frag6.c revision 121345
162587Sitojun/*	$FreeBSD: head/sys/netinet6/frag6.c 121345 2003-10-22 15:29:42Z ume $	*/
295023Ssuz/*	$KAME: frag6.c,v 1.33 2002/01/07 11:34:48 kjc Exp $	*/
362587Sitojun
453541Sshin/*
553541Sshin * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
653541Sshin * All rights reserved.
753541Sshin *
853541Sshin * Redistribution and use in source and binary forms, with or without
953541Sshin * modification, are permitted provided that the following conditions
1053541Sshin * are met:
1153541Sshin * 1. Redistributions of source code must retain the above copyright
1253541Sshin *    notice, this list of conditions and the following disclaimer.
1353541Sshin * 2. Redistributions in binary form must reproduce the above copyright
1453541Sshin *    notice, this list of conditions and the following disclaimer in the
1553541Sshin *    documentation and/or other materials provided with the distribution.
1653541Sshin * 3. Neither the name of the project nor the names of its contributors
1753541Sshin *    may be used to endorse or promote products derived from this software
1853541Sshin *    without specific prior written permission.
1953541Sshin *
2053541Sshin * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
2153541Sshin * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
2253541Sshin * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
2353541Sshin * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
2453541Sshin * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
2553541Sshin * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
2653541Sshin * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
2753541Sshin * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
2853541Sshin * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
2953541Sshin * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
3053541Sshin * SUCH DAMAGE.
3153541Sshin */
3253541Sshin
33120645Sume#include "opt_random_ip_id.h"
34120645Sume
3553541Sshin#include <sys/param.h>
3653541Sshin#include <sys/systm.h>
3753541Sshin#include <sys/malloc.h>
3853541Sshin#include <sys/mbuf.h>
3953541Sshin#include <sys/domain.h>
4053541Sshin#include <sys/protosw.h>
4153541Sshin#include <sys/socket.h>
4253541Sshin#include <sys/errno.h>
4353541Sshin#include <sys/time.h>
4453541Sshin#include <sys/kernel.h>
4553541Sshin#include <sys/syslog.h>
4653541Sshin
4753541Sshin#include <net/if.h>
4853541Sshin#include <net/route.h>
4953541Sshin
5053541Sshin#include <netinet/in.h>
5153541Sshin#include <netinet/in_var.h>
5262587Sitojun#include <netinet/ip6.h>
5353541Sshin#include <netinet6/ip6_var.h>
5462587Sitojun#include <netinet/icmp6.h>
5553541Sshin
5653541Sshin#include <net/net_osdep.h>
5753541Sshin
5853541Sshin/*
5953541Sshin * Define it to get a correct behavior on per-interface statistics.
6053541Sshin * You will need to perform an extra routing table lookup, per fragment,
6153541Sshin * to do it.  This may, or may not be, a performance hit.
6253541Sshin */
6362587Sitojun#define IN6_IFSTAT_STRICT
6453541Sshin
6562587Sitojunstatic void frag6_enq __P((struct ip6asfrag *, struct ip6asfrag *));
6662587Sitojunstatic void frag6_deq __P((struct ip6asfrag *));
6762587Sitojunstatic void frag6_insque __P((struct ip6q *, struct ip6q *));
6862587Sitojunstatic void frag6_remque __P((struct ip6q *));
6962587Sitojunstatic void frag6_freef __P((struct ip6q *));
7053541Sshin
71121345Sumestatic int ip6q_locked;
7262587Sitojunu_int frag6_nfragpackets;
73121345Sumeu_int frag6_nfrags;
7462587Sitojunstruct	ip6q ip6q;	/* ip6 reassemble queue */
7553541Sshin
76121345Sumestatic __inline int ip6q_lock_try __P((void));
77121345Sumestatic __inline void ip6q_unlock __P((void));
78121345Sume
79121345Sumestatic __inline int
80121345Sumeip6q_lock_try()
81121345Sume{
82121345Sume	if (ip6q_locked)
83121345Sume		return (0);
84121345Sume	ip6q_locked = 1;
85121345Sume	return (1);
86121345Sume}
87121345Sume
88121345Sumestatic __inline void
89121345Sumeip6q_unlock()
90121345Sume{
91121345Sume	ip6q_locked = 0;
92121345Sume}
93121345Sume
94121345Sume#ifdef DIAGNOSTIC
95121345Sume#define	IP6Q_LOCK()							\
96121345Sumedo {									\
97121345Sume	if (ip6q_lock_try() == 0) {					\
98121345Sume		printf("%s:%d: ip6q already locked\n", __FILE__, __LINE__); \
99121345Sume		panic("ip6q_lock");					\
100121345Sume	}								\
101121345Sume} while (/*CONSTCOND*/ 0)
102121345Sume#define	IP6Q_LOCK_CHECK()						\
103121345Sumedo {									\
104121345Sume	if (ip6q_locked == 0) {						\
105121345Sume		printf("%s:%d: ip6q lock not held\n", __FILE__, __LINE__); \
106121345Sume		panic("ip6q lock check");				\
107121345Sume	}								\
108121345Sume} while (/*CONSTCOND*/ 0)
109121345Sume#else
110121345Sume#define	IP6Q_LOCK()		(void) ip6q_lock_try()
111121345Sume#define	IP6Q_LOCK_CHECK()	/* nothing */
112121345Sume#endif
113121345Sume
114121345Sume#define	IP6Q_UNLOCK()		ip6q_unlock()
115121345Sume
11669774Sphkstatic MALLOC_DEFINE(M_FTABLE, "fragment", "fragment reassembly header");
11762587Sitojun
11853541Sshin/*
11953541Sshin * Initialise reassembly queue and fragment identifier.
