Home | History | Annotate | Line # | Download | only in netinet6
      1 /*	$NetBSD: frag6.c,v 1.78 2024/04/19 05:04:06 ozaki-r 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 1.78 2024/04/19 05:04:06 ozaki-r Exp $");
     35 
     36 #ifdef _KERNEL_OPT
     37 #include "opt_net_mpsafe.h"
     38 #endif
     39 
     40 #include <sys/param.h>
     41 #include <sys/systm.h>
     42 #include <sys/mbuf.h>
     43 #include <sys/errno.h>
     44 #include <sys/time.h>
     45 #include <sys/kmem.h>
     46 #include <sys/kernel.h>
     47 #include <sys/syslog.h>
     48 
     49 #include <net/if.h>
     50 #include <net/route.h>
     51 
     52 #include <netinet/in.h>
     53 #include <netinet/in_var.h>
     54 #include <netinet/ip6.h>
     55 #include <netinet6/ip6_var.h>
     56 #include <netinet6/ip6_private.h>
     57 #include <netinet/icmp6.h>
     58 
     59 /*
     60  * IPv6 reassembly queue structure. Each fragment being reassembled is
     61  * attached to one of these structures.
     62  *
     63  * XXX: Would be better to use TAILQ.
     64  */
     65 struct	ip6q {
     66 	u_int32_t	ip6q_head;
     67 	u_int16_t	ip6q_len;
     68 	u_int8_t	ip6q_nxt;	/* ip6f_nxt in first fragment */
     69 	u_int8_t	ip6q_hlim;
     70 	struct ip6asfrag *ip6q_down;
     71 	struct ip6asfrag *ip6q_up;
     72 	u_int32_t	ip6q_ident;
     73 	u_int8_t	ip6q_ttl;
     74 	struct in6_addr	ip6q_src, ip6q_dst;
     75 	struct ip6q	*ip6q_next;
     76 	struct ip6q	*ip6q_prev;
     77 	int		ip6q_unfrglen;	/* len of unfragmentable part */
     78 	int		ip6q_nfrag;	/* # of fragments */
     79 	int		ip6q_ipsec;	/* IPsec flags */
     80 };
     81 
     82 struct	ip6asfrag {
     83 	u_int32_t	ip6af_head;
     84 	u_int16_t	ip6af_len;
     85 	u_int8_t	ip6af_nxt;
     86 	u_int8_t	ip6af_hlim;
     87 	/* must not override the above members during reassembling */
     88 	struct ip6asfrag *ip6af_down;
     89 	struct ip6asfrag *ip6af_up;
     90 	struct mbuf	*ip6af_m;
     91 	int		ip6af_offset;	/* offset in ip6af_m to next header */
     92 	int		ip6af_frglen;	/* fragmentable part length */
     93 	int		ip6af_off;	/* fragment offset */
     94 	bool		ip6af_mff;	/* more fragment bit in frag off */
     95 };
     96 
     97 static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *);
     98 static void frag6_deq(struct ip6asfrag *);
     99 static void frag6_insque(struct ip6q *, struct ip6q *);
    100 static void frag6_remque(struct ip6q *);
    101 static void frag6_freef(struct ip6q *);
    102 
    103 static int frag6_drainwanted;
    104 
    105 static u_int frag6_nfragpackets;
    106 static u_int frag6_nfrags;
    107 static struct ip6q ip6q;	/* ip6 reassembly queue */
    108 
    109 /* Protects ip6q */
    110 static kmutex_t	frag6_lock __cacheline_aligned;
    111 
    112 /*
    113  * Initialise reassembly queue and fragment identifier.
    114  */
    115 void
    116 frag6_init(void)
    117 {
    118 
    119 	ip6q.ip6q_next = ip6q.ip6q_prev = &ip6q;
    120 	mutex_init(&frag6_lock, MUTEX_DEFAULT, IPL_NONE);
    121 }
    122 
    123 static void
    124 frag6_dropfrag(struct ip6q *q6)
    125 {
    126 	frag6_remque(q6);
    127 	frag6_nfrags -= q6->ip6q_nfrag;
    128 	kmem_intr_free(q6, sizeof(*q6));
    129 	frag6_nfragpackets--;
    130 }
    131 
    132 /*
    133  * IPv6 fragment input.
    134  *
    135  * In RFC2460, fragment and reassembly rule do not agree with each other,
    136  * in terms of next header field handling in fragment header.
