xref: /freebsd/sys/netinet/ip_output.c (revision 35a04710d7286aa9538917fd7f8e417dbee95b82)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ip_output.c	8.3 (Berkeley) 1/21/94
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_ipfw.h"
36 #include "opt_ipsec.h"
37 #include "opt_mac.h"
38 #include "opt_mbuf_stress_test.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/priv.h>
46 #include <sys/protosw.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/sysctl.h>
50 
51 #include <net/if.h>
52 #include <net/netisr.h>
53 #include <net/pfil.h>
54 #include <net/route.h>
55 
56 #include <netinet/in.h>
57 #include <netinet/in_systm.h>
58 #include <netinet/ip.h>
59 #include <netinet/in_pcb.h>
60 #include <netinet/in_var.h>
61 #include <netinet/ip_var.h>
62 #include <netinet/ip_options.h>
63 
64 #ifdef IPSEC
65 #include <netinet/ip_ipsec.h>
66 #include <netipsec/ipsec.h>
67 #endif /* IPSEC*/
68 
69 #include <machine/in_cksum.h>
70 
71 #include <security/mac/mac_framework.h>
72 
73 #define print_ip(x, a, y)	 printf("%s %d.%d.%d.%d%s",\
74 				x, (ntohl(a.s_addr)>>24)&0xFF,\
75 				  (ntohl(a.s_addr)>>16)&0xFF,\
76 				  (ntohl(a.s_addr)>>8)&0xFF,\
77 				  (ntohl(a.s_addr))&0xFF, y);
78 
79 u_short ip_id;
80 
81 #ifdef MBUF_STRESS_TEST
82 int mbuf_frag_size = 0;
83 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
84 	&mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
85 #endif
86 
87 static void	ip_mloopback
88 	(struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
89 
90 
91 extern	struct protosw inetsw[];
92 
93 /*
94  * IP output.  The packet in mbuf chain m contains a skeletal IP
95  * header (with len, off, ttl, proto, tos, src, dst).
96  * The mbuf chain containing the packet will be freed.
97  * The mbuf opt, if present, will not be freed.
98  * In the IP forwarding case, the packet will arrive with options already
99  * inserted, so must have a NULL opt pointer.
100  */
101 int
102 ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags,
103     struct ip_moptions *imo, struct inpcb *inp)
104 {
105 	struct ip *ip;
106 	struct ifnet *ifp = NULL;	/* keep compiler happy */
107 	struct mbuf *m0;
108 	int hlen = sizeof (struct ip);
109 	int mtu;
110 	int len, error = 0;
111 	struct sockaddr_in *dst = NULL;	/* keep compiler happy */
112 	struct in_ifaddr *ia = NULL;
113 	int isbroadcast, sw_csum;
114 	struct route iproute;
115 	struct in_addr odst;
116 #ifdef IPFIREWALL_FORWARD
117 	struct m_tag *fwd_tag = NULL;
118 #endif
119 	M_ASSERTPKTHDR(m);
120 
121 	if (ro == NULL) {
122 		ro = &iproute;
123 		bzero(ro, sizeof (*ro));
124 	}
125 
126 	if (inp != NULL)
127 		INP_LOCK_ASSERT(inp);
128 
129 	if (opt) {
130 		len = 0;
131 		m = ip_insertoptions(m, opt, &len);
132 		if (len != 0)
133 			hlen = len;
134 	}
135 	ip = mtod(m, struct ip *);
136 
137 	/*
138 	 * Fill in IP header.  If we are not allowing fragmentation,
139 	 * then the ip_id field is meaningless, but we don't set it
140 	 * to zero.  Doing so causes various problems when devices along
141 	 * the path (routers, load balancers, firewalls, etc.) illegally
142 	 * disable DF on our packet.  Note that a 16-bit counter
143 	 * will wrap around in less than 10 seconds at 100 Mbit/s on a
144 	 * medium with MTU 1500.  See Steven M. Bellovin, "A Technique
145 	 * for Counting NATted Hosts", Proc. IMW'02, available at
146 	 * <http://www.cs.columbia.edu/~smb/papers/fnat.pdf>.
147 	 */
148 	if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
149 		ip->ip_v = IPVERSION;
150 		ip->ip_hl = hlen >> 2;
151 		ip->ip_id = ip_newid();
152 		ipstat.ips_localout++;
153 	} else {
154 		hlen = ip->ip_hl << 2;
155 	}
156 
157 	dst = (struct sockaddr_in *)&ro->ro_dst;
158 again:
159 	/*
160 	 * If there is a cached route,
161 	 * check that it is to the same destination
162 	 * and is still up.  If not, free it and try again.
163 	 * The address family should also be checked in case of sharing the
164 	 * cache with IPv6.
165 	 */
166 	if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
167 			  dst->sin_family != AF_INET ||
168 			  dst->sin_addr.s_addr != ip->ip_dst.s_addr)) {
169 		RTFREE(ro->ro_rt);
170 		ro->ro_rt = (struct rtentry *)NULL;
171 	}
172 #ifdef IPFIREWALL_FORWARD
173 	if (ro->ro_rt == NULL && fwd_tag == NULL) {
174 #else
175 	if (ro->ro_rt == NULL) {
176 #endif
177 		bzero(dst, sizeof(*dst));
178 		dst->sin_family = AF_INET;
179 		dst->sin_len = sizeof(*dst);
180 		dst->sin_addr = ip->ip_dst;
181 	}
182 	/*
183 	 * If routing to interface only, short circuit routing lookup.