12053541Sshin */
12153541Sshinvoid
12253541Sshinfrag6_init()
12353541Sshin{
12453541Sshin
12577969Sjesper	ip6_maxfragpackets = nmbclusters / 4;
126121345Sume	ip6_maxfrags = nmbclusters / 4;
12777969Sjesper
128120645Sume#ifndef RANDOM_IP_ID
129120648Sume	ip6_id = arc4random();
130120643Sume#endif
13153541Sshin	ip6q.ip6q_next = ip6q.ip6q_prev = &ip6q;
13253541Sshin}
13353541Sshin
13453541Sshin/*
13562587Sitojun * In RFC2460, fragment and reassembly rule do not agree with each other,
13662587Sitojun * in terms of next header field handling in fragment header.
13762587Sitojun * While the sender will use the same value for all of the fragmented packets,
13862587Sitojun * receiver is suggested not to check the consistency.
13962587Sitojun *
14062587Sitojun * fragment rule (p20):
14162587Sitojun *	(2) A Fragment header containing:
14262587Sitojun *	The Next Header value that identifies the first header of
14362587Sitojun *	the Fragmentable Part of the original packet.
14462587Sitojun *		-> next header field is same for all fragments
14562587Sitojun *
14662587Sitojun * reassembly rule (p21):
14762587Sitojun *	The Next Header field of the last header of the Unfragmentable
14862587Sitojun *	Part is obtained from the Next Header field of the first
14962587Sitojun *	fragment's Fragment header.
15062587Sitojun *		-> should grab it from the first fragment only
15162587Sitojun *
15262587Sitojun * The following note also contradicts with fragment rule - noone is going to
15362587Sitojun * send different fragment with different next header field.
15462587Sitojun *
15562587Sitojun * additional note (p22):
15662587Sitojun *	The Next Header values in the Fragment headers of different
15762587Sitojun *	fragments of the same original packet may differ.  Only the value
15862587Sitojun *	from the Offset zero fragment packet is used for reassembly.
15962587Sitojun *		-> should grab it from the first fragment only
16062587Sitojun *
16162587Sitojun * There is no explicit reason given in the RFC.  Historical reason maybe?
16262587Sitojun */
16362587Sitojun/*
16453541Sshin * Fragment input
16553541Sshin */
16653541Sshinint
16753541Sshinfrag6_input(mp, offp, proto)
16853541Sshin	struct mbuf **mp;
16953541Sshin	int *offp, proto;
17053541Sshin{
17153541Sshin	struct mbuf *m = *mp, *t;
17253541Sshin	struct ip6_hdr *ip6;
17353541Sshin	struct ip6_frag *ip6f;
17453541Sshin	struct ip6q *q6;
17562587Sitojun	struct ip6asfrag *af6, *ip6af, *af6dwn;
17653541Sshin	int offset = *offp, nxt, i, next;
17753541Sshin	int first_frag = 0;
17862587Sitojun	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
17953541Sshin	struct ifnet *dstifp;
18053541Sshin#ifdef IN6_IFSTAT_STRICT
18153541Sshin	static struct route_in6 ro;
18253541Sshin	struct sockaddr_in6 *dst;
18353541Sshin#endif
18453541Sshin
18562587Sitojun	ip6 = mtod(m, struct ip6_hdr *);
18662587Sitojun#ifndef PULLDOWN_TEST
18753541Sshin	IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE);
18853541Sshin	ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset);
18962587Sitojun#else
19062587Sitojun	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
19162587Sitojun	if (ip6f == NULL)
192120856Sume		return (IPPROTO_DONE);
19362587Sitojun#endif
19453541Sshin
19553541Sshin	dstifp = NULL;
19653541Sshin#ifdef IN6_IFSTAT_STRICT
19753541Sshin	/* find the destination interface of the packet. */