    137  * While the sender will use the same value for all of the fragmented packets,
    138  * receiver is suggested not to check the consistency.
    139  *
    140  * fragment rule (p20):
    141  *	(2) A Fragment header containing:
    142  *	The Next Header value that identifies the first header of
    143  *	the Fragmentable Part of the original packet.
    144  *		-> next header field is same for all fragments
    145  *
    146  * reassembly rule (p21):
    147  *	The Next Header field of the last header of the Unfragmentable
    148  *	Part is obtained from the Next Header field of the first
    149  *	fragment's Fragment header.
    150  *		-> should grab it from the first fragment only
    151  *
    152  * The following note also contradicts with fragment rule - noone is going to
    153  * send different fragment with different next header field.
    154  *
    155  * additional note (p22):
    156  *	The Next Header values in the Fragment headers of different
    157  *	fragments of the same original packet may differ.  Only the value
    158  *	from the Offset zero fragment packet is used for reassembly.
    159  *		-> should grab it from the first fragment only
    160  *
    161  * There is no explicit reason given in the RFC.  Historical reason maybe?
    162  *
    163  * XXX: It would be better to use a pool, rather than kmem.
    164  */
    165 int
    166 frag6_input(struct mbuf **mp, int *offp, int proto)
    167 {
    168 	struct rtentry *rt;
    169 	struct mbuf *m = *mp, *t;
    170 	struct ip6_hdr *ip6;
    171 	struct ip6_frag *ip6f;
    172 	struct ip6q *q6;
    173 	struct ip6asfrag *af6, *ip6af, *af6dwn;
    174 	int offset = *offp, nxt, i, next;
    175 	int ipsecflags = m->m_flags & (M_DECRYPTED|M_AUTHIPHDR);
    176 	int first_frag = 0;
    177 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
    178 	struct ifnet *dstifp;
    179 	static struct route ro;
    180 	union {
    181 		struct sockaddr		dst;
    182 		struct sockaddr_in6	dst6;
    183 	} u;
    184 
    185 	ip6 = mtod(m, struct ip6_hdr *);
    186 	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
    187 	if (ip6f == NULL)
    188 		return IPPROTO_DONE;
    189 
    190 	dstifp = NULL;
    191 	/* find the destination interface of the packet. */
    192 	sockaddr_in6_init(&u.dst6, &ip6->ip6_dst, 0, 0, 0);
    193 	if ((rt = rtcache_lookup(&ro, &u.dst)) != NULL)
    194 		dstifp = ((struct in6_ifaddr *)rt->rt_ifa)->ia_ifp;
    195 
    196 	/* jumbo payload can't contain a fragment header */
    197 	if (ip6->ip6_plen == 0) {
    198 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
    199 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
    200 		goto done;
    201 	}
    202 
    203 	/*
    204 	 * Check whether fragment packet's fragment length is non-zero and
    205 	 * multiple of 8 octets.
    206 	 * sizeof(struct ip6_frag) == 8
    207 	 * sizeof(struct ip6_hdr) = 40
    208 	 */
    209 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset
    210 	    - sizeof(struct ip6_frag);
    211 	if ((frgpartlen == 0) ||
    212 	    ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) && (frgpartlen & 0x7) != 0)) {
    213 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    214 		    offsetof(struct ip6_hdr, ip6_plen));
    215 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
    216 		goto done;
    217 	}
    218 
    219 	IP6_STATINC(IP6_STAT_FRAGMENTS);
    220 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
    221 
    222 	/* offset now points to data portion */
    223 	offset += sizeof(struct ip6_frag);
    224 
    225 	/*
    226 	 * RFC6946: A host that receives an IPv6 packet which includes
    227 	 * a Fragment Header with the "Fragment Offset" equal to 0 and
    228 	 * the "M" bit equal to 0 MUST process such packet in isolation
    229 	 * from any other packets/fragments.
    230 	 *
    231 	 * XXX: Would be better to remove this fragment header entirely,
    232 	 * for us not to get confused later when looking back at the
    233 	 * previous headers in the chain.
    234 	 */
    235 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
    236 	if (fragoff == 0 && !(ip6f->ip6f_offlg & IP6F_MORE_FRAG)) {
    237 		IP6_STATINC(IP6_STAT_REASSEMBLED);
    238 		in6_ifstat_inc(dstifp, ifs6_reass_ok);
    239 		*offp = offset;
    240 		rtcache_unref(rt, &ro);
    241 		return ip6f->ip6f_nxt;
    242 	}
    243 
    244 	mutex_enter(&frag6_lock);
    245 
    246 	/*
    247 	 * Enforce upper bound on number of fragments.
