Home | History | Annotate | Line # | Download | only in netinet6
frag6.c revision 1.70
      1 /*	$NetBSD: frag6.c,v 1.70 2018/04/13 11:19:09 maxv 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 1.70 2018/04/13 11:19:09 maxv 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 #include <net/net_osdep.h>
     60 
     61 /*
     62  * IP6 reassembly queue structure.  Each fragment
     63  * being reassembled is attached to one of these structures.
     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 };
     80 
     81 struct	ip6asfrag {
     82 	u_int32_t	ip6af_head;
     83 	u_int16_t	ip6af_len;
     84 	u_int8_t	ip6af_nxt;
     85 	u_int8_t	ip6af_hlim;
     86 	/* must not override the above members during reassembling */
     87 	struct ip6asfrag *ip6af_down;
     88 	struct ip6asfrag *ip6af_up;
     89 	struct mbuf	*ip6af_m;
     90 	int		ip6af_offset;	/* offset in ip6af_m to next header */
     91 	int		ip6af_frglen;	/* fragmentable part length */
     92 	int		ip6af_off;	/* fragment offset */
     93 	bool		ip6af_mff;	/* more fragment bit in frag off */
     94 };
     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 u_int frag6_nfragpackets;
    106 u_int frag6_nfrags;
    107 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_NET);
    121 }
    122 
    123 /*
    124  * IPv6 fragment input.
    125  *
    126  * In RFC2460, fragment and reassembly rule do not agree with each other,
    127  * in terms of next header field handling in fragment header.
    128  * While the sender will use the same value for all of the fragmented packets,
    129  * receiver is suggested not to check the consistency.
    130  *
    131  * fragment rule (p20):
    132  *	(2) A Fragment header containing:
    133  *	The Next Header value that identifies the first header of
    134  *	the Fragmentable Part of the original packet.
    135  *		-> next header field is same for all fragments
    136  *
    137  * reassembly rule (p21):
    138  *	The Next Header field of the last header of the Unfragmentable
    139  *	Part is obtained from the Next Header field of the first
    140  *	fragment's Fragment header.
    141  *		-> should grab it from the first fragment only
    142  *
    143  * The following note also contradicts with fragment rule - noone is going to
    144  * send different fragment with different next header field.
    145  *
    146  * additional note (p22):
    147  *	The Next Header values in the Fragment headers of different
    148  *	fragments of the same original packet may differ.  Only the value
    149  *	from the Offset zero fragment packet is used for reassembly.
    150  *		-> should grab it from the first fragment only
    151  *
    152  * There is no explicit reason given in the RFC.  Historical reason maybe?
    153  *
    154  * XXX: It would be better to use a pool, rather than kmem.
    155  */
    156 int
    157 frag6_input(struct mbuf **mp, int *offp, int proto)
    158 {
    159 	struct rtentry *rt;
    160 	struct mbuf *m = *mp, *t;
    161 	struct ip6_hdr *ip6;
    162 	struct ip6_frag *ip6f;
    163 	struct ip6q *q6;
    164 	struct ip6asfrag *af6, *ip6af, *af6dwn;
    165 	int offset = *offp, nxt, i, next;
    166 	int first_frag = 0;
    167 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
    168 	struct ifnet *dstifp;
    169 	static struct route ro;
    170 	union {
    171 		struct sockaddr		dst;
    172 		struct sockaddr_in6	dst6;
    173 	} u;
    174 
    175 	ip6 = mtod(m, struct ip6_hdr *);
    176 	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
    177 	if (ip6f == NULL)
    178 		return IPPROTO_DONE;
    179 
    180 	dstifp = NULL;
    181 	/* find the destination interface of the packet. */
    182 	sockaddr_in6_init(&u.dst6, &ip6->ip6_dst, 0, 0, 0);
    183 	if ((rt = rtcache_lookup(&ro, &u.dst)) != NULL && rt->rt_ifa != NULL)
    184 		dstifp = ((struct in6_ifaddr *)rt->rt_ifa)->ia_ifp;
    185 
    186 	/* jumbo payload can't contain a fragment header */
    187 	if (ip6->ip6_plen == 0) {
    188 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
    189 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
    190 		goto done;
    191 	}
    192 
    193 	/*
    194 	 * Check whether fragment packet's fragment length is non-zero and
    195 	 * multiple of 8 octets.
