
1906 	}
1907 	OSAddAtomic(nfrags, &ipstat.ips_ofragments);
1908 
1909 	/* set first/last markers for fragment chain */
1910 	m->m_flags |= M_LASTFRAG;
1911 	m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1912 	m0->m_pkthdr.csum_data = nfrags;
1913 
1914 	/*
1915 	 * Update first fragment by trimming what's been copied out
1916 	 * and updating header, then send each fragment (in order).
1917 	 */
1918 	m = m0;
1919 	m_adj(m, hlen + firstlen - (u_short)ip->ip_len);
1920 	m->m_pkthdr.len = hlen + firstlen;
1921 	ip->ip_len = htons((u_short)m->m_pkthdr.len);
1922 	ip->ip_off |= IP_MF;
1923 
1924 #if BYTE_ORDER != BIG_ENDIAN
1925 	HTONS(ip->ip_off);
1926 #endif
1927 
1928 	ip->ip_sum = 0;
1929 	if (sw_csum & CSUM_DELAY_IP) {
1930 		ip->ip_sum = ip_cksum_hdr_out(m, hlen);
1931 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_IP;
1932 	}
1933 sendorfree:
1934 	if (error) {
1935 		m_freem_list(m0);
1936 	}
1937 
1938 	return error;
1939 }
1940 
1941 static void
ip_out_cksum_stats(int proto,u_int32_t len)1942 ip_out_cksum_stats(int proto, u_int32_t len)
1943 {
1944 	switch (proto) {
1945 	case IPPROTO_TCP:
1946 		tcp_out_cksum_stats(len);
1947 		break;
1948 	case IPPROTO_UDP:
1949 		udp_out_cksum_stats(len);
1950 		break;
1951 	default:
1952 		/* keep only TCP or UDP stats for now */
1953 		break;
1954 	}
1955 }
1956 
1957 /*
1958  * Process a delayed payload checksum calculation (outbound path.)
1959  *
1960  * hoff is the number of bytes beyond the mbuf data pointer which
1961  * points to the IP header.
1962  *
1963  * Returns a bitmask representing all the work done in software.
1964  */
1965 uint32_t
in_finalize_cksum(struct mbuf * m,uint32_t hoff,uint32_t csum_flags)1966 in_finalize_cksum(struct mbuf *m, uint32_t hoff, uint32_t csum_flags)
1967 {
1968 	unsigned char buf[15 << 2] __attribute__((aligned(8)));
1969 	struct ip *ip;
1970 	uint32_t offset, _hlen, mlen, hlen, len, sw_csum;
1971 	uint16_t csum, ip_len;
1972 
1973 	_CASSERT(sizeof(csum) == sizeof(uint16_t));
1974 	VERIFY(m->m_flags & M_PKTHDR);
1975 
1976 	sw_csum = (csum_flags & m->m_pkthdr.csum_flags);
1977 
1978 	if ((sw_csum &= (CSUM_DELAY_IP | CSUM_DELAY_DATA)) == 0) {
1979 		goto done;
1980 	}
1981 
1982 	mlen = m->m_pkthdr.len;                         /* total mbuf len */
1983 
1984 	/* sanity check (need at least simple IP header) */
1985 	if (mlen < (hoff + sizeof(*ip))) {
1986 		panic("%s: mbuf %p pkt len (%u) < hoff+ip_hdr "
1987 		    "(%u+%u)\n", __func__, m, mlen, hoff,
1988 		    (uint32_t)sizeof(*ip));
1989 		/* NOTREACHED */
1990 	}
1991 
1992 	/*
1993 	 * In case the IP header is not contiguous, or not 32-bit aligned,
1994 	 * or if we're computing the IP header checksum, copy it to a local
1995 	 * buffer.  Copy only the simple IP header here (IP options case
1996 	 * is handled below.)
1997 	 */
1998 	if ((sw_csum & CSUM_DELAY_IP) || (hoff + sizeof(*ip)) > m->m_len ||
1999 	    !IP_HDR_ALIGNED_P(mtod(m, caddr_t) + hoff)) {
2000 		m_copydata(m, hoff, sizeof(*ip), (caddr_t)buf);
2001 		ip = (struct ip *)(void *)buf;
2002 		_hlen = sizeof(*ip);
2003 	} else {
2004 		ip = (struct ip *)(void *)(m->m_data + hoff);
2005 		_hlen = 0;
2006 	}
2007 
2008 	hlen = IP_VHL_HL(ip->ip_vhl) << 2;              /* IP header len */
2009 
2010 	/* sanity check */
2011 	if (mlen < (hoff + hlen)) {
2012 		panic("%s: mbuf %p pkt too short (%d) for IP header (%u), "
2013 		    "hoff %u", __func__, m, mlen, hlen, hoff);
2014 		/* NOTREACHED */
2015 	}
2016 
2017 	/*
2018 	 * We could be in the context of an IP or interface filter; in the
2019 	 * former case, ip_len would be in host (correct) order while for
2020 	 * the latter it would be in network order.  Because of this, we
2021 	 * attempt to interpret the length field by comparing it against
2022 	 * the actual packet length.  If the comparison fails, byte swap
2023 	 * the length and check again.  If it still fails, use the actual
2024 	 * packet length.  This also covers the trailing bytes case.
2025 	 */
2026 	ip_len = ip->ip_len;
2027 	if (ip_len != (mlen - hoff)) {
2028 		ip_len = OSSwapInt16(ip_len);
2029 		if (ip_len != (mlen - hoff)) {
2030 			printf("%s: mbuf 0x%llx proto %d IP len %d (%x) "
2031 			    "[swapped %d (%x)] doesn't match actual packet "
2032 			    "length; %d is used instead\n", __func__,
2033 			    (uint64_t)VM_KERNEL_ADDRPERM(m), ip->ip_p,
2034 			    ip->ip_len, ip->ip_len, ip_len, ip_len,
2035 			    (mlen - hoff));
2036 			if (mlen - hoff > UINT16_MAX) {
2037 				panic("%s: mlen %u - hoff %u > 65535",
2038 				    __func__, mlen, hoff);
2039 			}
2040 			ip_len = (uint16_t)(mlen - hoff);
2041 		}
2042 	}
2043 
2044 	len = ip_len - hlen;                            /* csum span */
2045 
2046 	if (sw_csum & CSUM_DELAY_DATA) {
2047 		uint16_t ulpoff;
2048 
2049 		/*
2050 		 * offset is added to the lower 16-bit value of csum_data,
2051 		 * which is expected to contain the ULP offset; therefore
2052 		 * CSUM_PARTIAL offset adjustment must be undone.
2053 		 */
2054 		if ((m->m_pkthdr.csum_flags & (CSUM_PARTIAL | CSUM_DATA_VALID)) ==
2055 		    (CSUM_PARTIAL | CSUM_DATA_VALID)) {
2056 			/*
2057 			 * Get back the original ULP offset (this will
2058 			 * undo the CSUM_PARTIAL logic in ip_output.)