184 	 * The use of an all-ones broadcast address implies this; an
185 	 * interface is specified by the broadcast address of an interface,
186 	 * or the destination address of a ptp interface.
187 	 */
188 	if (flags & IP_SENDONES) {
189 		if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst)))) == NULL &&
190 		    (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL) {
191 			ipstat.ips_noroute++;
192 			error = ENETUNREACH;
193 			goto bad;
194 		}
195 		ip->ip_dst.s_addr = INADDR_BROADCAST;
196 		dst->sin_addr = ip->ip_dst;
197 		ifp = ia->ia_ifp;
198 		ip->ip_ttl = 1;
199 		isbroadcast = 1;
200 	} else if (flags & IP_ROUTETOIF) {
201 		if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
202 		    (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
203 			ipstat.ips_noroute++;
204 			error = ENETUNREACH;
205 			goto bad;
206 		}
207 		ifp = ia->ia_ifp;
208 		ip->ip_ttl = 1;
209 		isbroadcast = in_broadcast(dst->sin_addr, ifp);
210 	} else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
211 	    imo != NULL && imo->imo_multicast_ifp != NULL) {
212 		/*
213 		 * Bypass the normal routing lookup for multicast
214 		 * packets if the interface is specified.
215 		 */
216 		ifp = imo->imo_multicast_ifp;
217 		IFP_TO_IA(ifp, ia);
218 		isbroadcast = 0;	/* fool gcc */
219 	} else {
220 		/*
221 		 * We want to do any cloning requested by the link layer,
222 		 * as this is probably required in all cases for correct
223 		 * operation (as it is for ARP).
224 		 */
225 		if (ro->ro_rt == NULL)
226 			rtalloc_ign(ro, 0);
227 		if (ro->ro_rt == NULL) {
228 			ipstat.ips_noroute++;
229 			error = EHOSTUNREACH;
230 			goto bad;
231 		}
232 		ia = ifatoia(ro->ro_rt->rt_ifa);
233 		ifp = ro->ro_rt->rt_ifp;
234 		ro->ro_rt->rt_rmx.rmx_pksent++;
235 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
236 			dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
237 		if (ro->ro_rt->rt_flags & RTF_HOST)
238 			isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
239 		else
240 			isbroadcast = in_broadcast(dst->sin_addr, ifp);
241 	}
242 	/*
243 	 * Calculate MTU.  If we have a route that is up, use that,
244 	 * otherwise use the interface's MTU.
245 	 */
246 	if (ro->ro_rt != NULL && (ro->ro_rt->rt_flags & (RTF_UP|RTF_HOST))) {
247 		/*
248 		 * This case can happen if the user changed the MTU
249 		 * of an interface after enabling IP on it.  Because
250 		 * most netifs don't keep track of routes pointing to
251 		 * them, there is no way for one to update all its
252 		 * routes when the MTU is changed.
253 		 */
254 		if (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)
255 			ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
256 		mtu = ro->ro_rt->rt_rmx.rmx_mtu;
257 	} else {
258 		mtu = ifp->if_mtu;
259 	}
260 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
261 		struct in_multi *inm;
262 
263 		m->m_flags |= M_MCAST;
264 		/*
265 		 * IP destination address is multicast.  Make sure "dst"
266 		 * still points to the address in "ro".  (It may have been
267 		 * changed to point to a gateway address, above.)
268 		 */
269 		dst = (struct sockaddr_in *)&ro->ro_dst;
270 		/*
271 		 * See if the caller provided any multicast options
272 		 */
273 		if (imo != NULL) {
274 			ip->ip_ttl = imo->imo_multicast_ttl;
275 			if (imo->imo_multicast_vif != -1)
276 				ip->ip_src.s_addr =
277 				    ip_mcast_src ?
278 				    ip_mcast_src(imo->imo_multicast_vif) :
279 				    INADDR_ANY;
280 		} else
281 			ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
282 		/*
283 		 * Confirm that the outgoing interface supports multicast.
284 		 */
285 		if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
286 			if ((ifp->if_flags & IFF_MULTICAST) == 0) {
287 				ipstat.ips_noroute++;
288 				error = ENETUNREACH;
289 				goto bad;
290 			}
291 		}
292 		/*
293 		 * If source address not specified yet, use address
294 		 * of outgoing interface.
295 		 */
296 		if (ip->ip_src.s_addr == INADDR_ANY) {
297 			/* Interface may have no addresses. */
298 			if (ia != NULL)
299 				ip->ip_src = IA_SIN(ia)->sin_addr;
300 		}
301 
302 		IN_MULTI_LOCK();
303 		IN_LOOKUP_MULTI(ip->ip_dst, ifp, inm);
304 		if (inm != NULL &&
305 		   (imo == NULL || imo->imo_multicast_loop)) {
306 			IN_MULTI_UNLOCK();
307 			/*
308 			 * If we belong to the destination multicast group
309 			 * on the outgoing interface, and the caller did not
310 			 * forbid loopback, loop back a copy.