19853541Sshin	dst = (struct sockaddr_in6 *)&ro.ro_dst;
19953541Sshin	if (ro.ro_rt
20053541Sshin	 && ((ro.ro_rt->rt_flags & RTF_UP) == 0
20153541Sshin	  || !IN6_ARE_ADDR_EQUAL(&dst->sin6_addr, &ip6->ip6_dst))) {
20253541Sshin		RTFREE(ro.ro_rt);
20353541Sshin		ro.ro_rt = (struct rtentry *)0;
20453541Sshin	}
20553541Sshin	if (ro.ro_rt == NULL) {
20653541Sshin		bzero(dst, sizeof(*dst));
20753541Sshin		dst->sin6_family = AF_INET6;
20853541Sshin		dst->sin6_len = sizeof(struct sockaddr_in6);
20953541Sshin		dst->sin6_addr = ip6->ip6_dst;
21053541Sshin	}
21154350Sshin	rtalloc((struct route *)&ro);
21253541Sshin	if (ro.ro_rt != NULL && ro.ro_rt->rt_ifa != NULL)
21353541Sshin		dstifp = ((struct in6_ifaddr *)ro.ro_rt->rt_ifa)->ia_ifp;
21453541Sshin#else
21553541Sshin	/* we are violating the spec, this is not the destination interface */
21653541Sshin	if ((m->m_flags & M_PKTHDR) != 0)
21753541Sshin		dstifp = m->m_pkthdr.rcvif;
21853541Sshin#endif
21953541Sshin
22053541Sshin	/* jumbo payload can't contain a fragment header */
22153541Sshin	if (ip6->ip6_plen == 0) {
22253541Sshin		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
22353541Sshin		in6_ifstat_inc(dstifp, ifs6_reass_fail);
22453541Sshin		return IPPROTO_DONE;
22553541Sshin	}
22653541Sshin
22753541Sshin	/*
22853541Sshin	 * check whether fragment packet's fragment length is
22953541Sshin	 * multiple of 8 octets.
23053541Sshin	 * sizeof(struct ip6_frag) == 8
23153541Sshin	 * sizeof(struct ip6_hdr) = 40
23253541Sshin	 */
23353541Sshin	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
23453541Sshin	    (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
235120891Sume		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
236120891Sume		    offsetof(struct ip6_hdr, ip6_plen));
23753541Sshin		in6_ifstat_inc(dstifp, ifs6_reass_fail);
23853541Sshin		return IPPROTO_DONE;
23953541Sshin	}
24053541Sshin
24153541Sshin	ip6stat.ip6s_fragments++;
24253541Sshin	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
243120891Sume
24462587Sitojun	/* offset now points to data portion */
24553541Sshin	offset += sizeof(struct ip6_frag);
24653541Sshin
247121345Sume	IP6Q_LOCK();
24878064Sume
249121345Sume	/*
250121345Sume	 * Enforce upper bound on number of fragments.
251121345Sume	 * If maxfrag is 0, never accept fragments.
252121345Sume	 * If maxfrag is -1, accept all fragments without limitation.
253121345Sume	 */
254121345Sume	if (ip6_maxfrags < 0)
255121345Sume		;
256121345Sume	else if (frag6_nfrags >= (u_int)ip6_maxfrags)
257121345Sume		goto dropfrag;
258121345Sume
25953541Sshin	for (q6 = ip6q.ip6q_next; q6 != &ip6q; q6 = q6->ip6q_next)
26053541Sshin		if (ip6f->ip6f_ident == q6->ip6q_ident &&
26153541Sshin		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
26253541Sshin		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst))
26353541Sshin			break;
26453541Sshin
26553541Sshin	if (q6 == &ip6q) {
26653541Sshin		/*
26753541Sshin		 * the first fragment to arrive, create a reassembly queue.
26853541Sshin		 */
26953541Sshin		first_frag = 1;
27053541Sshin
27153541Sshin		/*
27253541Sshin		 * Enforce upper bound on number of fragmented packets
27353541Sshin		 * for which we attempt reassembly;
274121345Sume		 * If maxfragpackets is 0, never accept fragments.
275121345Sume		 * If maxfragpackets is -1, accept all fragments without
276121345Sume		 * limitation.
