Home | History | Annotate | Line # | Download | only in netinet6
frag6.c revision 1.33
      1 /*	$NetBSD: frag6.c,v 1.33 2006/12/15 21:18:54 joerg Exp $	*/
      2 /*	$KAME: frag6.c,v 1.40 2002/05/27 21:40:31 itojun Exp $	*/
      3 
      4 /*
      5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
      6  * All rights reserved.
      7  *
      8  * Redistribution and use in source and binary forms, with or without
      9  * modification, are permitted provided that the following conditions
     10  * are met:
     11  * 1. Redistributions of source code must retain the above copyright
     12  *    notice, this list of conditions and the following disclaimer.
     13  * 2. Redistributions in binary form must reproduce the above copyright
     14  *    notice, this list of conditions and the following disclaimer in the
     15  *    documentation and/or other materials provided with the distribution.
     16  * 3. Neither the name of the project nor the names of its contributors
     17  *    may be used to endorse or promote products derived from this software
     18  *    without specific prior written permission.
     19  *
     20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
     21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
     24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     30  * SUCH DAMAGE.
     31  */
     32 
     33 #include <sys/cdefs.h>
     34 __KERNEL_RCSID(0, "$NetBSD: frag6.c,v 1.33 2006/12/15 21:18:54 joerg Exp $");
     35 
     36 #include <sys/param.h>
     37 #include <sys/systm.h>
     38 #include <sys/malloc.h>
     39 #include <sys/mbuf.h>
     40 #include <sys/domain.h>
     41 #include <sys/protosw.h>
     42 #include <sys/socket.h>
     43 #include <sys/errno.h>
     44 #include <sys/time.h>
     45 #include <sys/kernel.h>
     46 #include <sys/syslog.h>
     47 
     48 #include <net/if.h>
     49 #include <net/route.h>
     50 
     51 #include <netinet/in.h>
     52 #include <netinet/in_var.h>
     53 #include <netinet/ip6.h>
     54 #include <netinet6/ip6_var.h>
     55 #include <netinet/icmp6.h>
     56 
     57 #include <net/net_osdep.h>
     58 
     59 /*
     60  * Define it to get a correct behavior on per-interface statistics.
     61  * You will need to perform an extra routing table lookup, per fragment,
     62  * to do it.  This may, or may not be, a performance hit.
     63  */
     64 #define IN6_IFSTAT_STRICT
     65 
     66 static void frag6_enq __P((struct ip6asfrag *, struct ip6asfrag *));
     67 static void frag6_deq __P((struct ip6asfrag *));
     68 static void frag6_insque __P((struct ip6q *, struct ip6q *));
     69 static void frag6_remque __P((struct ip6q *));
     70 static void frag6_freef __P((struct ip6q *));
     71 
     72 static int ip6q_locked;
     73 u_int frag6_nfragpackets;
     74 u_int frag6_nfrags;
     75 struct	ip6q ip6q;	/* ip6 reassemble queue */
     76 
     77 static inline int ip6q_lock_try __P((void));
     78 static inline void ip6q_unlock __P((void));
     79 
     80 static inline int
     81 ip6q_lock_try()
     82 {
     83 	int s;
     84 
     85 	/*
     86 	 * Use splvm() -- we're bloking things that would cause
     87 	 * mbuf allocation.