    196 	 * sizeof(struct ip6_frag) == 8
    197 	 * sizeof(struct ip6_hdr) = 40
    198 	 */
    199 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
    200 	    (((ntohs(ip6->ip6_plen) - offset) == 0) ||
    201 	     ((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
    202 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    203 		    offsetof(struct ip6_hdr, ip6_plen));
    204 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
    205 		goto done;
    206 	}
    207 
    208 	IP6_STATINC(IP6_STAT_FRAGMENTS);
    209 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
    210 
    211 	/* offset now points to data portion */
    212 	offset += sizeof(struct ip6_frag);
    213 
    214 	/*
    215 	 * RFC6946: A host that receives an IPv6 packet which includes
    216 	 * a Fragment Header with the "Fragment Offset" equal to 0 and
    217 	 * the "M" bit equal to 0 MUST process such packet in isolation
    218 	 * from any other packets/fragments.
    219 	 */
    220 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
    221 	if (fragoff == 0 && !(ip6f->ip6f_offlg & IP6F_MORE_FRAG)) {
    222 		IP6_STATINC(IP6_STAT_REASSEMBLED);
    223 		in6_ifstat_inc(dstifp, ifs6_reass_ok);
    224 		*offp = offset;
    225 		rtcache_unref(rt, &ro);
    226 		return ip6f->ip6f_nxt;
    227 	}
    228 
    229 	mutex_enter(&frag6_lock);
    230 
    231 	/*
    232 	 * Enforce upper bound on number of fragments.
    233 	 * If maxfrag is 0, never accept fragments.
    234 	 * If maxfrag is -1, accept all fragments without limitation.
    235 	 */
    236 	if (ip6_maxfrags < 0)
    237 		;
    238 	else if (frag6_nfrags >= (u_int)ip6_maxfrags)
    239 		goto dropfrag;
    240 
    241 	for (q6 = ip6q.ip6q_next; q6 != &ip6q; q6 = q6->ip6q_next)
    242 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
    243 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
    244 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst))
    245 			break;
    246 
    247 	if (q6 == &ip6q) {
    248 		/*
    249 		 * the first fragment to arrive, create a reassembly queue.
    250 		 */
    251 		first_frag = 1;
    252 
    253 		/*
    254 		 * Enforce upper bound on number of fragmented packets
    255 		 * for which we attempt reassembly;
    256 		 * If maxfragpackets is 0, never accept fragments.
    257 		 * If maxfragpackets is -1, accept all fragments without
    258 		 * limitation.
    259 		 */
    260 		if (ip6_maxfragpackets < 0)
    261 			;
    262 		else if (frag6_nfragpackets >= (u_int)ip6_maxfragpackets)
    263 			goto dropfrag;
    264 		frag6_nfragpackets++;
    265 
    266 		q6 = kmem_intr_zalloc(sizeof(struct ip6q), KM_NOSLEEP);
    267 		if (q6 == NULL) {
    268 			goto dropfrag;
    269 		}
    270 		frag6_insque(q6, &ip6q);
    271 
    272 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
    273 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
    274 		q6->ip6q_ident	= ip6f->ip6f_ident;
    275 		q6->ip6q_ttl 	= IPV6_FRAGTTL;
    276 		q6->ip6q_src	= ip6->ip6_src;
    277 		q6->ip6q_dst	= ip6->ip6_dst;
    278 		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
    279 
    280 		q6->ip6q_nfrag = 0;
    281 	}
    282 
    283 	/*
    284 	 * If it's the 1st fragment, record the length of the
    285 	 * unfragmentable part and the next header of the fragment header.