2059 			 */
2060 			m->m_pkthdr.csum_data = (m->m_pkthdr.csum_tx_stuff -
2061 			    m->m_pkthdr.csum_tx_start);
2062 		}
2063 
2064 		ulpoff = (m->m_pkthdr.csum_data & 0xffff); /* ULP csum offset */
2065 		offset = hoff + hlen;                   /* ULP header */
2066 
2067 		if (mlen < (ulpoff + sizeof(csum))) {
2068 			panic("%s: mbuf %p pkt len (%u) proto %d invalid ULP "
2069 			    "cksum offset (%u) cksum flags 0x%x\n", __func__,
2070 			    m, mlen, ip->ip_p, ulpoff, m->m_pkthdr.csum_flags);
2071 			/* NOTREACHED */
2072 		}
2073 
2074 		csum = inet_cksum(m, 0, offset, len);
2075 
2076 		/* Update stats */
2077 		ip_out_cksum_stats(ip->ip_p, len);
2078 
2079 		/* RFC1122 4.1.3.4 */
2080 		if (csum == 0 &&
2081 		    (m->m_pkthdr.csum_flags & (CSUM_UDP | CSUM_ZERO_INVERT))) {
2082 			csum = 0xffff;
2083 		}
2084 
2085 		/* Insert the checksum in the ULP csum field */
2086 		offset += ulpoff;
2087 		if (offset + sizeof(csum) > m->m_len) {
2088 			m_copyback(m, offset, sizeof(csum), &csum);
2089 		} else if (IP_HDR_ALIGNED_P(mtod(m, char *) + hoff)) {
2090 			*(uint16_t *)(void *)(mtod(m, char *) + offset) = csum;
2091 		} else {
2092 			bcopy(&csum, (mtod(m, char *) + offset), sizeof(csum));
2093 		}
2094 		m->m_pkthdr.csum_flags &= ~(CSUM_DELAY_DATA | CSUM_DATA_VALID |
2095 		    CSUM_PARTIAL | CSUM_ZERO_INVERT);
2096 	}
2097 
2098 	if (sw_csum & CSUM_DELAY_IP) {
2099 		/* IP header must be in the local buffer */
2100 		VERIFY(_hlen == sizeof(*ip));
2101 		if (_hlen != hlen) {
2102 			VERIFY(hlen <= sizeof(buf));
2103 			m_copydata(m, hoff, hlen, (caddr_t)buf);
2104 			ip = (struct ip *)(void *)buf;
2105 			_hlen = hlen;
2106 		}
2107 
2108 		/*
2109 		 * Compute the IP header checksum as if the IP length
2110 		 * is the length which we believe is "correct"; see
2111 		 * how ip_len gets calculated above.  Note that this
2112 		 * is done on the local copy and not on the real one.
2113 		 */
2114 		ip->ip_len = htons(ip_len);
2115 		ip->ip_sum = 0;
2116 		csum = in_cksum_hdr_opt(ip);
2117 
2118 		/* Update stats */
2119 		ipstat.ips_snd_swcsum++;
2120 		ipstat.ips_snd_swcsum_bytes += hlen;
2121 
2122 		/*
2123 		 * Insert only the checksum in the existing IP header
2124 		 * csum field; all other fields are left unchanged.
2125 		 */
2126 		offset = hoff + offsetof(struct ip, ip_sum);
2127 		if (offset + sizeof(csum) > m->m_len) {
2128 			m_copyback(m, offset, sizeof(csum), &csum);
2129 		} else if (IP_HDR_ALIGNED_P(mtod(m, char *) + hoff)) {
2130 			*(uint16_t *)(void *)(mtod(m, char *) + offset) = csum;
2131 		} else {
2132 			bcopy(&csum, (mtod(m, char *) + offset), sizeof(csum));
2133 		}
2134 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_IP;
2135 	}
2136 
2137 done:
2138 	return sw_csum;
2139 }
2140 
2141 /*
2142  * Insert IP options into preformed packet.
2143  * Adjust IP destination as required for IP source routing,
2144  * as indicated by a non-zero in_addr at the start of the options.
2145  *
2146  * XXX This routine assumes that the packet has no options in place.
2147  */
2148 static struct mbuf *
ip_insertoptions(struct mbuf * m,struct mbuf * opt,int * phlen)2149 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
2150 {
2151 	struct ipoption *p = mtod(opt, struct ipoption *);
2152 	struct mbuf *n;
2153 	struct ip *ip = mtod(m, struct ip *);
2154 	unsigned optlen;
2155 
2156 	optlen = opt->m_len - sizeof(p->ipopt_dst);
2157 	if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
2158 		return m;             /* XXX should fail */
2159 	}
2160 	if (p->ipopt_dst.s_addr) {
2161 		ip->ip_dst = p->ipopt_dst;
2162 	}
2163 	if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
2164 		MGETHDR(n, M_DONTWAIT, MT_HEADER);      /* MAC-OK */
2165 		if (n == NULL) {
2166 			return m;
2167 		}
2168 		n->m_pkthdr.rcvif = 0;
2169 		n->m_pkthdr.len = m->m_pkthdr.len + optlen;
2170 		m->m_len -= sizeof(struct ip);
2171 		m->m_data += sizeof(struct ip);
2172 		n->m_next = m;
2173 		m = n;
2174 		m->m_len = optlen + sizeof(struct ip);
2175 		m->m_data += max_linkhdr;
2176 		(void) memcpy(mtod(m, void *), ip, sizeof(struct ip));
2177 	} else {
2178 		m->m_data -= optlen;
2179 		m->m_len += optlen;
2180 		m->m_pkthdr.len += optlen;
2181 		ovbcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
2182 	}
2183 	ip = mtod(m, struct ip *);
2184 	bcopy(p->ipopt_list, ip + 1, optlen);
2185 	*phlen = sizeof(struct ip) + optlen;
2186 	ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
2187 	ip->ip_len += optlen;
2188 	return m;
2189 }
2190 
2191 /*
2192  * Copy options from ip to jp,
2193  * omitting those not copied during fragmentation.
2194  */
2195 static int
ip_optcopy(struct ip * ip,struct ip * jp)2196 ip_optcopy(struct ip *ip, struct ip *jp)
2197 {
2198 	u_char *cp, *dp;
2199 	int opt, optlen, cnt;
2200 
2201 	cp = (u_char *)(ip + 1);
2202 	dp = (u_char *)(jp + 1);
2203 	cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
2204 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
2205 		opt = cp[0];
2206 		if (opt == IPOPT_EOL) {
2207 			break;
2208 		}
2209 		if (opt == IPOPT_NOP) {
2210 			/* Preserve for IP mcast tunnel's LSRR alignment. */
2211 			*dp++ = IPOPT_NOP;
2212 			optlen = 1;
2213 			continue;
2214 		}
2215 #if DIAGNOSTIC
2216 		if (cnt < IPOPT_OLEN + sizeof(*cp)) {
2217 			panic("malformed IPv4 option passed to ip_optcopy");
2218 			/* NOTREACHED */
2219 		}
2220 #endif
2221 		optlen = cp[IPOPT_OLEN];
2222 #if DIAGNOSTIC
2223 		if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
2224 			panic("malformed IPv4 option passed to ip_optcopy");
2225 			/* NOTREACHED */
2226 		}
2227 #endif
2228 		/* bogus lengths should have been caught by ip_dooptions */
2229 		if (optlen > cnt) {
2230 			optlen = cnt;
2231 		}
2232 		if (IPOPT_COPIED(opt)) {
2233 			bcopy(cp, dp, optlen);
2234 			dp += optlen;
2235 		}
2236 	}
2237 	for (optlen = (int)(dp - (u_char *)(jp + 1)); optlen & 0x3; optlen++) {
2238 		*dp++ = IPOPT_EOL;
2239 	}
2240 	return optlen;
2241 }
2242 
2243 /*
2244  * IP socket option processing.