     88 	 */
     89 	s = splvm();
     90 	if (ip6q_locked) {
     91 		splx(s);
     92 		return (0);
     93 	}
     94 	ip6q_locked = 1;
     95 	splx(s);
     96 	return (1);
     97 }
     98 
     99 static inline void
    100 ip6q_unlock()
    101 {
    102 	int s;
    103 
    104 	s = splvm();
    105 	ip6q_locked = 0;
    106 	splx(s);
    107 }
    108 
    109 #ifdef DIAGNOSTIC
    110 #define	IP6Q_LOCK()							\
    111 do {									\
    112 	if (ip6q_lock_try() == 0) {					\
    113 		printf("%s:%d: ip6q already locked\n", __FILE__, __LINE__); \
    114 		panic("ip6q_lock");					\
    115 	}								\
    116 } while (/*CONSTCOND*/ 0)
    117 #define	IP6Q_LOCK_CHECK()						\
    118 do {									\
    119 	if (ip6q_locked == 0) {						\
    120 		printf("%s:%d: ip6q lock not held\n", __FILE__, __LINE__); \
    121 		panic("ip6q lock check");				\
    122 	}								\
    123 } while (/*CONSTCOND*/ 0)
    124 #else
    125 #define	IP6Q_LOCK()		(void) ip6q_lock_try()
    126 #define	IP6Q_LOCK_CHECK()	/* nothing */
    127 #endif
    128 
    129 #define	IP6Q_UNLOCK()		ip6q_unlock()
    130 
    131 #ifndef offsetof		/* XXX */
    132 #define	offsetof(type, member)	((size_t)(&((type *)0)->member))
    133 #endif
    134 
    135 /*
    136  * Initialise reassembly queue and fragment identifier.
    137  */
    138 void
    139 frag6_init()
    140 {
    141 
    142 	ip6q.ip6q_next = ip6q.ip6q_prev = &ip6q;
    143 }
    144 
    145 /*
    146  * In RFC2460, fragment and reassembly rule do not agree with each other,
    147  * in terms of next header field handling in fragment header.
    148  * While the sender will use the same value for all of the fragmented packets,
    149  * receiver is suggested not to check the consistency.
    150  *
    151  * fragment rule (p20):
    152  *	(2) A Fragment header containing:
    153  *	The Next Header value that identifies the first header of
    154  *	the Fragmentable Part of the original packet.
    155  *		-> next header field is same for all fragments
    156  *
    157  * reassembly rule (p21):
    158  *	The Next Header field of the last header of the Unfragmentable
    159  *	Part is obtained from the Next Header field of the first
    160  *	fragment's Fragment header.
    161  *		-> should grab it from the first fragment only
    162  *
    163  * The following note also contradicts with fragment rule - noone is going to
    164  * send different fragment with different next header field.
    165  *
    166  * additional note (p22):
    167  *	The Next Header values in the Fragment headers of different
    168  *	fragments of the same original packet may differ.  Only the value
    169  *	from the Offset zero fragment packet is used for reassembly.
    170  *		-> should grab it from the first fragment only
    171  *
    172  * There is no explicit reason given in the RFC.  Historical reason maybe?
    173  */
    174 /*
    175  * Fragment input
    176  */
    177 int
    178 frag6_input(struct mbuf **mp, int *offp, int proto)
    179 {
    180 	struct mbuf *m = *mp, *t;
    181 	struct ip6_hdr *ip6;
    182 	struct ip6_frag *ip6f;
    183 	struct ip6q *q6;
    184 	struct ip6asfrag *af6, *ip6af, *af6dwn;
    185 	int offset = *offp, nxt, i, next;
    186 	int first_frag = 0;
    187 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
    188 	struct ifnet *dstifp;
    189 #ifdef IN6_IFSTAT_STRICT
    190 	static struct route_in6 ro;
    191 	struct sockaddr_in6 *dst;
    192 #endif
    193 
    194 	ip6 = mtod(m, struct ip6_hdr *);
    195 	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
    196 	if (ip6f == NULL)
    197 		return IPPROTO_DONE;
    198 
    199 	dstifp = NULL;
    200 #ifdef IN6_IFSTAT_STRICT
    201 	/* find the destination interface of the packet. */
    202 	dst = (struct sockaddr_in6 *)&ro.ro_dst;
    203 	if (!IN6_ARE_ADDR_EQUAL(&dst->sin6_addr, &ip6->ip6_dst))
    204 		rtcache_free((struct route *)&ro);
    205 	else
    206 		rtcache_check((struct route *)&ro);
    207 	if (ro.ro_rt == NULL) {
    208 		bzero(dst, sizeof(*dst));
    209 		dst->sin6_family = AF_INET6;
    210 		dst->sin6_len = sizeof(struct sockaddr_in6);
    211 		dst->sin6_addr = ip6->ip6_dst;
    212 		rtcache_init((struct route *)&ro);
    213 	}
    214 	if (ro.ro_rt != NULL && ro.ro_rt->rt_ifa != NULL)
    215 		dstifp = ((struct in6_ifaddr *)ro.ro_rt->rt_ifa)->ia_ifp;
    216 #else
    217 	/* we are violating the spec, this is not the destination interface */
    218 	if ((m->m_flags & M_PKTHDR) != 0)
    219 		dstifp = m->m_pkthdr.rcvif;
    220 #endif
    221 
    222 	/* jumbo payload can't contain a fragment header */
    223 	if (ip6->ip6_plen == 0) {
    224 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
    225 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
    226 		return IPPROTO_DONE;
    227 	}
    228 
    229 	/*
    230 	 * check whether fragment packet's fragment length is
    231 	 * multiple of 8 octets.