27753541Sshin		 */
27878064Sume		if (ip6_maxfragpackets < 0)
27978064Sume			;
28078064Sume		else if (frag6_nfragpackets >= (u_int)ip6_maxfragpackets)
28178064Sume			goto dropfrag;
28278064Sume		frag6_nfragpackets++;
28353541Sshin		q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FTABLE,
284120891Sume		    M_DONTWAIT);
28553541Sshin		if (q6 == NULL)
28653541Sshin			goto dropfrag;
28762587Sitojun		bzero(q6, sizeof(*q6));
28853541Sshin
28953541Sshin		frag6_insque(q6, &ip6q);
29053541Sshin
29162587Sitojun		/* ip6q_nxt will be filled afterwards, from 1st fragment */
29253541Sshin		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
29362587Sitojun#ifdef notyet
29462587Sitojun		q6->ip6q_nxtp	= (u_char *)nxtp;
29562587Sitojun#endif
29653541Sshin		q6->ip6q_ident	= ip6f->ip6f_ident;
29753541Sshin		q6->ip6q_arrive = 0; /* Is it used anywhere? */
29853541Sshin		q6->ip6q_ttl 	= IPV6_FRAGTTL;
29953541Sshin		q6->ip6q_src	= ip6->ip6_src;
30053541Sshin		q6->ip6q_dst	= ip6->ip6_dst;
30153541Sshin		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
302121345Sume
303121345Sume		q6->ip6q_nfrag = 0;
30453541Sshin	}
30553541Sshin
30653541Sshin	/*
30753541Sshin	 * If it's the 1st fragment, record the length of the
30853541Sshin	 * unfragmentable part and the next header of the fragment header.
30953541Sshin	 */
31053541Sshin	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
31153541Sshin	if (fragoff == 0) {
312120891Sume		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
313120891Sume		    sizeof(struct ip6_frag);
31453541Sshin		q6->ip6q_nxt = ip6f->ip6f_nxt;
31553541Sshin	}
31653541Sshin
31753541Sshin	/*
31853541Sshin	 * Check that the reassembled packet would not exceed 65535 bytes
31953541Sshin	 * in size.
32053541Sshin	 * If it would exceed, discard the fragment and return an ICMP error.
32153541Sshin	 */
32262587Sitojun	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
32353541Sshin	if (q6->ip6q_unfrglen >= 0) {
32453541Sshin		/* The 1st fragment has already arrived. */
32553541Sshin		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
32653541Sshin			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
327120891Sume			    offset - sizeof(struct ip6_frag) +
328120891Sume			    offsetof(struct ip6_frag, ip6f_offlg));
329121345Sume			IP6Q_UNLOCK();
330120856Sume			return (IPPROTO_DONE);
33153541Sshin		}
332120891Sume	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
33353541Sshin		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
334120891Sume		    offset - sizeof(struct ip6_frag) +
335120891Sume		    offsetof(struct ip6_frag, ip6f_offlg));
336121345Sume		IP6Q_UNLOCK();
337120856Sume		return (IPPROTO_DONE);
33853541Sshin	}
33953541Sshin	/*
34053541Sshin	 * If it's the first fragment, do the above check for each
34153541Sshin	 * fragment already stored in the reassembly queue.
34253541Sshin	 */
34353541Sshin	if (fragoff == 0) {
34453541Sshin		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
34553541Sshin		     af6 = af6dwn) {
34653541Sshin			af6dwn = af6->ip6af_down;
34753541Sshin
34853541Sshin			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
34953541Sshin			    IPV6_MAXPACKET) {
35053541Sshin				struct mbuf *merr = IP6_REASS_MBUF(af6);
35153541Sshin				struct ip6_hdr *ip6err;
35253541Sshin				int erroff = af6->ip6af_offset;
35353541Sshin
35453541Sshin				/* dequeue the fragment. */
35553541Sshin				frag6_deq(af6);
35662587Sitojun				free(af6, M_FTABLE);
35753541Sshin
35853541Sshin				/* adjust pointer. */
35953541Sshin				ip6err = mtod(merr, struct ip6_hdr *);
36053541Sshin
36153541Sshin				/*
36253541Sshin				 * Restore source and destination addresses
36353541Sshin				 * in the erroneous IPv6 header.