2245  */
2246 int
ip_ctloutput(struct socket * so,struct sockopt * sopt)2247 ip_ctloutput(struct socket *so, struct sockopt *sopt)
2248 {
2249 	struct  inpcb *inp = sotoinpcb(so);
2250 	int     error, optval;
2251 	lck_mtx_t *mutex_held = NULL;
2252 
2253 	error = optval = 0;
2254 	if (sopt->sopt_level != IPPROTO_IP) {
2255 		return EINVAL;
2256 	}
2257 
2258 	switch (sopt->sopt_dir) {
2259 	case SOPT_SET:
2260 		mutex_held = socket_getlock(so, PR_F_WILLUNLOCK);
2261 		/*
2262 		 *  Wait if we are in the middle of ip_output
2263 		 *  as we unlocked the socket there and don't
2264 		 *  want to overwrite the IP options
2265 		 */
2266 		if (inp->inp_sndinprog_cnt > 0) {
2267 			inp->inp_sndingprog_waiters++;
2268 
2269 			while (inp->inp_sndinprog_cnt > 0) {
2270 				msleep(&inp->inp_sndinprog_cnt, mutex_held,
2271 				    PSOCK | PCATCH, "inp_sndinprog_cnt", NULL);
2272 			}
2273 			inp->inp_sndingprog_waiters--;
2274 		}
2275 		switch (sopt->sopt_name) {
2276 #ifdef notyet
2277 		case IP_RETOPTS:
2278 #endif
2279 		case IP_OPTIONS: {
2280 			struct mbuf *m;
2281 
2282 			if (sopt->sopt_valsize > MLEN) {
2283 				error = EMSGSIZE;
2284 				break;
2285 			}
2286 			MGET(m, sopt->sopt_p != kernproc ? M_WAIT : M_DONTWAIT,
2287 			    MT_HEADER);
2288 			if (m == NULL) {
2289 				error = ENOBUFS;
2290 				break;
2291 			}
2292 			m->m_len = (int32_t)sopt->sopt_valsize;
2293 			error = sooptcopyin(sopt, mtod(m, char *),
2294 			    m->m_len, m->m_len);
2295 			if (error) {
2296 				m_freem(m);
2297 				break;
2298 			}
2299 
2300 			return ip_pcbopts(sopt->sopt_name,
2301 			           &inp->inp_options, m);
2302 		}
2303 
2304 		case IP_TOS:
2305 		case IP_TTL:
2306 		case IP_RECVOPTS:
2307 		case IP_RECVRETOPTS:
2308 		case IP_RECVDSTADDR:
2309 		case IP_RECVIF:
2310 		case IP_RECVTTL:
2311 		case IP_RECVPKTINFO:
2312 		case IP_RECVTOS:
2313 		case IP_DONTFRAG:
2314 			error = sooptcopyin(sopt, &optval, sizeof(optval),
2315 			    sizeof(optval));
2316 			if (error) {
2317 				break;
2318 			}
2319 
2320 			switch (sopt->sopt_name) {
2321 			case IP_TOS:
2322 				if (optval > UINT8_MAX) {
2323 					error = EINVAL;
2324 					break;
2325 				}
2326 				inp->inp_ip_tos = (uint8_t)optval;
2327 				break;
2328 
2329 			case IP_TTL:
2330 				if (optval > UINT8_MAX) {
2331 					error = EINVAL;
2332 					break;
2333 				}
2334 				inp->inp_ip_ttl = (uint8_t)optval;
2335 				break;
2336 #define OPTSET(bit) do {                                                \
2337 	if (optval) {                                                   \
2338 	    inp->inp_flags |= bit;                                      \
2339 	} else {                                                        \
2340 	    inp->inp_flags &= ~bit;                                     \
2341 	}                                                               \
2342 } while (0)
2343 
2344 #define OPTSET2(bit) do {                                               \
2345 	if (optval) {                                                   \
2346 	    inp->inp_flags2 |= bit;                                     \
2347 	} else {                                                        \
2348 	    inp->inp_flags2 &= ~bit;                                    \
2349 	}                                                               \
2350 } while (0)
2351 
2352 			case IP_RECVOPTS:
2353 				OPTSET(INP_RECVOPTS);
2354 				break;
2355 
2356 			case IP_RECVRETOPTS:
2357 				OPTSET(INP_RECVRETOPTS);
2358 				break;
2359 
2360 			case IP_RECVDSTADDR:
2361 				OPTSET(INP_RECVDSTADDR);
2362 				break;
2363 
2364 			case IP_RECVIF:
2365 				OPTSET(INP_RECVIF);
2366 				break;
2367 
2368 			case IP_RECVTTL:
2369 				OPTSET(INP_RECVTTL);
2370 				break;
2371 
2372 			case IP_RECVPKTINFO:
2373 				OPTSET(INP_PKTINFO);
2374 				break;
2375 
2376 			case IP_RECVTOS:
2377 				OPTSET(INP_RECVTOS);
2378 				break;
2379 
2380 			case IP_DONTFRAG:
2381 				/* This option is settable only for IPv4 */
2382 				if (!(inp->inp_vflag & INP_IPV4)) {
2383 					error = EINVAL;
2384 					break;
2385 				}
2386 				OPTSET2(INP2_DONTFRAG);
2387 				break;
2388 #undef OPTSET
2389 #undef OPTSET2
2390 			}
2391 			break;
2392 		/*
2393 		 * Multicast socket options are processed by the in_mcast
2394 		 * module.
2395 		 */
2396 		case IP_MULTICAST_IF:
2397 		case IP_MULTICAST_IFINDEX:
2398 		case IP_MULTICAST_VIF:
2399 		case IP_MULTICAST_TTL:
2400 		case IP_MULTICAST_LOOP:
2401 		case IP_ADD_MEMBERSHIP:
2402 		case IP_DROP_MEMBERSHIP:
2403 		case IP_ADD_SOURCE_MEMBERSHIP:
2404 		case IP_DROP_SOURCE_MEMBERSHIP:
2405 		case IP_BLOCK_SOURCE:
2406 		case IP_UNBLOCK_SOURCE:
2407 		case IP_MSFILTER:
2408 		case MCAST_JOIN_GROUP:
2409 		case MCAST_LEAVE_GROUP:
2410 		case MCAST_JOIN_SOURCE_GROUP:
2411 		case MCAST_LEAVE_SOURCE_GROUP:
2412 		case MCAST_BLOCK_SOURCE:
2413 		case MCAST_UNBLOCK_SOURCE:
2414 			error = inp_setmoptions(inp, sopt);
2415 			break;
2416 
2417 		case IP_PORTRANGE:
2418 			error = sooptcopyin(sopt, &optval, sizeof(optval),
2419 			    sizeof(optval));
2420 			if (error) {
2421 				break;
2422 			}
2423 
2424 			switch (optval) {
2425 			case IP_PORTRANGE_DEFAULT:
2426 				inp->inp_flags &= ~(INP_LOWPORT);
2427 				inp->inp_flags &= ~(INP_HIGHPORT);
2428 				break;
2429 
2430 			case IP_PORTRANGE_HIGH:
2431 				inp->inp_flags &= ~(INP_LOWPORT);
2432 				inp->inp_flags |= INP_HIGHPORT;
2433 				break;
2434 
2435 			case IP_PORTRANGE_LOW:
2436 				inp->inp_flags &= ~(INP_HIGHPORT);
2437 				inp->inp_flags |= INP_LOWPORT;
2438 				break;
2439 
2440 			default:
2441 				error = EINVAL;
2442 				break;
2443 			}
2444 			break;
2445 
2446 #if IPSEC
2447 		case IP_IPSEC_POLICY: {
2448 			caddr_t req = NULL;
2449 			size_t len = 0;
2450 			int priv;
2451 			struct mbuf *m;
2452 			int optname;
2453 
2454 			if ((error = soopt_getm(sopt, &m)) != 0) { /* XXX */
2455 				break;
2456 			}
2457 			if ((error = soopt_mcopyin(sopt, m)) != 0) { /* XXX */
2458 				break;
2459 			}
2460 			priv = (proc_suser(sopt->sopt_p) == 0);
2461 			if (m) {
2462 				req = mtod(m, caddr_t);
2463 				len = m->m_len;
2464 			}
2465 			optname = sopt->sopt_name;
2466 			error = ipsec4_set_policy(inp, optname, req, len, priv);
2467 			m_freem(m);
2468 			break;
2469 		}
2470 #endif /* IPSEC */
2471 
2472 #if TRAFFIC_MGT
2473 		case IP_TRAFFIC_MGT_BACKGROUND: {
2474 			unsigned background = 0;
2475 
2476 			error = sooptcopyin(sopt, &background,
2477 			    sizeof(background), sizeof(background));
2478 			if (error) {
2479 				break;
2480 			}
2481 
2482 			if (background) {
2483 				socket_set_traffic_mgt_flags_locked(so,
2484 				    TRAFFIC_MGT_SO_BACKGROUND);
2485 			} else {
2486 				socket_clear_traffic_mgt_flags_locked(so,
2487 				    TRAFFIC_MGT_SO_BACKGROUND);
2488 			}
2489 
2490 			break;
2491 		}
2492 #endif /* TRAFFIC_MGT */
2493 
2494 		/*
2495 		 * On a multihomed system, scoped routing can be used to
2496 		 * restrict the source interface used for sending packets.