    232 	 * sizeof(struct ip6_frag) == 8
    233 	 * sizeof(struct ip6_hdr) = 40
    234 	 */
    235 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
    236 	    (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
    237 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    238 		    offsetof(struct ip6_hdr, ip6_plen));
    239 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
    240 		return IPPROTO_DONE;
    241 	}
    242 
    243 	ip6stat.ip6s_fragments++;
    244 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
    245 
    246 	/* offset now points to data portion */
    247 	offset += sizeof(struct ip6_frag);
    248 
    249 	IP6Q_LOCK();
    250 
    251 	/*
    252 	 * Enforce upper bound on number of fragments.
    253 	 * If maxfrag is 0, never accept fragments.
    254 	 * If maxfrag is -1, accept all fragments without limitation.
    255 	 */
    256 	if (ip6_maxfrags < 0)
    257 		;
    258 	else if (frag6_nfrags >= (u_int)ip6_maxfrags)
    259 		goto dropfrag;
    260 
    261 	for (q6 = ip6q.ip6q_next; q6 != &ip6q; q6 = q6->ip6q_next)
    262 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
    263 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
    264 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst))
    265 			break;
    266 
    267 	if (q6 == &ip6q) {
    268 		/*
    269 		 * the first fragment to arrive, create a reassembly queue.
    270 		 */
    271 		first_frag = 1;
    272 
    273 		/*
    274 		 * Enforce upper bound on number of fragmented packets
    275 		 * for which we attempt reassembly;
    276 		 * If maxfragpackets is 0, never accept fragments.
    277 		 * If maxfragpackets is -1, accept all fragments without
    278 		 * limitation.
    279 		 */
    280 		if (ip6_maxfragpackets < 0)
    281 			;
    282 		else if (frag6_nfragpackets >= (u_int)ip6_maxfragpackets)
    283 			goto dropfrag;
    284 		frag6_nfragpackets++;
    285 		q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FTABLE,
    286 		    M_DONTWAIT);
    287 		if (q6 == NULL)
    288 			goto dropfrag;
    289 		bzero(q6, sizeof(*q6));
    290 
    291 		frag6_insque(q6, &ip6q);
    292 
    293 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
    294 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
    295 #ifdef notyet
    296 		q6->ip6q_nxtp	= (u_char *)nxtp;
    297 #endif
    298 		q6->ip6q_ident	= ip6f->ip6f_ident;
    299 		q6->ip6q_arrive = 0; /* Is it used anywhere? */
    300 		q6->ip6q_ttl 	= IPV6_FRAGTTL;
    301 		q6->ip6q_src	= ip6->ip6_src;
    302 		q6->ip6q_dst	= ip6->ip6_dst;
    303 		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
    304 
    305 		q6->ip6q_nfrag = 0;
    306 	}
    307 
    308 	/*
    309 	 * If it's the 1st fragment, record the length of the
    310 	 * unfragmentable part and the next header of the fragment header.