    286 	 */
    287 	if (fragoff == 0) {
    288 		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
    289 		    sizeof(struct ip6_frag);
    290 		q6->ip6q_nxt = ip6f->ip6f_nxt;
    291 	}
    292 
    293 	/*
    294 	 * Check that the reassembled packet would not exceed 65535 bytes
    295 	 * in size. If it would exceed, discard the fragment and return an
    296 	 * ICMP error.
    297 	 */
    298 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
    299 	if (q6->ip6q_unfrglen >= 0) {
    300 		/* The 1st fragment has already arrived. */
    301 		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
    302 			mutex_exit(&frag6_lock);
    303 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    304 			    offset - sizeof(struct ip6_frag) +
    305 			    offsetof(struct ip6_frag, ip6f_offlg));
    306 			goto done;
    307 		}
    308 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
    309 		mutex_exit(&frag6_lock);
    310 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    311 		    offset - sizeof(struct ip6_frag) +
    312 		    offsetof(struct ip6_frag, ip6f_offlg));
    313 		goto done;
    314 	}
    315 
    316 	/*
    317 	 * If it's the first fragment, do the above check for each
    318 	 * fragment already stored in the reassembly queue.
    319 	 */
    320 	if (fragoff == 0) {
    321 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    322 		     af6 = af6dwn) {
    323 			af6dwn = af6->ip6af_down;
    324 
    325 			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
    326 			    IPV6_MAXPACKET) {
    327 				struct mbuf *merr = af6->ip6af_m;
    328 				struct ip6_hdr *ip6err;
    329 				int erroff = af6->ip6af_offset;
    330 
    331 				/* dequeue the fragment. */
    332 				frag6_deq(af6);
    333 				kmem_intr_free(af6, sizeof(struct ip6asfrag));
    334 
    335 				/* adjust pointer. */
    336 				ip6err = mtod(merr, struct ip6_hdr *);
    337 
    338 				/*
    339 				 * Restore source and destination addresses
    340 				 * in the erroneous IPv6 header.
    341 				 */
    342 				ip6err->ip6_src = q6->ip6q_src;
    343 				ip6err->ip6_dst = q6->ip6q_dst;
    344 
    345 				icmp6_error(merr, ICMP6_PARAM_PROB,
    346 				    ICMP6_PARAMPROB_HEADER,
    347 				    erroff - sizeof(struct ip6_frag) +
    348 				    offsetof(struct ip6_frag, ip6f_offlg));
    349 			}
    350 		}
    351 	}
    352 
    353 	ip6af = kmem_intr_zalloc(sizeof(struct ip6asfrag), KM_NOSLEEP);
    354 	if (ip6af == NULL) {
    355 		goto dropfrag;
    356 	}
    357 	ip6af->ip6af_head = ip6->ip6_flow;
    358 	ip6af->ip6af_len = ip6->ip6_plen;
    359 	ip6af->ip6af_nxt = ip6->ip6_nxt;
    360 	ip6af->ip6af_hlim = ip6->ip6_hlim;
    361 	ip6af->ip6af_mff = (ip6f->ip6f_offlg & IP6F_MORE_FRAG) != 0;
    362 	ip6af->ip6af_off = fragoff;
    363 	ip6af->ip6af_frglen = frgpartlen;
    364 	ip6af->ip6af_offset = offset;
    365 	ip6af->ip6af_m = m;
    366 
    367 	if (first_frag) {
    368 		af6 = (struct ip6asfrag *)q6;
    369 		goto insert;
    370 	}
    371 
    372 	/*
    373 	 * Find a segment which begins after this one does.
    374 	 */
    375 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    376 	     af6 = af6->ip6af_down)
    377 		if (af6->ip6af_off > ip6af->ip6af_off)
    378 			break;
    379 
    380 	/*
    381 	 * If the incoming fragment overlaps some existing fragments in
    382 	 * the reassembly queue - drop it as per RFC 5722.