2497 		 * The socket option IP_BOUND_IF binds a particular AF_INET
2498 		 * socket to an interface such that data sent on the socket
2499 		 * is restricted to that interface.  This is unlike the
2500 		 * SO_DONTROUTE option where the routing table is bypassed;
2501 		 * therefore it allows for a greater flexibility and control
2502 		 * over the system behavior, and does not place any restriction
2503 		 * on the destination address type (e.g.  unicast, multicast,
2504 		 * or broadcast if applicable) or whether or not the host is
2505 		 * directly reachable.  Note that in the multicast transmit
2506 		 * case, IP_MULTICAST_{IF,IFINDEX} takes precedence over
2507 		 * IP_BOUND_IF, since the former practically bypasses the
2508 		 * routing table; in this case, IP_BOUND_IF sets the default
2509 		 * interface used for sending multicast packets in the absence
2510 		 * of an explicit multicast transmit interface.
2511 		 */
2512 		case IP_BOUND_IF:
2513 			/* This option is settable only for IPv4 */
2514 			if (!(inp->inp_vflag & INP_IPV4)) {
2515 				error = EINVAL;
2516 				break;
2517 			}
2518 
2519 			error = sooptcopyin(sopt, &optval, sizeof(optval),
2520 			    sizeof(optval));
2521 
2522 			if (error) {
2523 				break;
2524 			}
2525 
2526 			error = inp_bindif(inp, optval, NULL);
2527 			break;
2528 
2529 		case IP_NO_IFT_CELLULAR:
2530 			/* This option is settable only for IPv4 */
2531 			if (!(inp->inp_vflag & INP_IPV4)) {
2532 				error = EINVAL;
2533 				break;
2534 			}
2535 
2536 			error = sooptcopyin(sopt, &optval, sizeof(optval),
2537 			    sizeof(optval));
2538 
2539 			if (error) {
2540 				break;
2541 			}
2542 
2543 			/* once set, it cannot be unset */
2544 			if (!optval && INP_NO_CELLULAR(inp)) {
2545 				error = EINVAL;
2546 				break;
2547 			}
2548 
2549 			error = so_set_restrictions(so,
2550 			    SO_RESTRICT_DENY_CELLULAR);
2551 			break;
2552 
2553 		case IP_OUT_IF:
2554 			/* This option is not settable */
2555 			error = EINVAL;
2556 			break;
2557 
2558 		default:
2559 			error = ENOPROTOOPT;
2560 			break;
2561 		}
2562 		break;
2563 
2564 	case SOPT_GET:
2565 		switch (sopt->sopt_name) {
2566 		case IP_OPTIONS:
2567 		case IP_RETOPTS:
2568 			if (inp->inp_options) {
2569 				error = sooptcopyout(sopt,
2570 				    mtod(inp->inp_options, char *),
2571 				    inp->inp_options->m_len);
2572 			} else {
2573 				sopt->sopt_valsize = 0;
2574 			}
2575 			break;
2576 
2577 		case IP_TOS:
2578 		case IP_TTL:
2579 		case IP_RECVOPTS:
2580 		case IP_RECVRETOPTS:
2581 		case IP_RECVDSTADDR:
2582 		case IP_RECVIF:
2583 		case IP_RECVTTL:
2584 		case IP_PORTRANGE:
2585 		case IP_RECVPKTINFO:
2586 		case IP_RECVTOS:
2587 		case IP_DONTFRAG:
2588 			switch (sopt->sopt_name) {
2589 			case IP_TOS:
2590 				optval = inp->inp_ip_tos;
2591 				break;
2592 
2593 			case IP_TTL:
2594 				optval = inp->inp_ip_ttl;
2595 				break;
2596 
2597 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
2598 #define OPTBIT2(bit)    (inp->inp_flags2 & bit ? 1 : 0)
2599 			case IP_RECVOPTS:
2600 				optval = OPTBIT(INP_RECVOPTS);
2601 				break;
2602 
2603 			case IP_RECVRETOPTS:
2604 				optval = OPTBIT(INP_RECVRETOPTS);
2605 				break;
2606 
2607 			case IP_RECVDSTADDR:
2608 				optval = OPTBIT(INP_RECVDSTADDR);
2609 				break;
2610 
2611 			case IP_RECVIF:
2612 				optval = OPTBIT(INP_RECVIF);
2613 				break;
2614 
2615 			case IP_RECVTTL:
2616 				optval = OPTBIT(INP_RECVTTL);
2617 				break;
2618 
2619 			case IP_PORTRANGE:
2620 				if (inp->inp_flags & INP_HIGHPORT) {
2621 					optval = IP_PORTRANGE_HIGH;
2622 				} else if (inp->inp_flags & INP_LOWPORT) {
2623 					optval = IP_PORTRANGE_LOW;
2624 				} else {
2625 					optval = 0;
2626 				}
2627 				break;
2628 
2629 			case IP_RECVPKTINFO:
2630 				optval = OPTBIT(INP_PKTINFO);
2631 				break;
2632 
2633 			case IP_RECVTOS:
2634 				optval = OPTBIT(INP_RECVTOS);
2635 				break;
2636 			case IP_DONTFRAG:
2637 				optval = OPTBIT2(INP2_DONTFRAG);
2638 				break;
2639 			}
2640 			error = sooptcopyout(sopt, &optval, sizeof(optval));
2641 			break;
2642 
2643 		case IP_MULTICAST_IF:
2644 		case IP_MULTICAST_IFINDEX:
2645 		case IP_MULTICAST_VIF:
2646 		case IP_MULTICAST_TTL:
2647 		case IP_MULTICAST_LOOP:
2648 		case IP_MSFILTER:
2649 			error = inp_getmoptions(inp, sopt);
2650 			break;
2651 
2652 #if IPSEC
2653 		case IP_IPSEC_POLICY: {
2654 			error = 0; /* This option is no longer supported */
2655 			break;
2656 		}
2657 #endif /* IPSEC */
2658 
2659 #if TRAFFIC_MGT
2660 		case IP_TRAFFIC_MGT_BACKGROUND: {
2661 			unsigned background = (so->so_flags1 &
2662 			    SOF1_TRAFFIC_MGT_SO_BACKGROUND) ? 1 : 0;
2663 			return sooptcopyout(sopt, &background,
2664 			           sizeof(background));
2665 		}
2666 #endif /* TRAFFIC_MGT */
2667 
2668 		case IP_BOUND_IF:
2669 			if (inp->inp_flags & INP_BOUND_IF) {
2670 				optval = inp->inp_boundifp->if_index;
2671 			}
2672 			error = sooptcopyout(sopt, &optval, sizeof(optval));
2673 			break;
2674 
2675 		case IP_NO_IFT_CELLULAR:
2676 			optval = INP_NO_CELLULAR(inp) ? 1 : 0;
2677 			error = sooptcopyout(sopt, &optval, sizeof(optval));
2678 			break;
2679 
2680 		case IP_OUT_IF:
2681 			optval = (inp->inp_last_outifp != NULL) ?
2682 			    inp->inp_last_outifp->if_index : 0;
2683 			error = sooptcopyout(sopt, &optval, sizeof(optval));
2684 			break;
2685 
2686 		default:
2687 			error = ENOPROTOOPT;
2688 			break;
2689 		}
2690 		break;
2691 	}
2692 	return error;
2693 }
2694 
2695 /*
2696  * Set up IP options in pcb for insertion in output packets.
2697  * Store in mbuf with pointer in pcbopt, adding pseudo-option
2698  * with destination address if source routed.
2699  */
2700 static int
ip_pcbopts(int optname,struct mbuf ** pcbopt,struct mbuf * m)2701 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
2702 {
2703 #pragma unused(optname)
2704 	int cnt, optlen;
2705 	u_char *cp;
2706 	u_char opt;
2707 
2708 	/* turn off any old options */
2709 	if (*pcbopt) {
2710 		(void) m_free(*pcbopt);
2711 	}
2712 	*pcbopt = 0;
2713 	if (m == (struct mbuf *)0 || m->m_len == 0) {
2714 		/*
2715 		 * Only turning off any previous options.