    311 	 */
    312 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
    313 	if (fragoff == 0) {
    314 		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
    315 		    sizeof(struct ip6_frag);
    316 		q6->ip6q_nxt = ip6f->ip6f_nxt;
    317 	}
    318 
    319 	/*
    320 	 * Check that the reassembled packet would not exceed 65535 bytes
    321 	 * in size.
    322 	 * If it would exceed, discard the fragment and return an ICMP error.
    323 	 */
    324 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
    325 	if (q6->ip6q_unfrglen >= 0) {
    326 		/* The 1st fragment has already arrived. */
    327 		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
    328 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    329 			    offset - sizeof(struct ip6_frag) +
    330 			    offsetof(struct ip6_frag, ip6f_offlg));
    331 			IP6Q_UNLOCK();
    332 			return (IPPROTO_DONE);
    333 		}
    334 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
    335 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    336 			    offset - sizeof(struct ip6_frag) +
    337 				offsetof(struct ip6_frag, ip6f_offlg));
    338 		IP6Q_UNLOCK();
    339 		return (IPPROTO_DONE);
    340 	}
    341 	/*
    342 	 * If it's the first fragment, do the above check for each
    343 	 * fragment already stored in the reassembly queue.
    344 	 */
    345 	if (fragoff == 0) {
    346 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    347 		     af6 = af6dwn) {
    348 			af6dwn = af6->ip6af_down;
    349 
    350 			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
    351 			    IPV6_MAXPACKET) {
    352 				struct mbuf *merr = IP6_REASS_MBUF(af6);
    353 				struct ip6_hdr *ip6err;
    354 				int erroff = af6->ip6af_offset;
    355 
    356 				/* dequeue the fragment. */
    357 				frag6_deq(af6);
    358 				free(af6, M_FTABLE);
    359 
    360 				/* adjust pointer. */
    361 				ip6err = mtod(merr, struct ip6_hdr *);
    362 
    363 				/*
    364 				 * Restore source and destination addresses
    365 				 * in the erroneous IPv6 header.
    366 				 */
    367 				ip6err->ip6_src = q6->ip6q_src;
    368 				ip6err->ip6_dst = q6->ip6q_dst;
    369 
    370 				icmp6_error(merr, ICMP6_PARAM_PROB,
    371 				    ICMP6_PARAMPROB_HEADER,
    372 				    erroff - sizeof(struct ip6_frag) +
    373 				    offsetof(struct ip6_frag, ip6f_offlg));
    374 			}
    375 		}
    376 	}
    377 
    378 	ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FTABLE,
    379 	    M_DONTWAIT);
    380 	if (ip6af == NULL)
    381 		goto dropfrag;
    382 	bzero(ip6af, sizeof(*ip6af));
    383 	ip6af->ip6af_head = ip6->ip6_flow;
    384 	ip6af->ip6af_len = ip6->ip6_plen;
    385 	ip6af->ip6af_nxt = ip6->ip6_nxt;
    386 	ip6af->ip6af_hlim = ip6->ip6_hlim;
    387 	ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
    388 	ip6af->ip6af_off = fragoff;
    389 	ip6af->ip6af_frglen = frgpartlen;
    390 	ip6af->ip6af_offset = offset;
    391 	IP6_REASS_MBUF(ip6af) = m;
    392 
    393 	if (first_frag) {
    394 		af6 = (struct ip6asfrag *)q6;
    395 		goto insert;
    396 	}
    397 
    398 	/*
    399 	 * Find a segment which begins after this one does.
    400 	 */
    401 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    402 	     af6 = af6->ip6af_down)
    403 		if (af6->ip6af_off > ip6af->ip6af_off)
    404 			break;
    405 
    406 #if 0
    407 	/*
    408 	 * If there is a preceding segment, it may provide some of
    409 	 * our data already.  If so, drop the data from the incoming
    410 	 * segment.  If it provides all of our data, drop us.
    411 	 */
    412 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
    413 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
    414 			- ip6af->ip6af_off;
    415 		if (i > 0) {
    416 			if (i >= ip6af->ip6af_frglen)
    417 				goto dropfrag;
    418 			m_adj(IP6_REASS_MBUF(ip6af), i);
    419 			ip6af->ip6af_off += i;
    420 			ip6af->ip6af_frglen -= i;
    421 		}
    422 	}
    423 
    424 	/*
    425 	 * While we overlap succeeding segments trim them or,
    426 	 * if they are completely covered, dequeue them.