    383 	 */
    384 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
    385 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
    386 			- ip6af->ip6af_off;
    387 		if (i > 0) {
    388 			kmem_intr_free(ip6af, sizeof(struct ip6asfrag));
    389 			goto dropfrag;
    390 		}
    391 	}
    392 	if (af6 != (struct ip6asfrag *)q6) {
    393 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
    394 		if (i > 0) {
    395 			kmem_intr_free(ip6af, sizeof(struct ip6asfrag));
    396 			goto dropfrag;
    397 		}
    398 	}
    399 
    400 insert:
    401 	/*
    402 	 * Stick new segment in its place.
    403 	 */
    404 	frag6_enq(ip6af, af6->ip6af_up);
    405 	frag6_nfrags++;
    406 	q6->ip6q_nfrag++;
    407 
    408 	/*
    409 	 * Check for complete reassembly.
    410 	 */
    411 	next = 0;
    412 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    413 	     af6 = af6->ip6af_down) {
    414 		if (af6->ip6af_off != next) {
    415 			mutex_exit(&frag6_lock);
    416 			goto done;
    417 		}
    418 		next += af6->ip6af_frglen;
    419 	}
    420 	if (af6->ip6af_up->ip6af_mff) {
    421 		mutex_exit(&frag6_lock);
    422 		goto done;
    423 	}
    424 
    425 	/*
    426 	 * Reassembly is complete; concatenate fragments.
    427 	 */
    428 	ip6af = q6->ip6q_down;
    429 	t = m = ip6af->ip6af_m;
    430 	af6 = ip6af->ip6af_down;
    431 	frag6_deq(ip6af);
    432 	while (af6 != (struct ip6asfrag *)q6) {
    433 		af6dwn = af6->ip6af_down;
    434 		frag6_deq(af6);
    435 		while (t->m_next)
    436 			t = t->m_next;
    437 		t->m_next = af6->ip6af_m;
    438 		m_adj(t->m_next, af6->ip6af_offset);
    439 		m_pkthdr_remove(t->m_next);
    440 		kmem_intr_free(af6, sizeof(struct ip6asfrag));
    441 		af6 = af6dwn;
    442 	}
    443 
    444 	/* adjust offset to point where the original next header starts */
    445 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
    446 	kmem_intr_free(ip6af, sizeof(struct ip6asfrag));
    447 	ip6 = mtod(m, struct ip6_hdr *);
    448 	ip6->ip6_plen = htons(next + offset - sizeof(struct ip6_hdr));
    449 	ip6->ip6_src = q6->ip6q_src;
    450 	ip6->ip6_dst = q6->ip6q_dst;
    451 	nxt = q6->ip6q_nxt;
    452 
    453 	/*
    454 	 * Delete frag6 header.
    455 	 */
    456 	if (m->m_len >= offset + sizeof(struct ip6_frag)) {
    457 		memmove((char *)ip6 + sizeof(struct ip6_frag), ip6, offset);
    458 		m->m_data += sizeof(struct ip6_frag);
    459 		m->m_len -= sizeof(struct ip6_frag);
    460 	} else {
    461 		/* this comes with no copy if the boundary is on cluster */
    462 		if ((t = m_split(m, offset, M_DONTWAIT)) == NULL) {
    463 			frag6_remque(q6);
    464 			frag6_nfrags -= q6->ip6q_nfrag;
    465 			kmem_intr_free(q6, sizeof(struct ip6q));
    466 			frag6_nfragpackets--;
    467 			goto dropfrag;
    468 		}
    469 		m_adj(t, sizeof(struct ip6_frag));
    470 		m_cat(m, t);
    471 	}
    472 
    473 	frag6_remque(q6);
    474 	frag6_nfrags -= q6->ip6q_nfrag;
    475 	kmem_intr_free(q6, sizeof(struct ip6q));
    476 	frag6_nfragpackets--;
    477 
    478 	{
    479 		KASSERT(m->m_flags & M_PKTHDR);
    480 		int plen = 0;
    481 		for (t = m; t; t = t->m_next) {
    482 			plen += t->m_len;
    483 		}
    484 		m->m_pkthdr.len = plen;
    485 		/* XXX XXX: clear csum_flags? */
    486 	}
    487 
    488 	/*
    489 	 * Restore NXT to the original.
