Home | History | Annotate | Line # | Download | only in netinet6
frag6.c revision 1.28
      1 /*	$NetBSD: frag6.c,v 1.28 2005/12/24 20:45:09 perry 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 1.28 2005/12/24 20:45:09 perry 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/in6_pcb.h>
     55 #include <netinet6/ip6_var.h>
     56 #include <netinet/icmp6.h>
     57 
     58 #include <net/net_osdep.h>
     59 
     60 /*
     61  * Define it to get a correct behavior on per-interface statistics.
     62  * You will need to perform an extra routing table lookup, per fragment,
     63  * to do it.  This may, or may not be, a performance hit.
     64  */
     65 #define IN6_IFSTAT_STRICT
     66 
     67 static void frag6_enq __P((struct ip6asfrag *, struct ip6asfrag *));
     68 static void frag6_deq __P((struct ip6asfrag *));
     69 static void frag6_insque __P((struct ip6q *, struct ip6q *));
     70 static void frag6_remque __P((struct ip6q *));
     71 static void frag6_freef __P((struct ip6q *));
     72 
     73 static int ip6q_locked;
     74 u_int frag6_nfragpackets;
     75 u_int frag6_nfrags;
     76 struct	ip6q ip6q;	/* ip6 reassemble queue */
     77 
     78 static inline int ip6q_lock_try __P((void));
     79 static inline void ip6q_unlock __P((void));
     80 
     81 static inline int
     82 ip6q_lock_try()
     83 {
     84 	int s;
     85 
     86 	/*
     87 	 * Use splvm() -- we're bloking things that would cause
     88 	 * mbuf allocation.
     89 	 */
     90 	s = splvm();
     91 	if (ip6q_locked) {
     92 		splx(s);
     93 		return (0);
     94 	}
     95 	ip6q_locked = 1;
     96 	splx(s);
     97 	return (1);
     98 }
     99 
    100 static inline void
    101 ip6q_unlock()
    102 {
    103 	int s;
    104 
    105 	s = splvm();
    106 	ip6q_locked = 0;
    107 	splx(s);
    108 }
    109 
    110 #ifdef DIAGNOSTIC
    111 #define	IP6Q_LOCK()							\
    112 do {									\
    113 	if (ip6q_lock_try() == 0) {					\
    114 		printf("%s:%d: ip6q already locked\n", __FILE__, __LINE__); \
    115 		panic("ip6q_lock");					\
    116 	}								\
    117 } while (/*CONSTCOND*/ 0)
    118 #define	IP6Q_LOCK_CHECK()						\
    119 do {									\
    120 	if (ip6q_locked == 0) {						\
    121 		printf("%s:%d: ip6q lock not held\n", __FILE__, __LINE__); \
    122 		panic("ip6q lock check");				\
    123 	}								\
    124 } while (/*CONSTCOND*/ 0)
    125 #else
    126 #define	IP6Q_LOCK()		(void) ip6q_lock_try()
    127 #define	IP6Q_LOCK_CHECK()	/* nothing */
    128 #endif
    129 
    130 #define	IP6Q_UNLOCK()		ip6q_unlock()
    131 
    132 #ifndef offsetof		/* XXX */
    133 #define	offsetof(type, member)	((size_t)(&((type *)0)->member))
    134 #endif
    135 
    136 /*
    137  * Initialise reassembly queue and fragment identifier.
    138  */
    139 void
    140 frag6_init()
    141 {
    142 
    143 	ip6q.ip6q_next = ip6q.ip6q_prev = &ip6q;
    144 }
    145 
    146 /*
    147  * In RFC2460, fragment and reassembly rule do not agree with each other,
    148  * in terms of next header field handling in fragment header.
    149  * While the sender will use the same value for all of the fragmented packets,
    150  * receiver is suggested not to check the consistency.
    151  *
    152  * fragment rule (p20):
    153  *	(2) A Fragment header containing:
    154  *	The Next Header value that identifies the first header of
    155  *	the Fragmentable Part of the original packet.
    156  *		-> next header field is same for all fragments
    157  *
    158  * reassembly rule (p21):
    159  *	The Next Header field of the last header of the Unfragmentable
    160  *	Part is obtained from the Next Header field of the first
    161  *	fragment's Fragment header.
    162  *		-> should grab it from the first fragment only
    163  *
    164  * The following note also contradicts with fragment rule - noone is going to
    165  * send different fragment with different next header field.