    427 	 */
    428 	while (af6 != (struct ip6asfrag *)q6 &&
    429 	       ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) {
    430 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
    431 		if (i < af6->ip6af_frglen) {
    432 			af6->ip6af_frglen -= i;
    433 			af6->ip6af_off += i;
    434 			m_adj(IP6_REASS_MBUF(af6), i);
    435 			break;
    436 		}
    437 		af6 = af6->ip6af_down;
    438 		m_freem(IP6_REASS_MBUF(af6->ip6af_up));
    439 		frag6_deq(af6->ip6af_up);
    440 	}
    441 #else
    442 	/*
    443 	 * If the incoming framgent overlaps some existing fragments in
    444 	 * the reassembly queue, drop it, since it is dangerous to override
    445 	 * existing fragments from a security point of view.
    446 	 * We don't know which fragment is the bad guy - here we trust
    447 	 * fragment that came in earlier, with no real reason.
    448 	 */
    449 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
    450 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
    451 			- ip6af->ip6af_off;
    452 		if (i > 0) {
    453 #if 0				/* suppress the noisy log */
    454 			log(LOG_ERR, "%d bytes of a fragment from %s "
    455 			    "overlaps the previous fragment\n",
    456 			    i, ip6_sprintf(&q6->ip6q_src));
    457 #endif
    458 			free(ip6af, M_FTABLE);
    459 			goto dropfrag;
    460 		}
    461 	}
    462 	if (af6 != (struct ip6asfrag *)q6) {
    463 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
    464 		if (i > 0) {
    465 #if 0				/* suppress the noisy log */
    466 			log(LOG_ERR, "%d bytes of a fragment from %s "
    467 			    "overlaps the succeeding fragment",
    468 			    i, ip6_sprintf(&q6->ip6q_src));
    469 #endif
    470 			free(ip6af, M_FTABLE);
    471 			goto dropfrag;
    472 		}
    473 	}
    474 #endif
    475 
    476 insert:
    477 
    478 	/*
    479 	 * Stick new segment in its place;
    480 	 * check for complete reassembly.
    481 	 * Move to front of packet queue, as we are
    482 	 * the most recently active fragmented packet.
    483 	 */
    484 	frag6_enq(ip6af, af6->ip6af_up);
    485 	frag6_nfrags++;
    486 	q6->ip6q_nfrag++;
    487 #if 0 /* xxx */
    488 	if (q6 != ip6q.ip6q_next) {
    489 		frag6_remque(q6);
    490 		frag6_insque(q6, &ip6q);
    491 	}
    492 #endif
    493 	next = 0;
    494 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    495 	     af6 = af6->ip6af_down) {
    496 		if (af6->ip6af_off != next) {
    497 			IP6Q_UNLOCK();
    498 			return IPPROTO_DONE;
    499 		}
    500 		next += af6->ip6af_frglen;
    501 	}
    502 	if (af6->ip6af_up->ip6af_mff) {
    503 		IP6Q_UNLOCK();
    504 		return IPPROTO_DONE;
    505 	}
    506 
    507 	/*
    508 	 * Reassembly is complete; concatenate fragments.