36453541Sshin				 */
36553541Sshin				ip6err->ip6_src = q6->ip6q_src;
36653541Sshin				ip6err->ip6_dst = q6->ip6q_dst;
36753541Sshin
36853541Sshin				icmp6_error(merr, ICMP6_PARAM_PROB,
369120891Sume				    ICMP6_PARAMPROB_HEADER,
370120891Sume				    erroff - sizeof(struct ip6_frag) +
371120891Sume				    offsetof(struct ip6_frag, ip6f_offlg));
37253541Sshin			}
37353541Sshin		}
37453541Sshin	}
37553541Sshin
37662587Sitojun	ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FTABLE,
377111119Simp	    M_DONTWAIT);
37862587Sitojun	if (ip6af == NULL)
37962587Sitojun		goto dropfrag;
38062587Sitojun	bzero(ip6af, sizeof(*ip6af));
38162587Sitojun	ip6af->ip6af_head = ip6->ip6_flow;
38262587Sitojun	ip6af->ip6af_len = ip6->ip6_plen;
38362587Sitojun	ip6af->ip6af_nxt = ip6->ip6_nxt;
38462587Sitojun	ip6af->ip6af_hlim = ip6->ip6_hlim;
38553541Sshin	ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
38653541Sshin	ip6af->ip6af_off = fragoff;
38753541Sshin	ip6af->ip6af_frglen = frgpartlen;
38853541Sshin	ip6af->ip6af_offset = offset;
38953541Sshin	IP6_REASS_MBUF(ip6af) = m;
39053541Sshin
39153541Sshin	if (first_frag) {
39253541Sshin		af6 = (struct ip6asfrag *)q6;
39353541Sshin		goto insert;
39453541Sshin	}
39553541Sshin
39653541Sshin	/*
39753541Sshin	 * Find a segment which begins after this one does.
39853541Sshin	 */
39953541Sshin	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
40053541Sshin	     af6 = af6->ip6af_down)
40153541Sshin		if (af6->ip6af_off > ip6af->ip6af_off)
40253541Sshin			break;
40353541Sshin
40462587Sitojun#if 0
40553541Sshin	/*
40662587Sitojun	 * If there is a preceding segment, it may provide some of
40762587Sitojun	 * our data already.  If so, drop the data from the incoming
40862587Sitojun	 * segment.  If it provides all of our data, drop us.
40962587Sitojun	 */
41062587Sitojun	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
41162587Sitojun		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
41262587Sitojun			- ip6af->ip6af_off;
41362587Sitojun		if (i > 0) {
41462587Sitojun			if (i >= ip6af->ip6af_frglen)
41562587Sitojun				goto dropfrag;
41662587Sitojun			m_adj(IP6_REASS_MBUF(ip6af), i);
41762587Sitojun			ip6af->ip6af_off += i;
41862587Sitojun			ip6af->ip6af_frglen -= i;
41962587Sitojun		}
42062587Sitojun	}
42162587Sitojun
42262587Sitojun	/*
42362587Sitojun	 * While we overlap succeeding segments trim them or,
42462587Sitojun	 * if they are completely covered, dequeue them.
42562587Sitojun	 */
42662587Sitojun	while (af6 != (struct ip6asfrag *)q6 &&
42762587Sitojun	       ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) {
42862587Sitojun		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
42962587Sitojun		if (i < af6->ip6af_frglen) {
43062587Sitojun			af6->ip6af_frglen -= i;
43162587Sitojun			af6->ip6af_off += i;
43262587Sitojun			m_adj(IP6_REASS_MBUF(af6), i);
43362587Sitojun			break;
43462587Sitojun		}
43562587Sitojun		af6 = af6->ip6af_down;
43662587Sitojun		m_freem(IP6_REASS_MBUF(af6->ip6af_up));
43762587Sitojun		frag6_deq(af6->ip6af_up);
43862587Sitojun	}
43962587Sitojun#else
44062587Sitojun	/*
44153541Sshin	 * If the incoming framgent overlaps some existing fragments in
44253541Sshin	 * the reassembly queue, drop it, since it is dangerous to override
44353541Sshin	 * existing fragments from a security point of view.
444121345Sume	 * We don't know which fragment is the bad guy - here we trust
445121345Sume	 * fragment that came in earlier, with no real reason.
44653541Sshin	 */
44753541Sshin	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
44853541Sshin		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
44953541Sshin			- ip6af->ip6af_off;
45053541Sshin		if (i > 0) {
45176899Ssumikawa#if 0				/* suppress the noisy log */
45253541Sshin			log(LOG_ERR, "%d bytes of a fragment from %s "
45353541Sshin			    "overlaps the previous fragment\n",
45453541Sshin			    i, ip6_sprintf(&q6->ip6q_src));
45576899Ssumikawa#endif
45676899Ssumikawa			free(ip6af, M_FTABLE);
45753541Sshin			goto dropfrag;
45853541Sshin		}
45953541Sshin	}
46053541Sshin	if (af6 != (struct ip6asfrag *)q6) {
46153541Sshin		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
46253541Sshin		if (i > 0) {
46376899Ssumikawa#if 0				/* suppress the noisy log */
46453541Sshin			log(LOG_ERR, "%d bytes of a fragment from %s "
46553541Sshin			    "overlaps the succeeding fragment",
46653541Sshin			    i, ip6_sprintf(&q6->ip6q_src));
46776899Ssumikawa#endif
46876899Ssumikawa			free(ip6af, M_FTABLE);
46953541Sshin			goto dropfrag;
47053541Sshin		}
47153541Sshin	}
47262587Sitojun#endif
47353541Sshin
47453541Sshininsert:
47553541Sshin
47653541Sshin	/*
47753541Sshin	 * Stick new segment in its place;
47853541Sshin	 * check for complete reassembly.