311 			 */
312 			ip_mloopback(ifp, m, dst, hlen);
313 		}
314 		else {
315 			IN_MULTI_UNLOCK();
316 			/*
317 			 * If we are acting as a multicast router, perform
318 			 * multicast forwarding as if the packet had just
319 			 * arrived on the interface to which we are about
320 			 * to send.  The multicast forwarding function
321 			 * recursively calls this function, using the
322 			 * IP_FORWARDING flag to prevent infinite recursion.
323 			 *
324 			 * Multicasts that are looped back by ip_mloopback(),
325 			 * above, will be forwarded by the ip_input() routine,
326 			 * if necessary.
327 			 */
328 			if (ip_mrouter && (flags & IP_FORWARDING) == 0) {
329 				/*
330 				 * If rsvp daemon is not running, do not
331 				 * set ip_moptions. This ensures that the packet
332 				 * is multicast and not just sent down one link
333 				 * as prescribed by rsvpd.
334 				 */
335 				if (!rsvp_on)
336 					imo = NULL;
337 				if (ip_mforward &&
338 				    ip_mforward(ip, ifp, m, imo) != 0) {
339 					m_freem(m);
340 					goto done;
341 				}
342 			}
343 		}
344 
345 		/*
346 		 * Multicasts with a time-to-live of zero may be looped-
347 		 * back, above, but must not be transmitted on a network.
348 		 * Also, multicasts addressed to the loopback interface
349 		 * are not sent -- the above call to ip_mloopback() will
350 		 * loop back a copy if this host actually belongs to the
351 		 * destination group on the loopback interface.
352 		 */
353 		if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
354 			m_freem(m);
355 			goto done;
356 		}
357 
358 		goto sendit;
359 	}
360 
361 	/*
362 	 * If the source address is not specified yet, use the address
363 	 * of the outoing interface.
364 	 */
365 	if (ip->ip_src.s_addr == INADDR_ANY) {
366 		/* Interface may have no addresses. */
367 		if (ia != NULL) {
368 			ip->ip_src = IA_SIN(ia)->sin_addr;
369 		}
370 	}
371 
372 	/*
373 	 * Verify that we have any chance at all of being able to queue the
374 	 * packet or packet fragments, unless ALTQ is enabled on the given
375 	 * interface in which case packetdrop should be done by queueing.
376 	 */
377 #ifdef ALTQ
378 	if ((!ALTQ_IS_ENABLED(&ifp->if_snd)) &&
379 	    ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >=
380 	    ifp->if_snd.ifq_maxlen))
381 #else
382 	if ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >=
383 	    ifp->if_snd.ifq_maxlen)
384 #endif /* ALTQ */
385 	{
386 		error = ENOBUFS;
387 		ipstat.ips_odropped++;
388 		ifp->if_snd.ifq_drops += (ip->ip_len / ifp->if_mtu + 1);
389 		goto bad;
390 	}
391 
392 	/*
393 	 * Look for broadcast address and
394 	 * verify user is allowed to send
395 	 * such a packet.
396 	 */
397 	if (isbroadcast) {
398 		if ((ifp->if_flags & IFF_BROADCAST) == 0) {
399 			error = EADDRNOTAVAIL;
400 			goto bad;
401 		}
402 		if ((flags & IP_ALLOWBROADCAST) == 0) {
403 			error = EACCES;
404 			goto bad;
405 		}
406 		/* don't allow broadcast messages to be fragmented */
407 		if (ip->ip_len > mtu) {
408 			error = EMSGSIZE;
409 			goto bad;
410 		}
411 		m->m_flags |= M_BCAST;
412 	} else {
413 		m->m_flags &= ~M_BCAST;
414 	}
415 
416 sendit:
417 #ifdef IPSEC
418 	switch(ip_ipsec_output(&m, inp, &flags, &error, &ro, &iproute, &dst, &ia, &ifp)) {
419 	case 1:
420 		goto bad;
421 	case -1:
422 		goto done;
423 	case 0:
424 	default:
425 		break;	/* Continue with packet processing. */
426 	}
427 	/* Update variables that are affected by ipsec4_output(). */
428 	ip = mtod(m, struct ip *);
429 	hlen = ip->ip_hl << 2;
430 #endif /* IPSEC */
431 
432 	/* Jump over all PFIL processing if hooks are not active. */
433 	if (!PFIL_HOOKED(&inet_pfil_hook))
434 		goto passout;
435 
436 	/* Run through list of hooks for output packets. */
437 	odst.s_addr = ip->ip_dst.s_addr;
438 	error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, inp);
439 	if (error != 0 || m == NULL)
440 		goto done;
441 
442 	ip = mtod(m, struct ip *);
443 
444 	/* See if destination IP address was changed by packet filter. */
445 	if (odst.s_addr != ip->ip_dst.s_addr) {
446 		m->m_flags |= M_SKIP_FIREWALL;
447 		/* If destination is now ourself drop to ip_input(). */
448 		if (in_localip(ip->ip_dst)) {
449 			m->m_flags |= M_FASTFWD_OURS;
450 			if (m->m_pkthdr.rcvif == NULL)