    166  *
    167  * additional note (p22):
    168  *	The Next Header values in the Fragment headers of different
    169  *	fragments of the same original packet may differ.  Only the value
    170  *	from the Offset zero fragment packet is used for reassembly.
    171  *		-> should grab it from the first fragment only
    172  *
    173  * There is no explicit reason given in the RFC.  Historical reason maybe?
    174  */
    175 /*
    176  * Fragment input
    177  */
    178 int
    179 frag6_input(mp, offp, proto)
    180 	struct mbuf **mp;
    181 	int *offp, proto;
    182 {
    183 	struct mbuf *m = *mp, *t;
    184 	struct ip6_hdr *ip6;
    185 	struct ip6_frag *ip6f;
    186 	struct ip6q *q6;
    187 	struct ip6asfrag *af6, *ip6af, *af6dwn;
    188 	int offset = *offp, nxt, i, next;
    189 	int first_frag = 0;
    190 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
    191 	struct ifnet *dstifp;
    192 #ifdef IN6_IFSTAT_STRICT
    193 	static struct route_in6 ro;
    194 	struct sockaddr_in6 *dst;
    195 #endif
    196 
    197 	ip6 = mtod(m, struct ip6_hdr *);
    198 	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
    199 	if (ip6f == NULL)
    200 		return IPPROTO_DONE;
    201 
    202 	dstifp = NULL;
    203 #ifdef IN6_IFSTAT_STRICT
    204 	/* find the destination interface of the packet. */
    205 	dst = (struct sockaddr_in6 *)&ro.ro_dst;
    206 	if (ro.ro_rt
    207 	 && ((ro.ro_rt->rt_flags & RTF_UP) == 0
    208 	  || !IN6_ARE_ADDR_EQUAL(&dst->sin6_addr, &ip6->ip6_dst))) {
    209 		RTFREE(ro.ro_rt);
    210 		ro.ro_rt = (struct rtentry *)0;
    211 	}
    212 	if (ro.ro_rt == NULL) {
    213 		bzero(dst, sizeof(*dst));
    214 		dst->sin6_family = AF_INET6;
    215 		dst->sin6_len = sizeof(struct sockaddr_in6);
    216 		dst->sin6_addr = ip6->ip6_dst;
    217 	}
    218 	rtalloc((struct route *)&ro);
    219 	if (ro.ro_rt != NULL && ro.ro_rt->rt_ifa != NULL)
    220 		dstifp = ((struct in6_ifaddr *)ro.ro_rt->rt_ifa)->ia_ifp;
    221 #else
    222 	/* we are violating the spec, this is not the destination interface */
    223 	if ((m->m_flags & M_PKTHDR) != 0)
    224 		dstifp = m->m_pkthdr.rcvif;
    225 #endif
    226 
    227 	/* jumbo payload can't contain a fragment header */
    228 	if (ip6->ip6_plen == 0) {
    229 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
    230 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
    231 		return IPPROTO_DONE;
    232 	}
    233 
    234 	/*
    235 	 * check whether fragment packet's fragment length is
    236 	 * multiple of 8 octets.
    237 	 * sizeof(struct ip6_frag) == 8
    238 	 * sizeof(struct ip6_hdr) = 40
    239 	 */
    240 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
    241 	    (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
    242 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    243 		    offsetof(struct ip6_hdr, ip6_plen));
    244 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
    245 		return IPPROTO_DONE;
    246 	}
    247 
    248 	ip6stat.ip6s_fragments++;
    249 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
    250 
    251 	/* offset now points to data portion */
    252 	offset += sizeof(struct ip6_frag);
    253 
    254 	IP6Q_LOCK();
    255 
    256 	/*
    257 	 * Enforce upper bound on number of fragments.
    258 	 * If maxfrag is 0, never accept fragments.
    259 	 * If maxfrag is -1, accept all fragments without limitation.
    260 	 */
    261 	if (ip6_maxfrags < 0)
    262 		;
    263 	else if (frag6_nfrags >= (u_int)ip6_maxfrags)
    264 		goto dropfrag;
    265 
    266 	for (q6 = ip6q.ip6q_next; q6 != &ip6q; q6 = q6->ip6q_next)
    267 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
    268 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
    269 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst))
    270 			break;
    271 
    272 	if (q6 == &ip6q) {
    273 		/*
    274 		 * the first fragment to arrive, create a reassembly queue.
    275 		 */
    276 		first_frag = 1;
    277 
    278 		/*
    279 		 * Enforce upper bound on number of fragmented packets
    280 		 * for which we attempt reassembly;
    281 		 * If maxfragpackets is 0, never accept fragments.