47953541Sshin	 * Move to front of packet queue, as we are
48053541Sshin	 * the most recently active fragmented packet.
48153541Sshin	 */
48253541Sshin	frag6_enq(ip6af, af6->ip6af_up);
483121345Sume	frag6_nfrags++;
484121345Sume	q6->ip6q_nfrag++;
48562587Sitojun#if 0 /* xxx */
48662587Sitojun	if (q6 != ip6q.ip6q_next) {
48762587Sitojun		frag6_remque(q6);
48862587Sitojun		frag6_insque(q6, &ip6q);
48962587Sitojun	}
49062587Sitojun#endif
49153541Sshin	next = 0;
49253541Sshin	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
49353541Sshin	     af6 = af6->ip6af_down) {
49453541Sshin		if (af6->ip6af_off != next) {
495121345Sume			IP6Q_UNLOCK();
49653541Sshin			return IPPROTO_DONE;
49753541Sshin		}
49853541Sshin		next += af6->ip6af_frglen;
49953541Sshin	}
50053541Sshin	if (af6->ip6af_up->ip6af_mff) {
501121345Sume		IP6Q_UNLOCK();
50253541Sshin		return IPPROTO_DONE;
50353541Sshin	}
50453541Sshin
50553541Sshin	/*
50653541Sshin	 * Reassembly is complete; concatenate fragments.
50753541Sshin	 */
50853541Sshin	ip6af = q6->ip6q_down;
50953541Sshin	t = m = IP6_REASS_MBUF(ip6af);
51053541Sshin	af6 = ip6af->ip6af_down;
51162587Sitojun	frag6_deq(ip6af);
51253541Sshin	while (af6 != (struct ip6asfrag *)q6) {
51362587Sitojun		af6dwn = af6->ip6af_down;
51462587Sitojun		frag6_deq(af6);
51553541Sshin		while (t->m_next)
51653541Sshin			t = t->m_next;
51753541Sshin		t->m_next = IP6_REASS_MBUF(af6);
51862587Sitojun		m_adj(t->m_next, af6->ip6af_offset);
51962587Sitojun		free(af6, M_FTABLE);
52062587Sitojun		af6 = af6dwn;
52153541Sshin	}
52253541Sshin
52353541Sshin	/* adjust offset to point where the original next header starts */
52453541Sshin	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
52562587Sitojun	free(ip6af, M_FTABLE);
52662587Sitojun	ip6 = mtod(m, struct ip6_hdr *);
52753541Sshin	ip6->ip6_plen = htons((u_short)next + offset - sizeof(struct ip6_hdr));
52853541Sshin	ip6->ip6_src = q6->ip6q_src;
52953541Sshin	ip6->ip6_dst = q6->ip6q_dst;
53053541Sshin	nxt = q6->ip6q_nxt;
53162587Sitojun#ifdef notyet
53262587Sitojun	*q6->ip6q_nxtp = (u_char)(nxt & 0xff);
53362587Sitojun#endif
53453541Sshin
53553541Sshin	/*
53653541Sshin	 * Delete frag6 header with as a few cost as possible.
53753541Sshin	 */
53862587Sitojun	if (offset < m->m_len) {
53953541Sshin		ovbcopy((caddr_t)ip6, (caddr_t)ip6 + sizeof(struct ip6_frag),
54053541Sshin			offset);
54162587Sitojun		m->m_data += sizeof(struct ip6_frag);
54262587Sitojun		m->m_len -= sizeof(struct ip6_frag);
54362587Sitojun	} else {
54462587Sitojun		/* this comes with no copy if the boundary is on cluster */
545111119Simp		if ((t = m_split(m, offset, M_DONTWAIT)) == NULL) {
54662587Sitojun			frag6_remque(q6);
547121345Sume			frag6_nfrags -= q6->ip6q_nfrag;
54862587Sitojun			free(q6, M_FTABLE);
54962587Sitojun			frag6_nfragpackets--;
55062587Sitojun			goto dropfrag;
55162587Sitojun		}
55262587Sitojun		m_adj(t, sizeof(struct ip6_frag));
55362587Sitojun		m_cat(m, t);
55453541Sshin	}
55553541Sshin
55653541Sshin	/*
55753541Sshin	 * Store NXT to the original.