2716 		 */
2717 		if (m) {
2718 			(void) m_free(m);
2719 		}
2720 		return 0;
2721 	}
2722 
2723 	if (m->m_len % sizeof(int32_t)) {
2724 		goto bad;
2725 	}
2726 
2727 	/*
2728 	 * IP first-hop destination address will be stored before
2729 	 * actual options; move other options back
2730 	 * and clear it when none present.
2731 	 */
2732 	if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN]) {
2733 		goto bad;
2734 	}
2735 	cnt = m->m_len;
2736 	m->m_len += sizeof(struct in_addr);
2737 	cp = mtod(m, u_char *) + sizeof(struct in_addr);
2738 	ovbcopy(mtod(m, caddr_t), (caddr_t)cp, (unsigned)cnt);
2739 	bzero(mtod(m, caddr_t), sizeof(struct in_addr));
2740 
2741 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
2742 		opt = cp[IPOPT_OPTVAL];
2743 		if (opt == IPOPT_EOL) {
2744 			break;
2745 		}
2746 		if (opt == IPOPT_NOP) {
2747 			optlen = 1;
2748 		} else {
2749 			if (cnt < IPOPT_OLEN + sizeof(*cp)) {
2750 				goto bad;
2751 			}
2752 			optlen = cp[IPOPT_OLEN];
2753 			if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
2754 				goto bad;
2755 			}
2756 		}
2757 		switch (opt) {
2758 		default:
2759 			break;
2760 
2761 		case IPOPT_LSRR:
2762 		case IPOPT_SSRR:
2763 			/*
2764 			 * user process specifies route as:
2765 			 *	->A->B->C->D
2766 			 * D must be our final destination (but we can't
2767 			 * check that since we may not have connected yet).
2768 			 * A is first hop destination, which doesn't appear in
2769 			 * actual IP option, but is stored before the options.
2770 			 */
2771 			if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr)) {
2772 				goto bad;
2773 			}
2774 			if (optlen > UINT8_MAX) {
2775 				goto bad;
2776 			}
2777 			m->m_len -= sizeof(struct in_addr);
2778 			cnt -= sizeof(struct in_addr);
2779 			optlen -= sizeof(struct in_addr);
2780 			cp[IPOPT_OLEN] = (uint8_t)optlen;
2781 			/*
2782 			 * Move first hop before start of options.
2783 			 */
2784 			bcopy((caddr_t)&cp[IPOPT_OFFSET + 1], mtod(m, caddr_t),
2785 			    sizeof(struct in_addr));
2786 			/*
2787 			 * Then copy rest of options back
2788 			 * to close up the deleted entry.
2789 			 */
2790 			ovbcopy((caddr_t)(&cp[IPOPT_OFFSET + 1] +
2791 			    sizeof(struct in_addr)),
2792 			    (caddr_t)&cp[IPOPT_OFFSET + 1],
2793 			    (unsigned)cnt - (IPOPT_MINOFF - 1));
2794 			break;
2795 		}
2796 	}
2797 	if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr)) {
2798 		goto bad;
2799 	}
2800 	*pcbopt = m;
2801 	return 0;
2802 
2803 bad:
2804 	(void) m_free(m);
2805 	return EINVAL;
2806 }
2807 
2808 void
ip_moptions_init(void)2809 ip_moptions_init(void)
2810 {
2811 	PE_parse_boot_argn("ifa_debug", &imo_debug, sizeof(imo_debug));
2812 
2813 	vm_size_t imo_size = (imo_debug == 0) ? sizeof(struct ip_moptions) :
2814 	    sizeof(struct ip_moptions_dbg);
2815 
2816 	imo_zone = zone_create(IMO_ZONE_NAME, imo_size, ZC_ZFREE_CLEARMEM);
2817 }
2818 
2819 void
imo_addref(struct ip_moptions * imo,int locked)2820 imo_addref(struct ip_moptions *imo, int locked)
2821 {
2822 	if (!locked) {
2823 		IMO_LOCK(imo);
2824 	} else {
2825 		IMO_LOCK_ASSERT_HELD(imo);
2826 	}
2827 
2828 	if (++imo->imo_refcnt == 0) {
2829 		panic("%s: imo %p wraparound refcnt", __func__, imo);
2830 		/* NOTREACHED */
2831 	} else if (imo->imo_trace != NULL) {
2832 		(*imo->imo_trace)(imo, TRUE);
2833 	}
2834 
2835 	if (!locked) {
2836 		IMO_UNLOCK(imo);
2837 	}
2838 }
2839 
2840 void
imo_remref(struct ip_moptions * imo)2841 imo_remref(struct ip_moptions *imo)
2842 {
2843 	IMO_LOCK(imo);
2844 	if (imo->imo_refcnt == 0) {
2845 		panic("%s: imo %p negative refcnt", __func__, imo);
2846 		/* NOTREACHED */
2847 	} else if (imo->imo_trace != NULL) {
2848 		(*imo->imo_trace)(imo, FALSE);
2849 	}
2850 
2851 	--imo->imo_refcnt;
2852 	if (imo->imo_refcnt > 0) {
2853 		IMO_UNLOCK(imo);
2854 		return;
2855 	}
2856 
2857 	IMO_PURGE_LOCKED(imo);
2858 
2859 	IMO_UNLOCK(imo);
2860 
2861 	kfree_type(struct in_multi *, imo->imo_max_memberships, imo->imo_membership);
2862 	kfree_type(struct in_mfilter, imo->imo_max_memberships, imo->imo_mfilters);
2863 	lck_mtx_destroy(&imo->imo_lock, &ifa_mtx_grp);
2864 
2865 	if (!(imo->imo_debug & IFD_ALLOC)) {
2866 		panic("%s: imo %p cannot be freed", __func__, imo);
2867 		/* NOTREACHED */
2868 	}
2869 	zfree(imo_zone, imo);
2870 }
2871 
2872 static void
imo_trace(struct ip_moptions * imo,int refhold)2873 imo_trace(struct ip_moptions *imo, int refhold)
2874 {
2875 	struct ip_moptions_dbg *imo_dbg = (struct ip_moptions_dbg *)imo;
2876 	ctrace_t *tr;
2877 	u_int32_t idx;
2878 	u_int16_t *cnt;
2879 
2880 	if (!(imo->imo_debug & IFD_DEBUG)) {
2881 		panic("%s: imo %p has no debug structure", __func__, imo);
2882 		/* NOTREACHED */
2883 	}
2884 	if (refhold) {
2885 		cnt = &imo_dbg->imo_refhold_cnt;
2886 		tr = imo_dbg->imo_refhold;
2887 	} else {
2888 		cnt = &imo_dbg->imo_refrele_cnt;
2889 		tr = imo_dbg->imo_refrele;
2890 	}
2891 
2892 	idx = atomic_add_16_ov(cnt, 1) % IMO_TRACE_HIST_SIZE;
2893 	ctrace_record(&tr[idx]);
2894 }
2895 
2896 struct ip_moptions *
ip_allocmoptions(zalloc_flags_t how)2897 ip_allocmoptions(zalloc_flags_t how)
2898 {
2899 	struct ip_moptions *imo;
2900 
2901 	imo = zalloc_flags(imo_zone, how | Z_ZERO);
2902 	if (imo != NULL) {
2903 		lck_mtx_init(&imo->imo_lock, &ifa_mtx_grp, &ifa_mtx_attr);
2904 		imo->imo_debug |= IFD_ALLOC;
2905 		if (imo_debug != 0) {
2906 			imo->imo_debug |= IFD_DEBUG;
2907 			imo->imo_trace = imo_trace;
2908 		}
2909 		IMO_ADDREF(imo);
2910 	}
2911 
2912 	return imo;
2913 }
2914 
2915 /*
2916  * Routine called from ip_output() to loop back a copy of an IP multicast
2917  * packet to the input queue of a specified interface.  Note that this
2918  * calls the output routine of the loopback "driver", but with an interface
2919  * pointer that might NOT be a loopback interface -- evil, but easier than
2920  * replicating that code here.