    282 		 * If maxfragpackets is -1, accept all fragments without
    283 		 * limitation.
    284 		 */
    285 		if (ip6_maxfragpackets < 0)
    286 			;
    287 		else if (frag6_nfragpackets >= (u_int)ip6_maxfragpackets)
    288 			goto dropfrag;
    289 		frag6_nfragpackets++;
    290 		q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FTABLE,
    291 		    M_DONTWAIT);
    292 		if (q6 == NULL)
    293 			goto dropfrag;
    294 		bzero(q6, sizeof(*q6));
    295 
    296 		frag6_insque(q6, &ip6q);
    297 
    298 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
    299 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
    300 #ifdef notyet
    301 		q6->ip6q_nxtp	= (u_char *)nxtp;
    302 #endif
    303 		q6->ip6q_ident	= ip6f->ip6f_ident;
    304 		q6->ip6q_arrive = 0; /* Is it used anywhere? */
    305 		q6->ip6q_ttl 	= IPV6_FRAGTTL;
    306 		q6->ip6q_src	= ip6->ip6_src;
    307 		q6->ip6q_dst	= ip6->ip6_dst;
    308 		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
    309 
    310 		q6->ip6q_nfrag = 0;
    311 	}
    312 
    313 	/*
    314 	 * If it's the 1st fragment, record the length of the
    315 	 * unfragmentable part and the next header of the fragment header.
    316 	 */
    317 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
    318 	if (fragoff == 0) {
    319 		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
    320 		    sizeof(struct ip6_frag);
    321 		q6->ip6q_nxt = ip6f->ip6f_nxt;
    322 	}
    323 
    324 	/*
    325 	 * Check that the reassembled packet would not exceed 65535 bytes
    326 	 * in size.
    327 	 * If it would exceed, discard the fragment and return an ICMP error.
    328 	 */
    329 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
    330 	if (q6->ip6q_unfrglen >= 0) {
    331 		/* The 1st fragment has already arrived. */
    332 		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
    333 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    334 			    offset - sizeof(struct ip6_frag) +
    335 			    offsetof(struct ip6_frag, ip6f_offlg));
    336 			IP6Q_UNLOCK();
    337 			return (IPPROTO_DONE);
    338 		}
    339 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
    340 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    341 			    offset - sizeof(struct ip6_frag) +
    342 				offsetof(struct ip6_frag, ip6f_offlg));
    343 		IP6Q_UNLOCK();
    344 		return (IPPROTO_DONE);
    345 	}
    346 	/*
    347 	 * If it's the first fragment, do the above check for each
    348 	 * fragment already stored in the reassembly queue.
    349 	 */
    350 	if (fragoff == 0) {
    351 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    352 		     af6 = af6dwn) {
    353 			af6dwn = af6->ip6af_down;
    354 
    355 			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
    356 			    IPV6_MAXPACKET) {
    357 				struct mbuf *merr = IP6_REASS_MBUF(af6);
    358 				struct ip6_hdr *ip6err;
    359 				int erroff = af6->ip6af_offset;
    360 
    361 				/* dequeue the fragment. */
    362 				frag6_deq(af6);
    363 				free(af6, M_FTABLE);
    364 
    365 				/* adjust pointer. */
    366 				ip6err = mtod(merr, struct ip6_hdr *);
    367 
    368 				/*
    369 				 * Restore source and destination addresses
    370 				 * in the erroneous IPv6 header.
    371 				 */
    372 				ip6err->ip6_src = q6->ip6q_src;
    373 				ip6err->ip6_dst = q6->ip6q_dst;
    374 
    375 				icmp6_error(merr, ICMP6_PARAM_PROB,
    376 				    ICMP6_PARAMPROB_HEADER,
    377 				    erroff - sizeof(struct ip6_frag) +
    378 				    offsetof(struct ip6_frag, ip6f_offlg));
    379 			}
    380 		}
    381 	}
    382 
    383 	ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FTABLE,
    384 	    M_DONTWAIT);
    385 	if (ip6af == NULL)
    386 		goto dropfrag;
    387 	bzero(ip6af, sizeof(*ip6af));
    388 	ip6af->ip6af_head = ip6->ip6_flow;
    389 	ip6af->ip6af_len = ip6->ip6_plen;
    390 	ip6af->ip6af_nxt = ip6->ip6_nxt;
    391 	ip6af->ip6af_hlim = ip6->ip6_hlim;
    392 	ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
    393 	ip6af->ip6af_off = fragoff;
    394 	ip6af->ip6af_frglen = frgpartlen;
    395 	ip6af->ip6af_offset = offset;
    396 	IP6_REASS_MBUF(ip6af) = m;
    397 
    398 	if (first_frag) {
    399 		af6 = (struct ip6asfrag *)q6;
    400 		goto insert;
    401 	}
    402 
    403 	/*
    404 	 * Find a segment which begins after this one does.