55853541Sshin	 */
55953541Sshin	{
56053541Sshin		char *prvnxtp = ip6_get_prevhdr(m, offset); /* XXX */
56153541Sshin		*prvnxtp = nxt;
56253541Sshin	}
56353541Sshin
56453541Sshin	frag6_remque(q6);
565121345Sume	frag6_nfrags -= q6->ip6q_nfrag;
56653541Sshin	free(q6, M_FTABLE);
56753541Sshin	frag6_nfragpackets--;
56853541Sshin
56953541Sshin	if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
57053541Sshin		int plen = 0;
57153541Sshin		for (t = m; t; t = t->m_next)
57253541Sshin			plen += t->m_len;
57353541Sshin		m->m_pkthdr.len = plen;
57453541Sshin	}
575120891Sume
57653541Sshin	ip6stat.ip6s_reassembled++;
57753541Sshin	in6_ifstat_inc(dstifp, ifs6_reass_ok);
57853541Sshin
57953541Sshin	/*
58053541Sshin	 * Tell launch routine the next header
58153541Sshin	 */
58253541Sshin
58353541Sshin	*mp = m;
58453541Sshin	*offp = offset;
58553541Sshin
586121345Sume	IP6Q_UNLOCK();
58753541Sshin	return nxt;
58853541Sshin
58953541Sshin dropfrag:
59053541Sshin	in6_ifstat_inc(dstifp, ifs6_reass_fail);
59153541Sshin	ip6stat.ip6s_fragdropped++;
59253541Sshin	m_freem(m);
593121345Sume	IP6Q_UNLOCK();
59453541Sshin	return IPPROTO_DONE;
59553541Sshin}
59653541Sshin
59753541Sshin/*
59853541Sshin * Free a fragment reassembly header and all
59953541Sshin * associated datagrams.
60053541Sshin */
60153541Sshinvoid
60253541Sshinfrag6_freef(q6)
60353541Sshin	struct ip6q *q6;
60453541Sshin{
60553541Sshin	struct ip6asfrag *af6, *down6;
60653541Sshin
607121345Sume	IP6Q_LOCK_CHECK();
608121345Sume
60953541Sshin	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
61053541Sshin	     af6 = down6) {
61153541Sshin		struct mbuf *m = IP6_REASS_MBUF(af6);
61253541Sshin
61353541Sshin		down6 = af6->ip6af_down;
61453541Sshin		frag6_deq(af6);
61553541Sshin
61653541Sshin		/*
61753541Sshin		 * Return ICMP time exceeded error for the 1st fragment.
61853541Sshin		 * Just free other fragments.
61953541Sshin		 */
62053541Sshin		if (af6->ip6af_off == 0) {
62153541Sshin			struct ip6_hdr *ip6;
62253541Sshin
62353541Sshin			/* adjust pointer */
62453541Sshin			ip6 = mtod(m, struct ip6_hdr *);
62553541Sshin
626120891Sume			/* restore source and destination addresses */
62753541Sshin			ip6->ip6_src = q6->ip6q_src;
62853541Sshin			ip6->ip6_dst = q6->ip6q_dst;
62953541Sshin
63053541Sshin			icmp6_error(m, ICMP6_TIME_EXCEEDED,
63153541Sshin				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
63262587Sitojun		} else
63353541Sshin			m_freem(m);
63462587Sitojun		free(af6, M_FTABLE);
63553541Sshin	}
63653541Sshin	frag6_remque(q6);
637121345Sume	frag6_nfrags -= q6->ip6q_nfrag;
63853541Sshin	free(q6, M_FTABLE);
63953541Sshin	frag6_nfragpackets--;
64053541Sshin}
64153541Sshin
64253541Sshin/*
64353541Sshin * Put an ip fragment on a reassembly chain.
64453541Sshin * Like insque, but pointers in middle of structure.
64553541Sshin */
64653541Sshinvoid
64753541Sshinfrag6_enq(af6, up6)
64853541Sshin	struct ip6asfrag *af6, *up6;
64953541Sshin{
650121345Sume
651121345Sume	IP6Q_LOCK_CHECK();
652121345Sume
65353541Sshin	af6->ip6af_up = up6;
65453541Sshin	af6->ip6af_down = up6->ip6af_down;
65553541Sshin	up6->ip6af_down->ip6af_up = af6;
65653541Sshin	up6->ip6af_down = af6;
65753541Sshin}
65853541Sshin
65953541Sshin/*
66053541Sshin * To frag6_enq as remque is to insque.