2921  */
2922 static void
ip_mloopback(struct ifnet * srcifp,struct ifnet * origifp,struct mbuf * m,struct sockaddr_in * dst,int hlen)2923 ip_mloopback(struct ifnet *srcifp, struct ifnet *origifp, struct mbuf *m,
2924     struct sockaddr_in *dst, int hlen)
2925 {
2926 	struct mbuf *copym;
2927 	struct ip *ip;
2928 
2929 	if (lo_ifp == NULL) {
2930 		return;
2931 	}
2932 
2933 	/*
2934 	 * Copy the packet header as it's needed for the checksum
2935 	 * Make sure to deep-copy IP header portion in case the data
2936 	 * is in an mbuf cluster, so that we can safely override the IP
2937 	 * header portion later.
2938 	 */
2939 	copym = m_copym_mode(m, 0, M_COPYALL, M_DONTWAIT, M_COPYM_COPY_HDR);
2940 	if (copym != NULL && ((copym->m_flags & M_EXT) || copym->m_len < hlen)) {
2941 		copym = m_pullup(copym, hlen);
2942 	}
2943 
2944 	if (copym == NULL) {
2945 		return;
2946 	}
2947 
2948 	/*
2949 	 * We don't bother to fragment if the IP length is greater
2950 	 * than the interface's MTU.  Can this possibly matter?
2951 	 */
2952 	ip = mtod(copym, struct ip *);
2953 #if BYTE_ORDER != BIG_ENDIAN
2954 	HTONS(ip->ip_len);
2955 	HTONS(ip->ip_off);
2956 #endif
2957 	ip->ip_sum = 0;
2958 	ip->ip_sum = ip_cksum_hdr_out(copym, hlen);
2959 
2960 	/*
2961 	 * Mark checksum as valid unless receive checksum offload is
2962 	 * disabled; if so, compute checksum in software.  If the
2963 	 * interface itself is lo0, this will be overridden by if_loop.
2964 	 */
2965 	if (hwcksum_rx) {
2966 		copym->m_pkthdr.csum_flags &= ~(CSUM_PARTIAL | CSUM_ZERO_INVERT);
2967 		copym->m_pkthdr.csum_flags |=
2968 		    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2969 		copym->m_pkthdr.csum_data = 0xffff;
2970 	} else if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2971 #if BYTE_ORDER != BIG_ENDIAN
2972 		NTOHS(ip->ip_len);
2973 #endif
2974 		in_delayed_cksum(copym);
2975 #if BYTE_ORDER != BIG_ENDIAN
2976 		HTONS(ip->ip_len);
2977 #endif
2978 	}
2979 
2980 	/*
2981 	 * Stuff the 'real' ifp into the pkthdr, to be used in matching
2982 	 * in ip_input(); we need the loopback ifp/dl_tag passed as args
2983 	 * to make the loopback driver compliant with the data link
2984 	 * requirements.
2985 	 */
2986 	copym->m_pkthdr.rcvif = origifp;
2987 
2988 	/*
2989 	 * Also record the source interface (which owns the source address).
2990 	 * This is basically a stripped down version of ifa_foraddr().
2991 	 */
2992 	if (srcifp == NULL) {
2993 		struct in_ifaddr *ia;
2994 
2995 		lck_rw_lock_shared(&in_ifaddr_rwlock);
2996 		TAILQ_FOREACH(ia, INADDR_HASH(ip->ip_src.s_addr), ia_hash) {
2997 			IFA_LOCK_SPIN(&ia->ia_ifa);
2998 			if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_src.s_addr) {
2999 				srcifp = ia->ia_ifp;
3000 				IFA_UNLOCK(&ia->ia_ifa);
3001 				break;
3002 			}
3003 			IFA_UNLOCK(&ia->ia_ifa);
3004 		}
3005 		lck_rw_done(&in_ifaddr_rwlock);
3006 	}
3007 	if (srcifp != NULL) {
3008 		ip_setsrcifaddr_info(copym, srcifp->if_index, NULL);
3009 	}
3010 	ip_setdstifaddr_info(copym, origifp->if_index, NULL);
3011 
3012 	dlil_output(lo_ifp, PF_INET, copym, NULL, SA(dst), 0, NULL);
3013 }
3014 
3015 /*
3016  * Given a source IP address (and route, if available), determine the best
3017  * interface to send the packet from.  Checking for (and updating) the
3018  * ROF_SRCIF_SELECTED flag in the pcb-supplied route placeholder is done
3019  * without any locks based on the assumption that ip_output() is single-
3020  * threaded per-pcb, i.e. for any given pcb there can only be one thread
3021  * performing output at the IP layer.
3022  *
3023  * This routine is analogous to in6_selectroute() for IPv6.
3024  */
3025 static struct ifaddr *
in_selectsrcif(struct ip * ip,struct route * ro,unsigned int ifscope)3026 in_selectsrcif(struct ip *ip, struct route *ro, unsigned int ifscope)
3027 {
3028 	struct ifaddr *ifa = NULL;
3029 	struct in_addr src = ip->ip_src;
3030 	struct in_addr dst = ip->ip_dst;
3031 	struct ifnet *rt_ifp;
3032 	char s_src[MAX_IPv4_STR_LEN], s_dst[MAX_IPv4_STR_LEN];
3033 
3034 	VERIFY(src.s_addr != INADDR_ANY);
3035 
3036 	if (ip_select_srcif_debug) {
3037 		(void) inet_ntop(AF_INET, &src.s_addr, s_src, sizeof(s_src));
3038 		(void) inet_ntop(AF_INET, &dst.s_addr, s_dst, sizeof(s_dst));
3039 	}
3040 
3041 	if (ro->ro_rt != NULL) {
3042 		RT_LOCK(ro->ro_rt);
3043 	}
3044 
3045 	rt_ifp = (ro->ro_rt != NULL) ? ro->ro_rt->rt_ifp : NULL;
3046 
3047 	/*
3048 	 * Given the source IP address, find a suitable source interface
3049 	 * to use for transmission; if the caller has specified a scope,
3050 	 * optimize the search by looking at the addresses only for that
3051 	 * interface.  This is still suboptimal, however, as we need to
3052 	 * traverse the per-interface list.
3053 	 */
3054 	if (ifscope != IFSCOPE_NONE || ro->ro_rt != NULL) {
3055 		unsigned int scope = ifscope;
3056 
3057 		/*
3058 		 * If no scope is specified and the route is stale (pointing
3059 		 * to a defunct interface) use the current primary interface;
3060 		 * this happens when switching between interfaces configured
3061 		 * with the same IP address.  Otherwise pick up the scope
3062 		 * information from the route; the ULP may have looked up a
3063 		 * correct route and we just need to verify it here and mark
3064 		 * it with the ROF_SRCIF_SELECTED flag below.
3065 		 */
3066 		if (scope == IFSCOPE_NONE) {
3067 			scope = rt_ifp->if_index;
3068 			if (scope != get_primary_ifscope(AF_INET) &&
3069 			    ROUTE_UNUSABLE(ro)) {
3070 				scope = get_primary_ifscope(AF_INET);
3071 			}
3072 		}
3073 
3074 		ifa = (struct ifaddr *)ifa_foraddr_scoped(src.s_addr, scope);
3075 
3076 		if (ifa == NULL && ip->ip_p != IPPROTO_UDP &&
3077 		    ip->ip_p != IPPROTO_TCP && ipforwarding) {
3078 			/*
3079 			 * If forwarding is enabled, and if the packet isn't
3080 			 * TCP or UDP, check if the source address belongs
3081 			 * to one of our own interfaces; if so, demote the
3082 			 * interface scope and do a route lookup right below.