    490 	 */
    491 	{
    492 		const int prvnxt = ip6_get_prevhdr(m, offset);
    493 		uint8_t *prvnxtp;
    494 
    495 		IP6_EXTHDR_GET(prvnxtp, uint8_t *, m, prvnxt,
    496 		    sizeof(*prvnxtp));
    497 		if (prvnxtp == NULL) {
    498 			goto dropfrag;
    499 		}
    500 		*prvnxtp = nxt;
    501 	}
    502 
    503 	IP6_STATINC(IP6_STAT_REASSEMBLED);
    504 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
    505 	rtcache_unref(rt, &ro);
    506 	mutex_exit(&frag6_lock);
    507 
    508 	/*
    509 	 * Tell launch routine the next header.
    510 	 */
    511 	*mp = m;
    512 	*offp = offset;
    513 	return nxt;
    514 
    515  dropfrag:
    516 	mutex_exit(&frag6_lock);
    517 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
    518 	IP6_STATINC(IP6_STAT_FRAGDROPPED);
    519 	m_freem(m);
    520  done:
    521 	rtcache_unref(rt, &ro);
    522 	return IPPROTO_DONE;
    523 }
    524 
    525 int
    526 ip6_reass_packet(struct mbuf **mp, int offset)
    527 {
    528 
    529 	if (frag6_input(mp, &offset, IPPROTO_IPV6) == IPPROTO_DONE) {
    530 		*mp = NULL;
    531 		return EINVAL;
    532 	}
    533 	return 0;
    534 }
    535 
    536 /*
    537  * Free a fragment reassembly header and all
    538  * associated datagrams.
    539  */
    540 static void
    541 frag6_freef(struct ip6q *q6)
    542 {
    543 	struct ip6asfrag *af6, *down6;
    544 
    545 	KASSERT(mutex_owned(&frag6_lock));
    546 
    547 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    548 	     af6 = down6) {
    549 		struct mbuf *m = af6->ip6af_m;
    550 
    551 		down6 = af6->ip6af_down;
    552 		frag6_deq(af6);
    553 
    554 		/*
    555 		 * Return ICMP time exceeded error for the 1st fragment.
    556 		 * Just free other fragments.
    557 		 */
    558 		if (af6->ip6af_off == 0) {
    559 			struct ip6_hdr *ip6;
    560 
    561 			/* adjust pointer */
    562 			ip6 = mtod(m, struct ip6_hdr *);
    563 
    564 			/* restore source and destination addresses */
    565 			ip6->ip6_src = q6->ip6q_src;
    566 			ip6->ip6_dst = q6->ip6q_dst;
    567 
    568 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
    569 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
    570 		} else {
    571 			m_freem(m);
    572 		}
    573 		kmem_intr_free(af6, sizeof(struct ip6asfrag));
    574 	}
    575 
    576 	frag6_remque(q6);
    577 	frag6_nfrags -= q6->ip6q_nfrag;
    578 	kmem_intr_free(q6, sizeof(struct ip6q));
    579 	frag6_nfragpackets--;
    580 }
    581 
    582 /*
    583  * Put an ip fragment on a reassembly chain.
    584  * Like insque, but pointers in middle of structure.
    585  */
    586 void
    587 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6)
    588 {
    589 
    590 	KASSERT(mutex_owned(&frag6_lock));
    591 
    592 	af6->ip6af_up = up6;
    593 	af6->ip6af_down = up6->ip6af_down;
    594 	up6->ip6af_down->ip6af_up = af6;
    595 	up6->ip6af_down = af6;
    596 }
    597 
    598 /*
    599  * To frag6_enq as remque is to insque.