    405 	 */
    406 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    407 	     af6 = af6->ip6af_down)
    408 		if (af6->ip6af_off > ip6af->ip6af_off)
    409 			break;
    410 
    411 #if 0
    412 	/*
    413 	 * If there is a preceding segment, it may provide some of
    414 	 * our data already.  If so, drop the data from the incoming
    415 	 * segment.  If it provides all of our data, drop us.
    416 	 */
    417 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
    418 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
    419 			- ip6af->ip6af_off;
    420 		if (i > 0) {
    421 			if (i >= ip6af->ip6af_frglen)
    422 				goto dropfrag;
    423 			m_adj(IP6_REASS_MBUF(ip6af), i);
    424 			ip6af->ip6af_off += i;
    425 			ip6af->ip6af_frglen -= i;
    426 		}
    427 	}
    428 
    429 	/*
    430 	 * While we overlap succeeding segments trim them or,
    431 	 * if they are completely covered, dequeue them.
    432 	 */
    433 	while (af6 != (struct ip6asfrag *)q6 &&
    434 	       ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) {
    435 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
    436 		if (i < af6->ip6af_frglen) {
    437 			af6->ip6af_frglen -= i;
    438 			af6->ip6af_off += i;
    439 			m_adj(IP6_REASS_MBUF(af6), i);
    440 			break;
    441 		}
    442 		af6 = af6->ip6af_down;
    443 		m_freem(IP6_REASS_MBUF(af6->ip6af_up));
    444 		frag6_deq(af6->ip6af_up);
    445 	}
    446 #else
    447 	/*
    448 	 * If the incoming framgent overlaps some existing fragments in
    449 	 * the reassembly queue, drop it, since it is dangerous to override
    450 	 * existing fragments from a security point of view.
    451 	 * We don't know which fragment is the bad guy - here we trust
    452 	 * fragment that came in earlier, with no real reason.
    453 	 */
    454 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
    455 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
    456 			- ip6af->ip6af_off;
    457 		if (i > 0) {
    458 #if 0				/* suppress the noisy log */
    459 			log(LOG_ERR, "%d bytes of a fragment from %s "
    460 			    "overlaps the previous fragment\n",
    461 			    i, ip6_sprintf(&q6->ip6q_src));
    462 #endif
    463 			free(ip6af, M_FTABLE);
    464 			goto dropfrag;
    465 		}
    466 	}
    467 	if (af6 != (struct ip6asfrag *)q6) {
    468 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
    469 		if (i > 0) {
    470 #if 0				/* suppress the noisy log */
    471 			log(LOG_ERR, "%d bytes of a fragment from %s "
    472 			    "overlaps the succeeding fragment",
    473 			    i, ip6_sprintf(&q6->ip6q_src));
    474 #endif
    475 			free(ip6af, M_FTABLE);
    476 			goto dropfrag;
    477 		}
    478 	}
    479 #endif
    480 
    481 insert:
    482 
    483 	/*
    484 	 * Stick new segment in its place;
    485 	 * check for complete reassembly.
    486 	 * Move to front of packet queue, as we are
    487 	 * the most recently active fragmented packet.
    488 	 */
    489 	frag6_enq(ip6af, af6->ip6af_up);
    490 	frag6_nfrags++;
    491 	q6->ip6q_nfrag++;
    492 #if 0 /* xxx */
    493 	if (q6 != ip6q.ip6q_next) {
    494 		frag6_remque(q6);
    495 		frag6_insque(q6, &ip6q);
    496 	}
    497 #endif
    498 	next = 0;
    499 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    500 	     af6 = af6->ip6af_down) {
    501 		if (af6->ip6af_off != next) {
    502 			IP6Q_UNLOCK();
    503 			return IPPROTO_DONE;
    504 		}
    505 		next += af6->ip6af_frglen;
    506 	}
    507 	if (af6->ip6af_up->ip6af_mff) {
    508 		IP6Q_UNLOCK();
    509 		return IPPROTO_DONE;
    510 	}
    511 
    512 	/*
    513 	 * Reassembly is complete; concatenate fragments.