3083 			 */
3084 			ifa = (struct ifaddr *)ifa_foraddr(src.s_addr);
3085 			if (ifa != NULL) {
3086 				IFA_REMREF(ifa);
3087 				ifa = NULL;
3088 				ifscope = IFSCOPE_NONE;
3089 			}
3090 		}
3091 
3092 		if (ip_select_srcif_debug && ifa != NULL) {
3093 			if (ro->ro_rt != NULL) {
3094 				printf("%s->%s ifscope %d->%d ifa_if %s "
3095 				    "ro_if %s\n", s_src, s_dst, ifscope,
3096 				    scope, if_name(ifa->ifa_ifp),
3097 				    if_name(rt_ifp));
3098 			} else {
3099 				printf("%s->%s ifscope %d->%d ifa_if %s\n",
3100 				    s_src, s_dst, ifscope, scope,
3101 				    if_name(ifa->ifa_ifp));
3102 			}
3103 		}
3104 	}
3105 
3106 	/*
3107 	 * Slow path; search for an interface having the corresponding source
3108 	 * IP address if the scope was not specified by the caller, and:
3109 	 *
3110 	 *   1) There currently isn't any route, or,
3111 	 *   2) The interface used by the route does not own that source
3112 	 *	IP address; in this case, the route will get blown away
3113 	 *	and we'll do a more specific scoped search using the newly
3114 	 *	found interface.
3115 	 */
3116 	if (ifa == NULL && ifscope == IFSCOPE_NONE) {
3117 		ifa = (struct ifaddr *)ifa_foraddr(src.s_addr);
3118 
3119 		/*
3120 		 * If we have the IP address, but not the route, we don't
3121 		 * really know whether or not it belongs to the correct
3122 		 * interface (it could be shared across multiple interfaces.)
3123 		 * The only way to find out is to do a route lookup.
3124 		 */
3125 		if (ifa != NULL && ro->ro_rt == NULL) {
3126 			struct rtentry *rt;
3127 			struct sockaddr_in sin;
3128 			struct ifaddr *oifa = NULL;
3129 
3130 			bzero(&sin, sizeof(sin));
3131 			sin.sin_family = AF_INET;
3132 			sin.sin_len = sizeof(sin);
3133 			sin.sin_addr = dst;
3134 
3135 			lck_mtx_lock(rnh_lock);
3136 			if ((rt = rt_lookup(TRUE, SA(&sin), NULL,
3137 			    rt_tables[AF_INET], IFSCOPE_NONE)) != NULL) {
3138 				RT_LOCK(rt);
3139 				/*
3140 				 * If the route uses a different interface,
3141 				 * use that one instead.  The IP address of
3142 				 * the ifaddr that we pick up here is not
3143 				 * relevant.
3144 				 */
3145 				if (ifa->ifa_ifp != rt->rt_ifp) {
3146 					oifa = ifa;
3147 					ifa = rt->rt_ifa;
3148 					IFA_ADDREF(ifa);
3149 					RT_UNLOCK(rt);
3150 				} else {
3151 					RT_UNLOCK(rt);
3152 				}
3153 				rtfree_locked(rt);
3154 			}
3155 			lck_mtx_unlock(rnh_lock);
3156 
3157 			if (oifa != NULL) {
3158 				struct ifaddr *iifa;
3159 
3160 				/*
3161 				 * See if the interface pointed to by the
3162 				 * route is configured with the source IP
3163 				 * address of the packet.
3164 				 */
3165 				iifa = (struct ifaddr *)ifa_foraddr_scoped(
3166 					src.s_addr, ifa->ifa_ifp->if_index);
3167 
3168 				if (iifa != NULL) {
3169 					/*
3170 					 * Found it; drop the original one
3171 					 * as well as the route interface
3172 					 * address, and use this instead.
3173 					 */
3174 					IFA_REMREF(oifa);
3175 					IFA_REMREF(ifa);
3176 					ifa = iifa;
3177 				} else if (!ipforwarding ||
3178 				    (rt->rt_flags & RTF_GATEWAY)) {
3179 					/*
3180 					 * This interface doesn't have that
3181 					 * source IP address; drop the route
3182 					 * interface address and just use the
3183 					 * original one, and let the caller
3184 					 * do a scoped route lookup.
3185 					 */
3186 					IFA_REMREF(ifa);
3187 					ifa = oifa;
3188 				} else {
3189 					/*
3190 					 * Forwarding is enabled and the source
3191 					 * address belongs to one of our own
3192 					 * interfaces which isn't the outgoing
3193 					 * interface, and we have a route, and
3194 					 * the destination is on a network that
3195 					 * is directly attached (onlink); drop
3196 					 * the original one and use the route
3197 					 * interface address instead.
3198 					 */
3199 					IFA_REMREF(oifa);
3200 				}
3201 			}
3202 		} else if (ifa != NULL && ro->ro_rt != NULL &&
3203 		    !(ro->ro_rt->rt_flags & RTF_GATEWAY) &&
3204 		    ifa->ifa_ifp != ro->ro_rt->rt_ifp && ipforwarding) {
3205 			/*
3206 			 * Forwarding is enabled and the source address belongs
3207 			 * to one of our own interfaces which isn't the same
3208 			 * as the interface used by the known route; drop the
3209 			 * original one and use the route interface address.
3210 			 */
3211 			IFA_REMREF(ifa);
3212 			ifa = ro->ro_rt->rt_ifa;
3213 			IFA_ADDREF(ifa);
3214 		}
3215 
3216 		if (ip_select_srcif_debug && ifa != NULL) {
3217 			printf("%s->%s ifscope %d ifa_if %s\n",
3218 			    s_src, s_dst, ifscope, if_name(ifa->ifa_ifp));
3219 		}
3220 	}
3221 
3222 	if (ro->ro_rt != NULL) {
3223 		RT_LOCK_ASSERT_HELD(ro->ro_rt);
3224 	}
3225 	/*
3226 	 * If there is a non-loopback route with the wrong interface, or if
3227 	 * there is no interface configured with such an address, blow it
3228 	 * away.  Except for local/loopback, we look for one with a matching
3229 	 * interface scope/index.
3230 	 */
3231 	if (ro->ro_rt != NULL &&
3232 	    (ifa == NULL || (ifa->ifa_ifp != rt_ifp && rt_ifp != lo_ifp) ||
3233 	    !(ro->ro_rt->rt_flags & RTF_UP))) {
3234 		if (ip_select_srcif_debug) {
3235 			if (ifa != NULL) {
3236 				printf("%s->%s ifscope %d ro_if %s != "
3237 				    "ifa_if %s (cached route cleared)\n",
3238 				    s_src, s_dst, ifscope, if_name(rt_ifp),
3239 				    if_name(ifa->ifa_ifp));
3240 			} else {
3241 				printf("%s->%s ifscope %d ro_if %s "
3242 				    "(no ifa_if found)\n",
3243 				    s_src, s_dst, ifscope, if_name(rt_ifp));
3244 			}
3245 		}
3246 
3247 		RT_UNLOCK(ro->ro_rt);
3248 		ROUTE_RELEASE(ro);
3249 
3250 		/*
3251 		 * If the destination is IPv4 LLA and the route's interface
3252 		 * doesn't match the source interface, then the source IP
3253 		 * address is wrong; it most likely belongs to the primary
3254 		 * interface associated with the IPv4 LL subnet.  Drop the
3255 		 * packet rather than letting it go out and return an error
3256 		 * to the ULP.  This actually applies not only to IPv4 LL
3257 		 * but other shared subnets; for now we explicitly test only
3258 		 * for the former case and save the latter for future.
3259 		 */
3260 		if (IN_LINKLOCAL(ntohl(dst.s_addr)) &&
3261 		    !IN_LINKLOCAL(ntohl(src.s_addr)) && ifa != NULL) {
3262 			IFA_REMREF(ifa);
3263 			ifa = NULL;
3264 		}
3265 	}
3266 
3267 	if (ip_select_srcif_debug && ifa == NULL) {
3268 		printf("%s->%s ifscope %d (neither ro_if/ifa_if found)\n",
3269 		    s_src, s_dst, ifscope);
3270 	}
3271 
3272 	/*
3273 	 * If there is a route, mark it accordingly.  If there isn't one,
3274 	 * we'll get here again during the next transmit (possibly with a
3275 	 * route) and the flag will get set at that point.  For IPv4 LLA
3276 	 * destination, mark it only if the route has been fully resolved;
3277 	 * otherwise we want to come back here again when the route points
3278 	 * to the interface over which the ARP reply arrives on.