    509 	 */
    510 	ip6af = q6->ip6q_down;
    511 	t = m = IP6_REASS_MBUF(ip6af);
    512 	af6 = ip6af->ip6af_down;
    513 	frag6_deq(ip6af);
    514 	while (af6 != (struct ip6asfrag *)q6) {
    515 		af6dwn = af6->ip6af_down;
    516 		frag6_deq(af6);
    517 		while (t->m_next)
    518 			t = t->m_next;
    519 		t->m_next = IP6_REASS_MBUF(af6);
    520 		m_adj(t->m_next, af6->ip6af_offset);
    521 		free(af6, M_FTABLE);
    522 		af6 = af6dwn;
    523 	}
    524 
    525 	/* adjust offset to point where the original next header starts */
    526 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
    527 	free(ip6af, M_FTABLE);
    528 	ip6 = mtod(m, struct ip6_hdr *);
    529 	ip6->ip6_plen = htons(next + offset - sizeof(struct ip6_hdr));
    530 	ip6->ip6_src = q6->ip6q_src;
    531 	ip6->ip6_dst = q6->ip6q_dst;
    532 	nxt = q6->ip6q_nxt;
    533 #ifdef notyet
    534 	*q6->ip6q_nxtp = (u_char)(nxt & 0xff);
    535 #endif
    536 
    537 	/*
    538 	 * Delete frag6 header with as a few cost as possible.
    539 	 */
    540 	if (offset < m->m_len) {
    541 		ovbcopy((caddr_t)ip6, (caddr_t)ip6 + sizeof(struct ip6_frag),
    542 			offset);
    543 		m->m_data += sizeof(struct ip6_frag);
    544 		m->m_len -= sizeof(struct ip6_frag);
    545 	} else {
    546 		/* this comes with no copy if the boundary is on cluster */
    547 		if ((t = m_split(m, offset, M_DONTWAIT)) == NULL) {
    548 			frag6_remque(q6);
    549 			frag6_nfrags -= q6->ip6q_nfrag;
    550 			free(q6, M_FTABLE);
    551 			frag6_nfragpackets--;
    552 			goto dropfrag;
    553 		}
    554 		m_adj(t, sizeof(struct ip6_frag));
    555 		m_cat(m, t);
    556 	}
    557 
    558 	/*
    559 	 * Store NXT to the original.
    560 	 */
    561 	{
    562 		u_int8_t *prvnxtp = ip6_get_prevhdr(m, offset); /* XXX */
    563 		*prvnxtp = nxt;
    564 	}
    565 
    566 	frag6_remque(q6);
    567 	frag6_nfrags -= q6->ip6q_nfrag;
    568 	free(q6, M_FTABLE);
    569 	frag6_nfragpackets--;
    570 
    571 	if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
    572 		int plen = 0;
    573 		for (t = m; t; t = t->m_next)
    574 			plen += t->m_len;
    575 		m->m_pkthdr.len = plen;
    576 	}
    577 
    578 	ip6stat.ip6s_reassembled++;
    579 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
    580 
    581 	/*
    582 	 * Tell launch routine the next header
    583 	 */
    584 
    585 	*mp = m;
    586 	*offp = offset;
    587 
    588 	IP6Q_UNLOCK();
    589 	return nxt;
    590 
    591  dropfrag:
    592 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
    593 	ip6stat.ip6s_fragdropped++;
    594 	m_freem(m);
    595 	IP6Q_UNLOCK();
    596 	return IPPROTO_DONE;
    597 }
    598 
    599 /*
    600  * Free a fragment reassembly header and all
    601  * associated datagrams.
    602  */
    603 void
    604 frag6_freef(q6)
    605 	struct ip6q *q6;
    606 {
    607 	struct ip6asfrag *af6, *down6;
    608 
    609 	IP6Q_LOCK_CHECK();
    610 
    611 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    612 	     af6 = down6) {
    613 		struct mbuf *m = IP6_REASS_MBUF(af6);
    614 
    615 		down6 = af6->ip6af_down;
    616 		frag6_deq(af6);
    617 
    618 		/*
    619 		 * Return ICMP time exceeded error for the 1st fragment.
    620 		 * Just free other fragments.
    621 		 */
    622 		if (af6->ip6af_off == 0) {
    623 			struct ip6_hdr *ip6;
    624 
    625 			/* adjust pointer */
    626 			ip6 = mtod(m, struct ip6_hdr *);
    627 
    628 			/* restoure source and destination addresses */
    629 			ip6->ip6_src = q6->ip6q_src;
    630 			ip6->ip6_dst = q6->ip6q_dst;
    631 
    632 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
    633 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
    634 		} else
    635 			m_freem(m);
    636 		free(af6, M_FTABLE);
    637 	}
    638 	frag6_remque(q6);
    639 	frag6_nfrags -= q6->ip6q_nfrag;
    640 	free(q6, M_FTABLE);
    641 	frag6_nfragpackets--;
    642 }
    643 
    644 /*
    645  * Put an ip fragment on a reassembly chain.