66153541Sshin */
66253541Sshinvoid
66353541Sshinfrag6_deq(af6)
66453541Sshin	struct ip6asfrag *af6;
66553541Sshin{
666121345Sume
667121345Sume	IP6Q_LOCK_CHECK();
668121345Sume
66953541Sshin	af6->ip6af_up->ip6af_down = af6->ip6af_down;
67053541Sshin	af6->ip6af_down->ip6af_up = af6->ip6af_up;
67153541Sshin}
67253541Sshin
67353541Sshinvoid
67453541Sshinfrag6_insque(new, old)
67553541Sshin	struct ip6q *new, *old;
67653541Sshin{
677121345Sume
678121345Sume	IP6Q_LOCK_CHECK();
679121345Sume
68053541Sshin	new->ip6q_prev = old;
68153541Sshin	new->ip6q_next = old->ip6q_next;
68253541Sshin	old->ip6q_next->ip6q_prev= new;
68353541Sshin	old->ip6q_next = new;
68453541Sshin}
68553541Sshin
68653541Sshinvoid
68753541Sshinfrag6_remque(p6)
68853541Sshin	struct ip6q *p6;
68953541Sshin{
690121345Sume
691121345Sume	IP6Q_LOCK_CHECK();
692121345Sume
69353541Sshin	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
69453541Sshin	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
69553541Sshin}
69653541Sshin
69753541Sshin/*
69878064Sume * IPv6 reassembling timer processing;
69953541Sshin * if a timer expires on a reassembly
70053541Sshin * queue, discard it.
70153541Sshin */
70253541Sshinvoid
70353541Sshinfrag6_slowtimo()
70453541Sshin{
70553541Sshin	struct ip6q *q6;
70653541Sshin	int s = splnet();
70753541Sshin
708121345Sume	IP6Q_LOCK();
70953541Sshin	q6 = ip6q.ip6q_next;
71053541Sshin	if (q6)
71153541Sshin		while (q6 != &ip6q) {
71253541Sshin			--q6->ip6q_ttl;
71353541Sshin			q6 = q6->ip6q_next;
71453541Sshin			if (q6->ip6q_prev->ip6q_ttl == 0) {
71553541Sshin				ip6stat.ip6s_fragtimeout++;
71653541Sshin				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
71753541Sshin				frag6_freef(q6->ip6q_prev);
71853541Sshin			}
71953541Sshin		}
72053541Sshin	/*
72153541Sshin	 * If we are over the maximum number of fragments
72253541Sshin	 * (due to the limit being lowered), drain off
72353541Sshin	 * enough to get down to the new limit.
72453541Sshin	 */
72578064Sume	while (frag6_nfragpackets > (u_int)ip6_maxfragpackets &&
72678064Sume	    ip6q.ip6q_prev) {
72753541Sshin		ip6stat.ip6s_fragoverflow++;
72853541Sshin		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
72953541Sshin		frag6_freef(ip6q.ip6q_prev);
73053541Sshin	}
731121345Sume	IP6Q_UNLOCK();
73262587Sitojun
73362587Sitojun#if 0
73462587Sitojun	/*
73562587Sitojun	 * Routing changes might produce a better route than we last used;
73662587Sitojun	 * make sure we notice eventually, even if forwarding only for one
73762587Sitojun	 * destination and the cache is never replaced.
73862587Sitojun	 */
73962587Sitojun	if (ip6_forward_rt.ro_rt) {
74062587Sitojun		RTFREE(ip6_forward_rt.ro_rt);
74162587Sitojun		ip6_forward_rt.ro_rt = 0;
74262587Sitojun	}
74362587Sitojun	if (ipsrcchk_rt.ro_rt) {
74462587Sitojun		RTFREE(ipsrcchk_rt.ro_rt);
74562587Sitojun		ipsrcchk_rt.ro_rt = 0;
74662587Sitojun	}
74762587Sitojun#endif
74862587Sitojun
74953541Sshin	splx(s);
75053541Sshin}
75153541Sshin
75253541Sshin/*
75353541Sshin * Drain off all datagram fragments.
75453541Sshin */
75553541Sshinvoid
75653541Sshinfrag6_drain()
75753541Sshin{
758121345Sume
759121345Sume	if (ip6q_lock_try() == 0)
76053541Sshin		return;
76153541Sshin	while (ip6q.ip6q_next != &ip6q) {
76253541Sshin		ip6stat.ip6s_fragdropped++;
76353541Sshin		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
76453541Sshin		frag6_freef(ip6q.ip6q_next);
76553541Sshin	}
766121345Sume	IP6Q_UNLOCK();
76753541Sshin}
768