3279 	 */
3280 	if (ro->ro_rt != NULL && (!IN_LINKLOCAL(ntohl(dst.s_addr)) ||
3281 	    (ro->ro_rt->rt_gateway->sa_family == AF_LINK &&
3282 	    SDL(ro->ro_rt->rt_gateway)->sdl_alen != 0))) {
3283 		if (ifa != NULL) {
3284 			IFA_ADDREF(ifa);        /* for route */
3285 		}
3286 		if (ro->ro_srcia != NULL) {
3287 			IFA_REMREF(ro->ro_srcia);
3288 		}
3289 		ro->ro_srcia = ifa;
3290 		ro->ro_flags |= ROF_SRCIF_SELECTED;
3291 		RT_GENID_SYNC(ro->ro_rt);
3292 	}
3293 
3294 	if (ro->ro_rt != NULL) {
3295 		RT_UNLOCK(ro->ro_rt);
3296 	}
3297 
3298 	return ifa;
3299 }
3300 
3301 /*
3302  * @brief	Given outgoing interface it determines what checksum needs
3303  *      to be computed in software and what needs to be offloaded to the
3304  *      interface.
3305  *
3306  * @param	ifp Pointer to the outgoing interface
3307  * @param	m Pointer to the packet
3308  * @param	hlen IP header length
3309  * @param	ip_len Total packet size i.e. headers + data payload
3310  * @param	sw_csum Pointer to a software checksum flag set
3311  *
3312  * @return	void
3313  */
3314 void
ip_output_checksum(struct ifnet * ifp,struct mbuf * m,int hlen,int ip_len,uint32_t * sw_csum)3315 ip_output_checksum(struct ifnet *ifp, struct mbuf *m, int hlen, int ip_len,
3316     uint32_t *sw_csum)
3317 {
3318 	int tso = TSO_IPV4_OK(ifp, m);
3319 	uint32_t hwcap = ifp->if_hwassist;
3320 
3321 	m->m_pkthdr.csum_flags |= CSUM_IP;
3322 
3323 	if (!hwcksum_tx) {
3324 		/* do all in software; hardware checksum offload is disabled */
3325 		*sw_csum = (CSUM_DELAY_DATA | CSUM_DELAY_IP) &
3326 		    m->m_pkthdr.csum_flags;
3327 	} else {
3328 		/* do in software what the hardware cannot */
3329 		*sw_csum = m->m_pkthdr.csum_flags &
3330 		    ~IF_HWASSIST_CSUM_FLAGS(hwcap);
3331 	}
3332 
3333 	if (hlen != sizeof(struct ip)) {
3334 		*sw_csum |= ((CSUM_DELAY_DATA | CSUM_DELAY_IP) &
3335 		    m->m_pkthdr.csum_flags);
3336 	} else if (!(*sw_csum & CSUM_DELAY_DATA) && (hwcap & CSUM_PARTIAL)) {
3337 		int interface_mtu = ifp->if_mtu;
3338 
3339 		if (INTF_ADJUST_MTU_FOR_CLAT46(ifp)) {
3340 			interface_mtu = IN6_LINKMTU(ifp);
3341 			/* Further adjust the size for CLAT46 expansion */
3342 			interface_mtu -= CLAT46_HDR_EXPANSION_OVERHD;
3343 		}
3344 
3345 		/*
3346 		 * Partial checksum offload, if non-IP fragment, and TCP only
3347 		 * (no UDP support, as the hardware may not be able to convert
3348 		 * +0 to -0 (0xffff) per RFC1122 4.1.3.4. unless the interface
3349 		 * supports "invert zero" capability.)
3350 		 */
3351 		if (hwcksum_tx && !tso &&
3352 		    ((m->m_pkthdr.csum_flags & CSUM_TCP) ||
3353 		    ((hwcap & CSUM_ZERO_INVERT) &&
3354 		    (m->m_pkthdr.csum_flags & CSUM_ZERO_INVERT))) &&
3355 		    ip_len <= interface_mtu) {
3356 			uint16_t start = sizeof(struct ip);
3357 			uint16_t ulpoff = m->m_pkthdr.csum_data & 0xffff;
3358 			m->m_pkthdr.csum_flags |=
3359 			    (CSUM_DATA_VALID | CSUM_PARTIAL);
3360 			m->m_pkthdr.csum_tx_stuff = (ulpoff + start);
3361 			m->m_pkthdr.csum_tx_start = start;
3362 			/* do IP hdr chksum in software */
3363 			*sw_csum = CSUM_DELAY_IP;
3364 		} else {
3365 			*sw_csum |= (CSUM_DELAY_DATA & m->m_pkthdr.csum_flags);
3366 		}
3367 	}
3368 
3369 	if (*sw_csum & CSUM_DELAY_DATA) {
3370 		in_delayed_cksum(m);
3371 		*sw_csum &= ~CSUM_DELAY_DATA;
3372 	}
3373 
3374 	if (hwcksum_tx) {
3375 		/*
3376 		 * Drop off bits that aren't supported by hardware;
3377 		 * also make sure to preserve non-checksum related bits.
3378 		 */
3379 		m->m_pkthdr.csum_flags =
3380 		    ((m->m_pkthdr.csum_flags &
3381 		    (IF_HWASSIST_CSUM_FLAGS(hwcap) | CSUM_DATA_VALID)) |
3382 		    (m->m_pkthdr.csum_flags & ~IF_HWASSIST_CSUM_MASK));
3383 	} else {
3384 		/* drop all bits; hardware checksum offload is disabled */
3385 		m->m_pkthdr.csum_flags = 0;
3386 	}
3387 }
3388 
3389 /*
3390  * GRE protocol output for PPP/PPTP
3391  */
3392 int
ip_gre_output(struct mbuf * m)3393 ip_gre_output(struct mbuf *m)
3394 {
3395 	struct route ro;
3396 	int error;
3397 
3398 	bzero(&ro, sizeof(ro));
3399 
3400 	error = ip_output(m, NULL, &ro, 0, NULL, NULL);
3401 
3402 	ROUTE_RELEASE(&ro);
3403 
3404 	return error;
3405 }
3406 
3407 static int
3408 sysctl_reset_ip_output_stats SYSCTL_HANDLER_ARGS
3409 {
3410 #pragma unused(arg1, arg2)
3411 	int error, i;
3412 
3413 	i = ip_output_measure;
3414 	error = sysctl_handle_int(oidp, &i, 0, req);
3415 	if (error || req->newptr == USER_ADDR_NULL) {
3416 		goto done;
3417 	}
3418 	/* impose bounds */
3419 	if (i < 0 || i > 1) {
3420 		error = EINVAL;
3421 		goto done;
3422 	}
3423 	if (ip_output_measure != i && i == 1) {
3424 		net_perf_initialize(&net_perf, ip_output_measure_bins);
3425 	}
3426 	ip_output_measure = i;
3427 done:
3428 	return error;
3429 }
3430 
3431 static int
3432 sysctl_ip_output_measure_bins SYSCTL_HANDLER_ARGS
3433 {
3434 #pragma unused(arg1, arg2)
3435 	int error;
3436 	uint64_t i;
3437 
3438 	i = ip_output_measure_bins;
3439 	error = sysctl_handle_quad(oidp, &i, 0, req);
3440 	if (error || req->newptr == USER_ADDR_NULL) {
3441 		goto done;
3442 	}
3443 	/* validate data */
3444 	if (!net_perf_validate_bins(i)) {
3445 		error = EINVAL;
3446 		goto done;
3447 	}
3448 	ip_output_measure_bins = i;
3449 done:
3450 	return error;
3451 }
3452 
3453 static int
3454 sysctl_ip_output_getperf SYSCTL_HANDLER_ARGS
3455 {
3456 #pragma unused(oidp, arg1, arg2)
3457 	if (req->oldptr == USER_ADDR_NULL) {
3458 		req->oldlen = (size_t)sizeof(struct ipstat);
3459 	}
3460 
3461 	return SYSCTL_OUT(req, &net_perf, MIN(sizeof(net_perf), req->oldlen));
3462 }
3463