451 				m->m_pkthdr.rcvif = loif;
452 			if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
453 				m->m_pkthdr.csum_flags |=
454 				    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
455 				m->m_pkthdr.csum_data = 0xffff;
456 			}
457 			m->m_pkthdr.csum_flags |=
458 			    CSUM_IP_CHECKED | CSUM_IP_VALID;
459 
460 			error = netisr_queue(NETISR_IP, m);
461 			goto done;
462 		} else
463 			goto again;	/* Redo the routing table lookup. */
464 	}
465 
466 #ifdef IPFIREWALL_FORWARD
467 	/* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */
468 	if (m->m_flags & M_FASTFWD_OURS) {
469 		if (m->m_pkthdr.rcvif == NULL)
470 			m->m_pkthdr.rcvif = loif;
471 		if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
472 			m->m_pkthdr.csum_flags |=
473 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
474 			m->m_pkthdr.csum_data = 0xffff;
475 		}
476 		m->m_pkthdr.csum_flags |=
477 			    CSUM_IP_CHECKED | CSUM_IP_VALID;
478 
479 		error = netisr_queue(NETISR_IP, m);
480 		goto done;
481 	}
482 	/* Or forward to some other address? */
483 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
484 	if (fwd_tag) {
485 		dst = (struct sockaddr_in *)&ro->ro_dst;
486 		bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in));
487 		m->m_flags |= M_SKIP_FIREWALL;
488 		m_tag_delete(m, fwd_tag);
489 		goto again;
490 	}
491 #endif /* IPFIREWALL_FORWARD */
492 
493 passout:
494 	/* 127/8 must not appear on wire - RFC1122. */
495 	if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
496 	    (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
497 		if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
498 			ipstat.ips_badaddr++;
499 			error = EADDRNOTAVAIL;
500 			goto bad;
501 		}
502 	}
503 
504 	m->m_pkthdr.csum_flags |= CSUM_IP;
505 	sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
506 	if (sw_csum & CSUM_DELAY_DATA) {
507 		in_delayed_cksum(m);
508 		sw_csum &= ~CSUM_DELAY_DATA;
509 	}
510 	m->m_pkthdr.csum_flags &= ifp->if_hwassist;
511 
512 	/*
513 	 * If small enough for interface, or the interface will take
514 	 * care of the fragmentation for us, we can just send directly.
515 	 */
516 	if (ip->ip_len <= mtu ||
517 	    (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 ||
518 	    ((ip->ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) {
519 		ip->ip_len = htons(ip->ip_len);
520 		ip->ip_off = htons(ip->ip_off);
521 		ip->ip_sum = 0;
522 		if (sw_csum & CSUM_DELAY_IP)
523 			ip->ip_sum = in_cksum(m, hlen);
524 
525 		/*
526 		 * Record statistics for this interface address.
527 		 * With CSUM_TSO the byte/packet count will be slightly
528 		 * incorrect because we count the IP+TCP headers only
529 		 * once instead of for every generated packet.
530 		 */
531 		if (!(flags & IP_FORWARDING) && ia) {
532 			if (m->m_pkthdr.csum_flags & CSUM_TSO)
533 				ia->ia_ifa.if_opackets +=
534 				    m->m_pkthdr.len / m->m_pkthdr.tso_segsz;
535 			else
536 				ia->ia_ifa.if_opackets++;
537 			ia->ia_ifa.if_obytes += m->m_pkthdr.len;
538 		}
539 #ifdef MBUF_STRESS_TEST
540 		if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size)
541 			m = m_fragment(m, M_DONTWAIT, mbuf_frag_size);
542 #endif
543 		/*
544 		 * Reset layer specific mbuf flags
545 		 * to avoid confusing lower layers.
546 		 */
547 		m->m_flags &= ~(M_PROTOFLAGS);
548 
549 		error = (*ifp->if_output)(ifp, m,
550 				(struct sockaddr *)dst, ro->ro_rt);
551 		goto done;
552 	}
553 
554 	/* Balk when DF bit is set or the interface didn't support TSO. */
555 	if ((ip->ip_off & IP_DF) || (m->m_pkthdr.csum_flags & CSUM_TSO)) {
556 		error = EMSGSIZE;
557 		ipstat.ips_cantfrag++;
558 		goto bad;
559 	}
560 
561 	/*
562 	 * Too large for interface; fragment if possible. If successful,
563 	 * on return, m will point to a list of packets to be sent.
564 	 */
565 	error = ip_fragment(ip, &m, mtu, ifp->if_hwassist, sw_csum);
566 	if (error)
567 		goto bad;
568 	for (; m; m = m0) {
569 		m0 = m->m_nextpkt;
570 		m->m_nextpkt = 0;
571 		if (error == 0) {
572 			/* Record statistics for this interface address. */
573 			if (ia != NULL) {
574 				ia->ia_ifa.if_opackets++;
575 				ia->ia_ifa.if_obytes += m->m_pkthdr.len;
576 			}
577 			/*
578 			 * Reset layer specific mbuf flags
579 			 * to avoid confusing upper layers.