    600  */
    601 void
    602 frag6_deq(struct ip6asfrag *af6)
    603 {
    604 
    605 	KASSERT(mutex_owned(&frag6_lock));
    606 
    607 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
    608 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
    609 }
    610 
    611 /*
    612  * Insert newq after oldq.
    613  */
    614 void
    615 frag6_insque(struct ip6q *newq, struct ip6q *oldq)
    616 {
    617 
    618 	KASSERT(mutex_owned(&frag6_lock));
    619 
    620 	newq->ip6q_prev = oldq;
    621 	newq->ip6q_next = oldq->ip6q_next;
    622 	oldq->ip6q_next->ip6q_prev = newq;
    623 	oldq->ip6q_next = newq;
    624 }
    625 
    626 /*
    627  * Unlink p6.
    628  */
    629 void
    630 frag6_remque(struct ip6q *p6)
    631 {
    632 
    633 	KASSERT(mutex_owned(&frag6_lock));
    634 
    635 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
    636 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
    637 }
    638 
    639 void
    640 frag6_fasttimo(void)
    641 {
    642 
    643 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    644 
    645 	if (frag6_drainwanted) {
    646 		frag6_drain();
    647 		frag6_drainwanted = 0;
    648 	}
    649 
    650 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    651 }
    652 
    653 /*
    654  * IPv6 reassembling timer processing;
    655  * if a timer expires on a reassembly
    656  * queue, discard it.
    657  */
    658 void
    659 frag6_slowtimo(void)
    660 {
    661 	struct ip6q *q6;
    662 
    663 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    664 
    665 	mutex_enter(&frag6_lock);
    666 	q6 = ip6q.ip6q_next;
    667 	if (q6)
    668 		while (q6 != &ip6q) {
    669 			--q6->ip6q_ttl;
    670 			q6 = q6->ip6q_next;
    671 			if (q6->ip6q_prev->ip6q_ttl == 0) {
    672 				IP6_STATINC(IP6_STAT_FRAGTIMEOUT);
    673 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    674 				frag6_freef(q6->ip6q_prev);
    675 			}
    676 		}
    677 	/*
    678 	 * If we are over the maximum number of fragments
    679 	 * (due to the limit being lowered), drain off
    680 	 * enough to get down to the new limit.
    681 	 */
    682 	while (frag6_nfragpackets > (u_int)ip6_maxfragpackets &&
    683 	    ip6q.ip6q_prev) {
    684 		IP6_STATINC(IP6_STAT_FRAGOVERFLOW);
    685 		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    686 		frag6_freef(ip6q.ip6q_prev);
    687 	}
    688 	mutex_exit(&frag6_lock);
    689 
    690 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    691 
    692 #if 0
    693 	/*
    694 	 * Routing changes might produce a better route than we last used;
    695 	 * make sure we notice eventually, even if forwarding only for one
    696 	 * destination and the cache is never replaced.
    697 	 */
    698 	rtcache_free(&ip6_forward_rt);
    699 	rtcache_free(&ipsrcchk_rt);
    700 #endif
    701 
    702 }
    703 
    704 void
    705 frag6_drainstub(void)
    706 {
    707 	frag6_drainwanted = 1;
    708 }
    709 
    710 /*
    711  * Drain off all datagram fragments.
    712  */
    713 void
    714 frag6_drain(void)
    715 {
    716 
    717 	if (mutex_tryenter(&frag6_lock)) {
    718 		while (ip6q.ip6q_next != &ip6q) {
    719 			IP6_STATINC(IP6_STAT_FRAGDROPPED);
    720 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    721 			frag6_freef(ip6q.ip6q_next);
    722 		}
    723 		mutex_exit(&frag6_lock);
    724 	}
    725 }
    726