    248 	 * If maxfrag is 0, never accept fragments.
    249 	 * If maxfrag is -1, accept all fragments without limitation.
    250 	 */
    251 	if (ip6_maxfrags < 0)
    252 		;
    253 	else if (frag6_nfrags >= (u_int)ip6_maxfrags)
    254 		goto dropfrag;
    255 
    256 	for (q6 = ip6q.ip6q_next; q6 != &ip6q; q6 = q6->ip6q_next)
    257 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
    258 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
    259 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst))
    260 			break;
    261 
    262 	if (q6 != &ip6q) {
    263 		/* All fragments must have the same IPsec flags. */
    264 		if (q6->ip6q_ipsec != ipsecflags) {
    265 			goto dropfrag;
    266 		}
    267 	}
    268 
    269 	if (q6 == &ip6q) {
    270 		/*
    271 		 * the first fragment to arrive, create a reassembly queue.
    272 		 */
    273 		first_frag = 1;
    274 
    275 		/*
    276 		 * Enforce upper bound on number of fragmented packets
    277 		 * for which we attempt reassembly;
    278 		 * If maxfragpackets is 0, never accept fragments.
    279 		 * If maxfragpackets is -1, accept all fragments without
    280 		 * limitation.
    281 		 */
    282 		if (ip6_maxfragpackets < 0)
    283 			;
    284 		else if (frag6_nfragpackets >= (u_int)ip6_maxfragpackets)
    285 			goto dropfrag;
    286 		frag6_nfragpackets++;
    287 
    288 		q6 = kmem_intr_zalloc(sizeof(struct ip6q), KM_NOSLEEP);
    289 		if (q6 == NULL) {
    290 			goto dropfrag;
    291 		}
    292 		frag6_insque(q6, &ip6q);
    293 
    294 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
    295 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
    296 		q6->ip6q_ident	= ip6f->ip6f_ident;
    297 		q6->ip6q_ttl 	= IPV6_FRAGTTL;
    298 		q6->ip6q_src	= ip6->ip6_src;
    299 		q6->ip6q_dst	= ip6->ip6_dst;
    300 		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
    301 		q6->ip6q_nfrag = 0;
    302 		q6->ip6q_ipsec = ipsecflags;
    303 	}
    304 
    305 	/*
    306 	 * If it's the 1st fragment, record the length of the
    307 	 * unfragmentable part and the next header of the fragment header.
    308 	 */
    309 	if (fragoff == 0) {
    310 		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
    311 		    sizeof(struct ip6_frag);
    312 		q6->ip6q_nxt = ip6f->ip6f_nxt;
    313 	}
    314 
    315 	/*
    316 	 * Check that the reassembled packet would not exceed 65535 bytes
    317 	 * in size. If it would exceed, discard the fragment and return an
    318 	 * ICMP error.
    319 	 */
    320 	if (q6->ip6q_unfrglen >= 0) {
    321 		/* The 1st fragment has already arrived. */
    322 		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
    323 			mutex_exit(&frag6_lock);
    324 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    325 			    offset - sizeof(struct ip6_frag) +
    326 			    offsetof(struct ip6_frag, ip6f_offlg));
    327 			goto done;
    328 		}
    329 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
    330 		mutex_exit(&frag6_lock);
    331 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    332 		    offset - sizeof(struct ip6_frag) +
    333 		    offsetof(struct ip6_frag, ip6f_offlg));
    334 		goto done;
    335 	}
    336 
    337 	/*
    338 	 * If it's the first fragment, do the above check for each
    339 	 * fragment already stored in the reassembly queue.
    340 	 */
    341 	if (fragoff == 0) {
    342 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    343 		     af6 = af6dwn) {
    344 			af6dwn = af6->ip6af_down;
    345 
    346 			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
    347 			    IPV6_MAXPACKET) {
    348 				struct mbuf *merr = af6->ip6af_m;
    349 				struct ip6_hdr *ip6err;
    350 				int erroff = af6->ip6af_offset;
    351 
    352 				/* dequeue the fragment. */
    353 				frag6_deq(af6);
    354 				kmem_intr_free(af6, sizeof(struct ip6asfrag));
    355 
    356 				/* adjust pointer. */
    357 				ip6err = mtod(merr, struct ip6_hdr *);
    358 
    359 				/*
    360 				 * Restore source and destination addresses
    361 				 * in the erroneous IPv6 header.