580 			 */
581 			m->m_flags &= ~(M_PROTOFLAGS);
582 
583 			error = (*ifp->if_output)(ifp, m,
584 			    (struct sockaddr *)dst, ro->ro_rt);
585 		} else
586 			m_freem(m);
587 	}
588 
589 	if (error == 0)
590 		ipstat.ips_fragmented++;
591 
592 done:
593 	if (ro == &iproute && ro->ro_rt) {
594 		RTFREE(ro->ro_rt);
595 	}
596 	return (error);
597 bad:
598 	m_freem(m);
599 	goto done;
600 }
601 
602 /*
603  * Create a chain of fragments which fit the given mtu. m_frag points to the
604  * mbuf to be fragmented; on return it points to the chain with the fragments.
605  * Return 0 if no error. If error, m_frag may contain a partially built
606  * chain of fragments that should be freed by the caller.
607  *
608  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
609  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
610  */
611 int
612 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
613     u_long if_hwassist_flags, int sw_csum)
614 {
615 	int error = 0;
616 	int hlen = ip->ip_hl << 2;
617 	int len = (mtu - hlen) & ~7;	/* size of payload in each fragment */
618 	int off;
619 	struct mbuf *m0 = *m_frag;	/* the original packet		*/
620 	int firstlen;
621 	struct mbuf **mnext;
622 	int nfrags;
623 
624 	if (ip->ip_off & IP_DF) {	/* Fragmentation not allowed */
625 		ipstat.ips_cantfrag++;
626 		return EMSGSIZE;
627 	}
628 
629 	/*
630 	 * Must be able to put at least 8 bytes per fragment.
631 	 */
632 	if (len < 8)
633 		return EMSGSIZE;
634 
635 	/*
636 	 * If the interface will not calculate checksums on
637 	 * fragmented packets, then do it here.
638 	 */
639 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA &&
640 	    (if_hwassist_flags & CSUM_IP_FRAGS) == 0) {
641 		in_delayed_cksum(m0);
642 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
643 	}
644 
645 	if (len > PAGE_SIZE) {
646 		/*
647 		 * Fragment large datagrams such that each segment
648 		 * contains a multiple of PAGE_SIZE amount of data,
649 		 * plus headers. This enables a receiver to perform
650 		 * page-flipping zero-copy optimizations.
651 		 *
652 		 * XXX When does this help given that sender and receiver
653 		 * could have different page sizes, and also mtu could
654 		 * be less than the receiver's page size ?
655 		 */
656 		int newlen;
657 		struct mbuf *m;
658 
659 		for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
660 			off += m->m_len;
661 
662 		/*
663 		 * firstlen (off - hlen) must be aligned on an
664 		 * 8-byte boundary
665 		 */
666 		if (off < hlen)
667 			goto smart_frag_failure;
668 		off = ((off - hlen) & ~7) + hlen;
669 		newlen = (~PAGE_MASK) & mtu;
670 		if ((newlen + sizeof (struct ip)) > mtu) {
671 			/* we failed, go back the default */
672 smart_frag_failure:
673 			newlen = len;
674 			off = hlen + len;
675 		}
676 		len = newlen;
677 
678 	} else {
679 		off = hlen + len;
680 	}
681 
682 	firstlen = off - hlen;
683 	mnext = &m0->m_nextpkt;		/* pointer to next packet */
684 
685 	/*
686 	 * Loop through length of segment after first fragment,
687 	 * make new header and copy data of each part and link onto chain.
688 	 * Here, m0 is the original packet, m is the fragment being created.
689 	 * The fragments are linked off the m_nextpkt of the original
690 	 * packet, which after processing serves as the first fragment.
691 	 */
692 	for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
693 		struct ip *mhip;	/* ip header on the fragment */
694 		struct mbuf *m;
695 		int mhlen = sizeof (struct ip);
696 
697 		MGETHDR(m, M_DONTWAIT, MT_DATA);
698 		if (m == NULL) {
699 			error = ENOBUFS;
700 			ipstat.ips_odropped++;
701 			goto done;
702 		}
703 		m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
704 		/*
705 		 * In the first mbuf, leave room for the link header, then
706 		 * copy the original IP header including options. The payload
707 		 * goes into an additional mbuf chain returned by m_copy().
708 		 */
709 		m->m_data += max_linkhdr;
710 		mhip = mtod(m, struct ip *);
711 		*mhip = *ip;
712 		if (hlen > sizeof (struct ip)) {
713 			mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
714 			mhip->ip_v = IPVERSION;
715 			mhip->ip_hl = mhlen >> 2;
716 		}
717 		m->m_len = mhlen;
718 		/* XXX do we need to add ip->ip_off below ? */
719 		mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
720 		if (off + len >= ip->ip_len) {	/* last fragment */
721 			len = ip->ip_len - off;
722 			m->m_flags |= M_LASTFRAG;
723 		} else
724 			mhip->ip_off |= IP_MF;
725 		mhip->ip_len = htons((u_short)(len + mhlen));
726 		m->m_next = m_copy(m0, off, len);
727 		if (m->m_next == NULL) {	/* copy failed */
728 			m_free(m);
729 			error = ENOBUFS;	/* ??? */
730 			ipstat.ips_odropped++;
731 			goto done;
732 		}
733 		m->m_pkthdr.len = mhlen + len;
734 		m->m_pkthdr.rcvif = NULL;
735 #ifdef MAC
736 		mac_netinet_fragment(m0, m);
737 #endif
738 		m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
739 		mhip->ip_off = htons(mhip->ip_off);
740 		mhip->ip_sum = 0;
741 		if (sw_csum & CSUM_DELAY_IP)
742 			mhip->ip_sum = in_cksum(m, mhlen);
743 		*mnext = m;
744 		mnext = &m->m_nextpkt;
745 	}
746 	ipstat.ips_ofragments += nfrags;
747 
748 	/* set first marker for fragment chain */
749 	m0->m_flags |= M_FIRSTFRAG | M_FRAG;
750 	m0->m_pkthdr.csum_data = nfrags;
751 
752 	/*
753 	 * Update first fragment by trimming what's been copied out
754 	 * and updating header.