    514 	 */
    515 	ip6af = q6->ip6q_down;
    516 	t = m = IP6_REASS_MBUF(ip6af);
    517 	af6 = ip6af->ip6af_down;
    518 	frag6_deq(ip6af);
    519 	while (af6 != (struct ip6asfrag *)q6) {
    520 		af6dwn = af6->ip6af_down;
    521 		frag6_deq(af6);
    522 		while (t->m_next)
    523 			t = t->m_next;
    524 		t->m_next = IP6_REASS_MBUF(af6);
    525 		m_adj(t->m_next, af6->ip6af_offset);
    526 		free(af6, M_FTABLE);
    527 		af6 = af6dwn;
    528 	}
    529 
    530 	/* adjust offset to point where the original next header starts */
    531 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
    532 	free(ip6af, M_FTABLE);
    533 	ip6 = mtod(m, struct ip6_hdr *);
    534 	ip6->ip6_plen = htons(next + offset - sizeof(struct ip6_hdr));
    535 	ip6->ip6_src = q6->ip6q_src;
    536 	ip6->ip6_dst = q6->ip6q_dst;
    537 	nxt = q6->ip6q_nxt;
    538 #ifdef notyet
    539 	*q6->ip6q_nxtp = (u_char)(nxt & 0xff);
    540 #endif
    541 
    542 	/*
    543 	 * Delete frag6 header with as a few cost as possible.
    544 	 */
    545 	if (offset < m->m_len) {
    546 		ovbcopy((caddr_t)ip6, (caddr_t)ip6 + sizeof(struct ip6_frag),
    547 			offset);
    548 		m->m_data += sizeof(struct ip6_frag);
    549 		m->m_len -= sizeof(struct ip6_frag);
    550 	} else {
    551 		/* this comes with no copy if the boundary is on cluster */
    552 		if ((t = m_split(m, offset, M_DONTWAIT)) == NULL) {
    553 			frag6_remque(q6);
    554 			frag6_nfrags -= q6->ip6q_nfrag;
    555 			free(q6, M_FTABLE);
    556 			frag6_nfragpackets--;
    557 			goto dropfrag;
    558 		}
    559 		m_adj(t, sizeof(struct ip6_frag));
    560 		m_cat(m, t);
    561 	}
    562 
    563 	/*
    564 	 * Store NXT to the original.
    565 	 */
    566 	{
    567 		u_int8_t *prvnxtp = ip6_get_prevhdr(m, offset); /* XXX */
    568 		*prvnxtp = nxt;
    569 	}
    570 
    571 	frag6_remque(q6);
    572 	frag6_nfrags -= q6->ip6q_nfrag;
    573 	free(q6, M_FTABLE);
    574 	frag6_nfragpackets--;
    575 
    576 	if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
    577 		int plen = 0;
    578 		for (t = m; t; t = t->m_next)
    579 			plen += t->m_len;
    580 		m->m_pkthdr.len = plen;
    581 	}
    582 
    583 	ip6stat.ip6s_reassembled++;
    584 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
    585 
    586 	/*
    587 	 * Tell launch routine the next header
    588 	 */
    589 
    590 	*mp = m;
    591 	*offp = offset;
    592 
    593 	IP6Q_UNLOCK();
    594 	return nxt;
    595 
    596  dropfrag:
    597 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
    598 	ip6stat.ip6s_fragdropped++;
    599 	m_freem(m);
    600 	IP6Q_UNLOCK();
    601 	return IPPROTO_DONE;
    602 }
    603 
    604 /*
    605  * Free a fragment reassembly header and all
    606  * associated datagrams.
    607  */
    608 void
    609 frag6_freef(q6)
    610 	struct ip6q *q6;
    611 {
    612 	struct ip6asfrag *af6, *down6;
    613 
    614 	IP6Q_LOCK_CHECK();
    615 
    616 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    617 	     af6 = down6) {
    618 		struct mbuf *m = IP6_REASS_MBUF(af6);
    619 
    620 		down6 = af6->ip6af_down;
    621 		frag6_deq(af6);
    622 
    623 		/*
    624 		 * Return ICMP time exceeded error for the 1st fragment.
    625 		 * Just free other fragments.