    362 				 */
    363 				ip6err->ip6_src = q6->ip6q_src;
    364 				ip6err->ip6_dst = q6->ip6q_dst;
    365 
    366 				icmp6_error(merr, ICMP6_PARAM_PROB,
    367 				    ICMP6_PARAMPROB_HEADER,
    368 				    erroff - sizeof(struct ip6_frag) +
    369 				    offsetof(struct ip6_frag, ip6f_offlg));
    370 			}
    371 		}
    372 	}
    373 
    374 	ip6af = kmem_intr_zalloc(sizeof(struct ip6asfrag), KM_NOSLEEP);
    375 	if (ip6af == NULL) {
    376 		goto dropfrag;
    377 	}
    378 	ip6af->ip6af_head = ip6->ip6_flow;
    379 	ip6af->ip6af_len = ip6->ip6_plen;
    380 	ip6af->ip6af_nxt = ip6->ip6_nxt;
    381 	ip6af->ip6af_hlim = ip6->ip6_hlim;
    382 	ip6af->ip6af_mff = (ip6f->ip6f_offlg & IP6F_MORE_FRAG) != 0;
    383 	ip6af->ip6af_off = fragoff;
    384 	ip6af->ip6af_frglen = frgpartlen;
    385 	ip6af->ip6af_offset = offset;
    386 	ip6af->ip6af_m = m;
    387 
    388 	if (first_frag) {
    389 		af6 = (struct ip6asfrag *)q6;
    390 		goto insert;
    391 	}
    392 
    393 	/*
    394 	 * Find a segment which begins after this one does.
    395 	 */
    396 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    397 	     af6 = af6->ip6af_down)
    398 		if (af6->ip6af_off > ip6af->ip6af_off)
    399 			break;
    400 
    401 	/*
    402 	 * If the incoming fragment overlaps some existing fragments in
    403 	 * the reassembly queue - drop it as per RFC 5722.
    404 	 */
    405 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
    406 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
    407 			- ip6af->ip6af_off;
    408 		if (i > 0) {
    409 			kmem_intr_free(ip6af, sizeof(struct ip6asfrag));
    410 			goto dropfrag;
    411 		}
    412 	}
    413 	if (af6 != (struct ip6asfrag *)q6) {
    414 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
    415 		if (i > 0) {
    416 			kmem_intr_free(ip6af, sizeof(struct ip6asfrag));
    417 			goto dropfrag;
    418 		}
    419 	}
    420 
    421 insert:
    422 	/*
    423 	 * Stick new segment in its place.
    424 	 */
    425 	frag6_enq(ip6af, af6->ip6af_up);
    426 	frag6_nfrags++;
    427 	q6->ip6q_nfrag++;
    428 
    429 	/*
    430 	 * Check for complete reassembly.
    431 	 */
    432 	next = 0;
    433 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    434 	     af6 = af6->ip6af_down) {
    435 		if (af6->ip6af_off != next) {
    436 			mutex_exit(&frag6_lock);
    437 			goto done;
    438 		}
    439 		next += af6->ip6af_frglen;
    440 	}
    441 	if (af6->ip6af_up->ip6af_mff) {
    442 		mutex_exit(&frag6_lock);
    443 		goto done;
    444 	}
    445 
    446 	/*
    447 	 * Reassembly is complete; concatenate fragments.
    448 	 */
    449 	ip6af = q6->ip6q_down;
    450 	t = m = ip6af->ip6af_m;
    451 	af6 = ip6af->ip6af_down;
    452 	frag6_deq(ip6af);
    453 	while (af6 != (struct ip6asfrag *)q6) {
    454 		af6dwn = af6->ip6af_down;
    455 		frag6_deq(af6);
    456 		while (t->m_next)
    457 			t = t->m_next;
    458 		t->m_next = af6->ip6af_m;
    459 		m_adj(t->m_next, af6->ip6af_offset);
    460 		m_remove_pkthdr(t->m_next);
    461 		kmem_intr_free(af6, sizeof(struct ip6asfrag));
    462 		af6 = af6dwn;
    463 	}
    464 
    465 	/* adjust offset to point where the original next header starts */
    466 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
    467 	kmem_intr_free(ip6af, sizeof(struct ip6asfrag));
    468 	next += offset - sizeof(struct ip6_hdr);
    469 	if ((u_int)next > IPV6_MAXPACKET) {
    470 		frag6_dropfrag(q6);
    471 		goto dropfrag;
    472 	}
    473 	ip6 = mtod(m, struct ip6_hdr *);
    474 	ip6->ip6_plen = htons(next);
    475 	ip6->ip6_src = q6->ip6q_src;
    476 	ip6->ip6_dst = q6->ip6q_dst;
    477 	nxt = q6->ip6q_nxt;
    478 
    479 	/*
    480 	 * Delete frag6 header.