755 	 */
756 	m_adj(m0, hlen + firstlen - ip->ip_len);
757 	m0->m_pkthdr.len = hlen + firstlen;
758 	ip->ip_len = htons((u_short)m0->m_pkthdr.len);
759 	ip->ip_off |= IP_MF;
760 	ip->ip_off = htons(ip->ip_off);
761 	ip->ip_sum = 0;
762 	if (sw_csum & CSUM_DELAY_IP)
763 		ip->ip_sum = in_cksum(m0, hlen);
764 
765 done:
766 	*m_frag = m0;
767 	return error;
768 }
769 
770 void
771 in_delayed_cksum(struct mbuf *m)
772 {
773 	struct ip *ip;
774 	u_short csum, offset;
775 
776 	ip = mtod(m, struct ip *);
777 	offset = ip->ip_hl << 2 ;
778 	csum = in_cksum_skip(m, ip->ip_len, offset);
779 	if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
780 		csum = 0xffff;
781 	offset += m->m_pkthdr.csum_data;	/* checksum offset */
782 
783 	if (offset + sizeof(u_short) > m->m_len) {
784 		printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
785 		    m->m_len, offset, ip->ip_p);
786 		/*
787 		 * XXX
788 		 * this shouldn't happen, but if it does, the
789 		 * correct behavior may be to insert the checksum
790 		 * in the appropriate next mbuf in the chain.
791 		 */
792 		return;
793 	}
794 	*(u_short *)(m->m_data + offset) = csum;
795 }
796 
797 /*
798  * IP socket option processing.
799  */
800 int
801 ip_ctloutput(struct socket *so, struct sockopt *sopt)
802 {
803 	struct	inpcb *inp = sotoinpcb(so);
804 	int	error, optval;
805 
806 	error = optval = 0;
807 	if (sopt->sopt_level != IPPROTO_IP) {
808 		return (EINVAL);
809 	}
810 
811 	switch (sopt->sopt_dir) {
812 	case SOPT_SET:
813 		switch (sopt->sopt_name) {
814 		case IP_OPTIONS:
815 #ifdef notyet
816 		case IP_RETOPTS:
817 #endif
818 		{
819 			struct mbuf *m;
820 			if (sopt->sopt_valsize > MLEN) {
821 				error = EMSGSIZE;
822 				break;
823 			}
824 			MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
825 			if (m == NULL) {
826 				error = ENOBUFS;
827 				break;
828 			}
829 			m->m_len = sopt->sopt_valsize;
830 			error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
831 					    m->m_len);
832 			if (error) {
833 				m_free(m);
834 				break;
835 			}
836 			INP_LOCK(inp);
837 			error = ip_pcbopts(inp, sopt->sopt_name, m);
838 			INP_UNLOCK(inp);
839 			return (error);
840 		}
841 
842 		case IP_TOS:
843 		case IP_TTL:
844 		case IP_MINTTL:
845 		case IP_RECVOPTS:
846 		case IP_RECVRETOPTS:
847 		case IP_RECVDSTADDR:
848 		case IP_RECVTTL:
849 		case IP_RECVIF:
850 		case IP_FAITH:
851 		case IP_ONESBCAST:
852 		case IP_DONTFRAG:
853 			error = sooptcopyin(sopt, &optval, sizeof optval,
854 					    sizeof optval);
855 			if (error)
856 				break;
857 
858 			switch (sopt->sopt_name) {
859 			case IP_TOS:
860 				inp->inp_ip_tos = optval;
861 				break;
862 
863 			case IP_TTL:
864 				inp->inp_ip_ttl = optval;
865 				break;
866 
867 			case IP_MINTTL:
868 				if (optval > 0 && optval <= MAXTTL)
869 					inp->inp_ip_minttl = optval;
870 				else
871 					error = EINVAL;
872 				break;
873 
874 #define	OPTSET(bit) do {						\
875 	INP_LOCK(inp);							\
876 	if (optval)							\
877 		inp->inp_flags |= bit;					\
878 	else								\
879 		inp->inp_flags &= ~bit;					\
880 	INP_UNLOCK(inp);						\
881 } while (0)
882 
883 			case IP_RECVOPTS:
884 				OPTSET(INP_RECVOPTS);
885 				break;
886 
887 			case IP_RECVRETOPTS:
888 				OPTSET(INP_RECVRETOPTS);
889 				break;
890 
891 			case IP_RECVDSTADDR:
892 				OPTSET(INP_RECVDSTADDR);
893 				break;
894 
895 			case IP_RECVTTL:
896 				OPTSET(INP_RECVTTL);
897 				break;
898 
899 			case IP_RECVIF:
900 				OPTSET(INP_RECVIF);
901 				break;
902 
903 			case IP_FAITH:
904 				OPTSET(INP_FAITH);
905 				break;
906 
907 			case IP_ONESBCAST:
908 				OPTSET(INP_ONESBCAST);
909 				break;
910 			case IP_DONTFRAG:
911 				OPTSET(INP_DONTFRAG);
912 				break;
913 			}
914 			break;
915 #undef OPTSET
916 
917 		/*
918 		 * Multicast socket options are processed by the in_mcast
919 		 * module.