    626 		 */
    627 		if (af6->ip6af_off == 0) {
    628 			struct ip6_hdr *ip6;
    629 
    630 			/* adjust pointer */
    631 			ip6 = mtod(m, struct ip6_hdr *);
    632 
    633 			/* restoure source and destination addresses */
    634 			ip6->ip6_src = q6->ip6q_src;
    635 			ip6->ip6_dst = q6->ip6q_dst;
    636 
    637 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
    638 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
    639 		} else
    640 			m_freem(m);
    641 		free(af6, M_FTABLE);
    642 	}
    643 	frag6_remque(q6);
    644 	frag6_nfrags -= q6->ip6q_nfrag;
    645 	free(q6, M_FTABLE);
    646 	frag6_nfragpackets--;
    647 }
    648 
    649 /*
    650  * Put an ip fragment on a reassembly chain.
    651  * Like insque, but pointers in middle of structure.
    652  */
    653 void
    654 frag6_enq(af6, up6)
    655 	struct ip6asfrag *af6, *up6;
    656 {
    657 
    658 	IP6Q_LOCK_CHECK();
    659 
    660 	af6->ip6af_up = up6;
    661 	af6->ip6af_down = up6->ip6af_down;
    662 	up6->ip6af_down->ip6af_up = af6;
    663 	up6->ip6af_down = af6;
    664 }
    665 
    666 /*
    667  * To frag6_enq as remque is to insque.
    668  */
    669 void
    670 frag6_deq(af6)
    671 	struct ip6asfrag *af6;
    672 {
    673 
    674 	IP6Q_LOCK_CHECK();
    675 
    676 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
    677 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
    678 }
    679 
    680 void
    681 frag6_insque(new, old)
    682 	struct ip6q *new, *old;
    683 {
    684 
    685 	IP6Q_LOCK_CHECK();
    686 
    687 	new->ip6q_prev = old;
    688 	new->ip6q_next = old->ip6q_next;
    689 	old->ip6q_next->ip6q_prev= new;
    690 	old->ip6q_next = new;
    691 }
    692 
    693 void
    694 frag6_remque(p6)
    695 	struct ip6q *p6;
    696 {
    697 
    698 	IP6Q_LOCK_CHECK();
    699 
    700 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
    701 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
    702 }
    703 
    704 /*
    705  * IPv6 reassembling timer processing;
    706  * if a timer expires on a reassembly
    707  * queue, discard it.
    708  */
    709 void
    710 frag6_slowtimo()
    711 {
    712 	struct ip6q *q6;
    713 	int s = splsoftnet();
    714 
    715 	IP6Q_LOCK();
    716 	q6 = ip6q.ip6q_next;
    717 	if (q6)
    718 		while (q6 != &ip6q) {
    719 			--q6->ip6q_ttl;
    720 			q6 = q6->ip6q_next;
    721 			if (q6->ip6q_prev->ip6q_ttl == 0) {
    722 				ip6stat.ip6s_fragtimeout++;
    723 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    724 				frag6_freef(q6->ip6q_prev);
    725 			}
    726 		}
    727 	/*
    728 	 * If we are over the maximum number of fragments
    729 	 * (due to the limit being lowered), drain off
    730 	 * enough to get down to the new limit.
    731 	 */
    732 	while (frag6_nfragpackets > (u_int)ip6_maxfragpackets &&
    733 	    ip6q.ip6q_prev) {
    734 		ip6stat.ip6s_fragoverflow++;
    735 		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    736 		frag6_freef(ip6q.ip6q_prev);
    737 	}
    738 	IP6Q_UNLOCK();
    739 
    740 #if 0
    741 	/*
    742 	 * Routing changes might produce a better route than we last used;
    743 	 * make sure we notice eventually, even if forwarding only for one
    744 	 * destination and the cache is never replaced.
    745 	 */
    746 	if (ip6_forward_rt.ro_rt) {
    747 		RTFREE(ip6_forward_rt.ro_rt);
    748 		ip6_forward_rt.ro_rt = 0;
    749 	}
    750 	if (ipsrcchk_rt.ro_rt) {
    751 		RTFREE(ipsrcchk_rt.ro_rt);
    752 		ipsrcchk_rt.ro_rt = 0;
    753 	}
    754 #endif
    755 
    756 	splx(s);
    757 }
    758 
    759 /*
    760  * Drain off all datagram fragments.
    761  */
    762 void
    763 frag6_drain()
    764 {
    765 
    766 	if (ip6q_lock_try() == 0)
    767 		return;
    768 	while (ip6q.ip6q_next != &ip6q) {
    769 		ip6stat.ip6s_fragdropped++;
    770 		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    771 		frag6_freef(ip6q.ip6q_next);
    772 	}
    773 	IP6Q_UNLOCK();
    774 }
    775