    646  * Like insque, but pointers in middle of structure.
    647  */
    648 void
    649 frag6_enq(af6, up6)
    650 	struct ip6asfrag *af6, *up6;
    651 {
    652 
    653 	IP6Q_LOCK_CHECK();
    654 
    655 	af6->ip6af_up = up6;
    656 	af6->ip6af_down = up6->ip6af_down;
    657 	up6->ip6af_down->ip6af_up = af6;
    658 	up6->ip6af_down = af6;
    659 }
    660 
    661 /*
    662  * To frag6_enq as remque is to insque.
    663  */
    664 void
    665 frag6_deq(af6)
    666 	struct ip6asfrag *af6;
    667 {
    668 
    669 	IP6Q_LOCK_CHECK();
    670 
    671 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
    672 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
    673 }
    674 
    675 void
    676 frag6_insque(new, old)
    677 	struct ip6q *new, *old;
    678 {
    679 
    680 	IP6Q_LOCK_CHECK();
    681 
    682 	new->ip6q_prev = old;
    683 	new->ip6q_next = old->ip6q_next;
    684 	old->ip6q_next->ip6q_prev= new;
    685 	old->ip6q_next = new;
    686 }
    687 
    688 void
    689 frag6_remque(p6)
    690 	struct ip6q *p6;
    691 {
    692 
    693 	IP6Q_LOCK_CHECK();
    694 
    695 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
    696 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
    697 }
    698 
    699 /*
    700  * IPv6 reassembling timer processing;
    701  * if a timer expires on a reassembly
    702  * queue, discard it.
    703  */
    704 void
    705 frag6_slowtimo()
    706 {
    707 	struct ip6q *q6;
    708 	int s = splsoftnet();
    709 
    710 	IP6Q_LOCK();
    711 	q6 = ip6q.ip6q_next;
    712 	if (q6)
    713 		while (q6 != &ip6q) {
    714 			--q6->ip6q_ttl;
    715 			q6 = q6->ip6q_next;
    716 			if (q6->ip6q_prev->ip6q_ttl == 0) {
    717 				ip6stat.ip6s_fragtimeout++;
    718 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    719 				frag6_freef(q6->ip6q_prev);
    720 			}
    721 		}
    722 	/*
    723 	 * If we are over the maximum number of fragments
    724 	 * (due to the limit being lowered), drain off
    725 	 * enough to get down to the new limit.
    726 	 */
    727 	while (frag6_nfragpackets > (u_int)ip6_maxfragpackets &&
    728 	    ip6q.ip6q_prev) {
    729 		ip6stat.ip6s_fragoverflow++;
    730 		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    731 		frag6_freef(ip6q.ip6q_prev);
    732 	}
    733 	IP6Q_UNLOCK();
    734 
    735 #if 0
    736 	/*
    737 	 * Routing changes might produce a better route than we last used;
    738 	 * make sure we notice eventually, even if forwarding only for one
    739 	 * destination and the cache is never replaced.
    740 	 */
    741 	rtcache_free((struct route *)&ip6_forward_rt);
    742 	rtcache_free((struct route *)&ipsrcchk_rt);
    743 #endif
    744 
    745 	splx(s);
    746 }
    747 
    748 /*
    749  * Drain off all datagram fragments.
    750  */
    751 void
    752 frag6_drain()
    753 {
    754 
    755 	if (ip6q_lock_try() == 0)
    756 		return;
    757 	while (ip6q.ip6q_next != &ip6q) {
    758 		ip6stat.ip6s_fragdropped++;
    759 		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    760 		frag6_freef(ip6q.ip6q_next);
    761 	}
    762 	IP6Q_UNLOCK();
    763 }
    764