920 		 */
921 		case IP_MULTICAST_IF:
922 		case IP_MULTICAST_VIF:
923 		case IP_MULTICAST_TTL:
924 		case IP_MULTICAST_LOOP:
925 		case IP_ADD_MEMBERSHIP:
926 		case IP_DROP_MEMBERSHIP:
927 		case IP_ADD_SOURCE_MEMBERSHIP:
928 		case IP_DROP_SOURCE_MEMBERSHIP:
929 		case IP_BLOCK_SOURCE:
930 		case IP_UNBLOCK_SOURCE:
931 		case IP_MSFILTER:
932 		case MCAST_JOIN_GROUP:
933 		case MCAST_LEAVE_GROUP:
934 		case MCAST_JOIN_SOURCE_GROUP:
935 		case MCAST_LEAVE_SOURCE_GROUP:
936 		case MCAST_BLOCK_SOURCE:
937 		case MCAST_UNBLOCK_SOURCE:
938 			error = inp_setmoptions(inp, sopt);
939 			break;
940 
941 		case IP_PORTRANGE:
942 			error = sooptcopyin(sopt, &optval, sizeof optval,
943 					    sizeof optval);
944 			if (error)
945 				break;
946 
947 			INP_LOCK(inp);
948 			switch (optval) {
949 			case IP_PORTRANGE_DEFAULT:
950 				inp->inp_flags &= ~(INP_LOWPORT);
951 				inp->inp_flags &= ~(INP_HIGHPORT);
952 				break;
953 
954 			case IP_PORTRANGE_HIGH:
955 				inp->inp_flags &= ~(INP_LOWPORT);
956 				inp->inp_flags |= INP_HIGHPORT;
957 				break;
958 
959 			case IP_PORTRANGE_LOW:
960 				inp->inp_flags &= ~(INP_HIGHPORT);
961 				inp->inp_flags |= INP_LOWPORT;
962 				break;
963 
964 			default:
965 				error = EINVAL;
966 				break;
967 			}
968 			INP_UNLOCK(inp);
969 			break;
970 
971 #ifdef IPSEC
972 		case IP_IPSEC_POLICY:
973 		{
974 			caddr_t req;
975 			size_t len = 0;
976 			int priv;
977 			struct mbuf *m;
978 			int optname;
979 
980 			if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
981 				break;
982 			if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
983 				break;
984 			if (sopt->sopt_td != NULL) {
985 				/*
986 				 * XXXRW: Would be more desirable to do this
987 				 * one layer down so that we only exercise
988 				 * privilege if it is needed.
989 				 */
990 				error = priv_check(sopt->sopt_td,
991 				    PRIV_NETINET_IPSEC);
992 				if (error)
993 					priv = 0;
994 				else
995 					priv = 1;
996 			} else
997 				priv = 1;
998 			req = mtod(m, caddr_t);
999 			len = m->m_len;
1000 			optname = sopt->sopt_name;
1001 			error = ipsec4_set_policy(inp, optname, req, len, priv);
1002 			m_freem(m);
1003 			break;
1004 		}
1005 #endif /* IPSEC */
1006 
1007 		default:
1008 			error = ENOPROTOOPT;
1009 			break;
1010 		}
1011 		break;
1012 
1013 	case SOPT_GET:
1014 		switch (sopt->sopt_name) {
1015 		case IP_OPTIONS:
1016 		case IP_RETOPTS:
1017 			if (inp->inp_options)
1018 				error = sooptcopyout(sopt,
1019 						     mtod(inp->inp_options,
1020 							  char *),
1021 						     inp->inp_options->m_len);
1022 			else
1023 				sopt->sopt_valsize = 0;
1024 			break;
1025 
1026 		case IP_TOS:
1027 		case IP_TTL:
1028 		case IP_MINTTL:
1029 		case IP_RECVOPTS:
1030 		case IP_RECVRETOPTS:
1031 		case IP_RECVDSTADDR:
1032 		case IP_RECVTTL:
1033 		case IP_RECVIF:
1034 		case IP_PORTRANGE:
1035 		case IP_FAITH:
1036 		case IP_ONESBCAST:
1037 		case IP_DONTFRAG:
1038 			switch (sopt->sopt_name) {
1039 
1040 			case IP_TOS:
1041 				optval = inp->inp_ip_tos;
1042 				break;
1043 
1044 			case IP_TTL:
1045 				optval = inp->inp_ip_ttl;
1046 				break;
1047 
1048 			case IP_MINTTL:
1049 				optval = inp->inp_ip_minttl;
1050 				break;
1051 
1052 #define	OPTBIT(bit)	(inp->inp_flags & bit ? 1 : 0)
1053 
1054 			case IP_RECVOPTS:
1055 				optval = OPTBIT(INP_RECVOPTS);
1056 				break;
1057 
1058 			case IP_RECVRETOPTS:
1059 				optval = OPTBIT(INP_RECVRETOPTS);
1060 				break;
1061 
1062 			case IP_RECVDSTADDR:
1063 				optval = OPTBIT(INP_RECVDSTADDR);
1064 				break;
1065 
1066 			case IP_RECVTTL:
1067 				optval = OPTBIT(INP_RECVTTL);
1068 				break;
1069 
1070 			case IP_RECVIF:
1071 				optval = OPTBIT(INP_RECVIF);
1072 				break;
1073 