    481 	 */
    482 	if (m->m_len >= offset + sizeof(struct ip6_frag)) {
    483 		memmove((char *)ip6 + sizeof(struct ip6_frag), ip6, offset);
    484 		m->m_data += sizeof(struct ip6_frag);
    485 		m->m_len -= sizeof(struct ip6_frag);
    486 	} else {
    487 		/* this comes with no copy if the boundary is on cluster */
    488 		if ((t = m_split(m, offset, M_DONTWAIT)) == NULL) {
    489 			frag6_dropfrag(q6);
    490 			goto dropfrag;
    491 		}
    492 		m_adj(t, sizeof(struct ip6_frag));
    493 		m_cat(m, t);
    494 	}
    495 
    496 	frag6_dropfrag(q6);
    497 
    498 	{
    499 		KASSERT(m->m_flags & M_PKTHDR);
    500 		int plen = 0;
    501 		for (t = m; t; t = t->m_next) {
    502 			plen += t->m_len;
    503 		}
    504 		m->m_pkthdr.len = plen;
    505 		/* XXX XXX: clear csum_flags? */
    506 	}
    507 
    508 	/*
    509 	 * Restore NXT to the original.
    510 	 */
    511 	{
    512 		const int prvnxt = ip6_get_prevhdr(m, offset);
    513 		uint8_t *prvnxtp;
    514 
    515 		IP6_EXTHDR_GET(prvnxtp, uint8_t *, m, prvnxt,
    516 		    sizeof(*prvnxtp));
    517 		if (prvnxtp == NULL) {
    518 			goto dropfrag;
    519 		}
    520 		*prvnxtp = nxt;
    521 	}
    522 
    523 	IP6_STATINC(IP6_STAT_REASSEMBLED);
    524 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
    525 	rtcache_unref(rt, &ro);
    526 	mutex_exit(&frag6_lock);
    527 
    528 	/*
    529 	 * Tell launch routine the next header.
    530 	 */
    531 	*mp = m;
    532 	*offp = offset;
    533 	return nxt;
    534 
    535  dropfrag:
    536 	mutex_exit(&frag6_lock);
    537 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
    538 	IP6_STATINC(IP6_STAT_FRAGDROPPED);
    539 	m_freem(m);
    540  done:
    541 	rtcache_unref(rt, &ro);
    542 	return IPPROTO_DONE;
    543 }
    544 
    545 int
    546 ip6_reass_packet(struct mbuf **mp, int offset)
    547 {
    548 
    549 	if (frag6_input(mp, &offset, IPPROTO_IPV6) == IPPROTO_DONE) {
    550 		*mp = NULL;
    551 		return EINVAL;
    552 	}
    553 	return 0;
    554 }
    555 
    556 /*
    557  * Free a fragment reassembly header and all
    558  * associated datagrams.
    559  */
    560 static void
    561 frag6_freef(struct ip6q *q6)
    562 {
    563 	struct ip6asfrag *af6, *down6;
    564 
    565 	KASSERT(mutex_owned(&frag6_lock));
    566 
    567 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    568 	     af6 = down6) {
    569 		struct mbuf *m = af6->ip6af_m;
    570 
    571 		down6 = af6->ip6af_down;
    572 		frag6_deq(af6);
    573 
    574 		/*
    575 		 * Return ICMP time exceeded error for the 1st fragment.
    576 		 * Just free other fragments.
    577 		 */
    578 		if (af6->ip6af_off == 0) {
    579 			struct ip6_hdr *ip6;
    580 
    581 			/* adjust pointer */
    582 			ip6 = mtod(m, struct ip6_hdr *);
    583 
    584 			/* restore source and destination addresses */
    585 			ip6->ip6_src = q6->ip6q_src;
    586 			ip6->ip6_dst = q6->ip6q_dst;
    587 
    588 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
    589 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
    590 		} else {
    591 			m_freem(m);
    592 		}
    593 		kmem_intr_free(af6, sizeof(struct ip6asfrag));
    594 	}
    595 
    596 	frag6_dropfrag(q6);
    597 }
    598 
    599 /*
    600  * Put an ip fragment on a reassembly chain.