1074 			case IP_PORTRANGE:
1075 				if (inp->inp_flags & INP_HIGHPORT)
1076 					optval = IP_PORTRANGE_HIGH;
1077 				else if (inp->inp_flags & INP_LOWPORT)
1078 					optval = IP_PORTRANGE_LOW;
1079 				else
1080 					optval = 0;
1081 				break;
1082 
1083 			case IP_FAITH:
1084 				optval = OPTBIT(INP_FAITH);
1085 				break;
1086 
1087 			case IP_ONESBCAST:
1088 				optval = OPTBIT(INP_ONESBCAST);
1089 				break;
1090 			case IP_DONTFRAG:
1091 				optval = OPTBIT(INP_DONTFRAG);
1092 				break;
1093 			}
1094 			error = sooptcopyout(sopt, &optval, sizeof optval);
1095 			break;
1096 
1097 		/*
1098 		 * Multicast socket options are processed by the in_mcast
1099 		 * module.
1100 		 */
1101 		case IP_MULTICAST_IF:
1102 		case IP_MULTICAST_VIF:
1103 		case IP_MULTICAST_TTL:
1104 		case IP_MULTICAST_LOOP:
1105 		case IP_MSFILTER:
1106 			error = inp_getmoptions(inp, sopt);
1107 			break;
1108 
1109 #ifdef IPSEC
1110 		case IP_IPSEC_POLICY:
1111 		{
1112 			struct mbuf *m = NULL;
1113 			caddr_t req = NULL;
1114 			size_t len = 0;
1115 
1116 			if (m != 0) {
1117 				req = mtod(m, caddr_t);
1118 				len = m->m_len;
1119 			}
1120 			error = ipsec4_get_policy(sotoinpcb(so), req, len, &m);
1121 			if (error == 0)
1122 				error = soopt_mcopyout(sopt, m); /* XXX */
1123 			if (error == 0)
1124 				m_freem(m);
1125 			break;
1126 		}
1127 #endif /* IPSEC */
1128 
1129 		default:
1130 			error = ENOPROTOOPT;
1131 			break;
1132 		}
1133 		break;
1134 	}
1135 	return (error);
1136 }
1137 
1138 /*
1139  * Routine called from ip_output() to loop back a copy of an IP multicast
1140  * packet to the input queue of a specified interface.  Note that this
1141  * calls the output routine of the loopback "driver", but with an interface
1142  * pointer that might NOT be a loopback interface -- evil, but easier than
1143  * replicating that code here.
1144  */
1145 static void
1146 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
1147     int hlen)
1148 {
1149 	register struct ip *ip;
1150 	struct mbuf *copym;
1151 
1152 	copym = m_copy(m, 0, M_COPYALL);
1153 	if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
1154 		copym = m_pullup(copym, hlen);
1155 	if (copym != NULL) {
1156 		/* If needed, compute the checksum and mark it as valid. */
1157 		if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
1158 			in_delayed_cksum(copym);
1159 			copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1160 			copym->m_pkthdr.csum_flags |=
1161 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1162 			copym->m_pkthdr.csum_data = 0xffff;
1163 		}
1164 		/*
1165 		 * We don't bother to fragment if the IP length is greater
1166 		 * than the interface's MTU.  Can this possibly matter?
1167 		 */
1168 		ip = mtod(copym, struct ip *);
1169 		ip->ip_len = htons(ip->ip_len);
1170 		ip->ip_off = htons(ip->ip_off);
1171 		ip->ip_sum = 0;
1172 		ip->ip_sum = in_cksum(copym, hlen);
1173 		/*
1174 		 * NB:
1175 		 * It's not clear whether there are any lingering
1176 		 * reentrancy problems in other areas which might
1177 		 * be exposed by using ip_input directly (in
1178 		 * particular, everything which modifies the packet
1179 		 * in-place).  Yet another option is using the
1180 		 * protosw directly to deliver the looped back
1181 		 * packet.  For the moment, we'll err on the side
1182 		 * of safety by using if_simloop().
1183 		 */
1184 #if 1 /* XXX */
1185 		if (dst->sin_family != AF_INET) {
1186 			printf("ip_mloopback: bad address family %d\n",
1187 						dst->sin_family);
1188 			dst->sin_family = AF_INET;
1189 		}
1190 #endif
1191 
1192 #ifdef notdef
1193 		copym->m_pkthdr.rcvif = ifp;
1194 		ip_input(copym);
1195 #else
1196 		if_simloop(ifp, copym, dst->sin_family, 0);
1197 #endif
1198 	}
1199 }
1200