    601  * Like insque, but pointers in middle of structure.
    602  */
    603 void
    604 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6)
    605 {
    606 
    607 	KASSERT(mutex_owned(&frag6_lock));
    608 
    609 	af6->ip6af_up = up6;
    610 	af6->ip6af_down = up6->ip6af_down;
    611 	up6->ip6af_down->ip6af_up = af6;
    612 	up6->ip6af_down = af6;
    613 }
    614 
    615 /*
    616  * To frag6_enq as remque is to insque.
    617  */
    618 void
    619 frag6_deq(struct ip6asfrag *af6)
    620 {
    621 
    622 	KASSERT(mutex_owned(&frag6_lock));
    623 
    624 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
    625 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
    626 }
    627 
    628 /*
    629  * Insert newq after oldq.
    630  */
    631 void
    632 frag6_insque(struct ip6q *newq, struct ip6q *oldq)
    633 {
    634 
    635 	KASSERT(mutex_owned(&frag6_lock));
    636 
    637 	newq->ip6q_prev = oldq;
    638 	newq->ip6q_next = oldq->ip6q_next;
    639 	oldq->ip6q_next->ip6q_prev = newq;
    640 	oldq->ip6q_next = newq;
    641 }
    642 
    643 /*
    644  * Unlink p6.
    645  */
    646 void
    647 frag6_remque(struct ip6q *p6)
    648 {
    649 
    650 	KASSERT(mutex_owned(&frag6_lock));
    651 
    652 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
    653 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
    654 }
    655 
    656 void
    657 frag6_fasttimo(void)
    658 {
    659 
    660 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    661 
    662 	if (frag6_drainwanted) {
    663 		frag6_drain();
    664 		frag6_drainwanted = 0;
    665 	}
    666 
    667 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    668 }
    669 
    670 /*
    671  * IPv6 reassembling timer processing;
    672  * if a timer expires on a reassembly
    673  * queue, discard it.
    674  */
    675 void
    676 frag6_slowtimo(void)
    677 {
    678 	struct ip6q *q6;
    679 
    680 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    681 
    682 	mutex_enter(&frag6_lock);
    683 	q6 = ip6q.ip6q_next;
    684 	if (q6) {
    685 		while (q6 != &ip6q) {
    686 			--q6->ip6q_ttl;
    687 			q6 = q6->ip6q_next;
    688 			if (q6->ip6q_prev->ip6q_ttl == 0) {
    689 				IP6_STATINC(IP6_STAT_FRAGTIMEOUT);
    690 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    691 				frag6_freef(q6->ip6q_prev);
    692 			}
    693 		}
    694 	}
    695 
    696 	/*
    697 	 * If we are over the maximum number of fragments
    698 	 * (due to the limit being lowered), drain off
    699 	 * enough to get down to the new limit.
    700 	 */
    701 	while (frag6_nfragpackets > (u_int)ip6_maxfragpackets &&
    702 	    ip6q.ip6q_prev) {
    703 		IP6_STATINC(IP6_STAT_FRAGOVERFLOW);
    704 		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    705 		frag6_freef(ip6q.ip6q_prev);
    706 	}
    707 	mutex_exit(&frag6_lock);
    708 
    709 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    710 
    711 #if 0
    712 	/*
    713 	 * Routing changes might produce a better route than we last used;
    714 	 * make sure we notice eventually, even if forwarding only for one
    715 	 * destination and the cache is never replaced.
    716 	 */
    717 	rtcache_free(&ip6_forward_rt);
    718 	rtcache_free(&ipsrcchk_rt);
    719 #endif
    720 }
    721 
    722 void
    723 frag6_drainstub(void)
    724 {
    725 	frag6_drainwanted = 1;
    726 }
    727 
    728 /*
    729  * Drain off all datagram fragments.
    730  */
    731 void
    732 frag6_drain(void)
    733 {
    734 
    735 	if (mutex_tryenter(&frag6_lock)) {
    736 		while (ip6q.ip6q_next != &ip6q) {
    737 			IP6_STATINC(IP6_STAT_FRAGDROPPED);
    738 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    739 			frag6_freef(ip6q.ip6q_next);
    740 		}
    741 		mutex_exit(&frag6_lock);
    742 	}
    743 }
    744