Home | History | Annotate | Line # | Download | only in netinet6
frag6.c revision 1.73
      1 /*	$NetBSD: frag6.c,v 1.73 2018/05/03 07:25:49 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 1.73 2018/05/03 07:25:49 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 /*
     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 };
     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 static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *);
     97 static void frag6_deq(struct ip6asfrag *);
     98 static void frag6_insque(struct ip6q *, struct ip6q *);
     99 static void frag6_remque(struct ip6q *);
    100 static void frag6_freef(struct ip6q *);
    101 
    102 static int frag6_drainwanted;
    103 
    104 static u_int frag6_nfragpackets;
    105 static u_int frag6_nfrags;
    106 static struct ip6q ip6q;	/* ip6 reassembly queue */
    107 
    108 /* Protects ip6q */
    109 static kmutex_t	frag6_lock __cacheline_aligned;
    110 
    111 /*
    112  * Initialise reassembly queue and fragment identifier.
    113  */
    114 void
    115 frag6_init(void)
    116 {
    117 
    118 	ip6q.ip6q_next = ip6q.ip6q_prev = &ip6q;
    119 	mutex_init(&frag6_lock, MUTEX_DEFAULT, IPL_NET);
    120 }
    121 
    122 /*
    123  * IPv6 fragment input.
    124  *
    125  * In RFC2460, fragment and reassembly rule do not agree with each other,
    126  * in terms of next header field handling in fragment header.
    127  * While the sender will use the same value for all of the fragmented packets,
    128  * receiver is suggested not to check the consistency.
    129  *
    130  * fragment rule (p20):
    131  *	(2) A Fragment header containing:
    132  *	The Next Header value that identifies the first header of
    133  *	the Fragmentable Part of the original packet.
    134  *		-> next header field is same for all fragments
    135  *
    136  * reassembly rule (p21):
    137  *	The Next Header field of the last header of the Unfragmentable
    138  *	Part is obtained from the Next Header field of the first
    139  *	fragment's Fragment header.
    140  *		-> should grab it from the first fragment only
    141  *
    142  * The following note also contradicts with fragment rule - noone is going to
    143  * send different fragment with different next header field.
    144  *
    145  * additional note (p22):
    146  *	The Next Header values in the Fragment headers of different
    147  *	fragments of the same original packet may differ.  Only the value
    148  *	from the Offset zero fragment packet is used for reassembly.
    149  *		-> should grab it from the first fragment only
    150  *
    151  * There is no explicit reason given in the RFC.  Historical reason maybe?
    152  *
    153  * XXX: It would be better to use a pool, rather than kmem.
    154  */
    155 int
    156 frag6_input(struct mbuf **mp, int *offp, int proto)
    157 {
    158 	struct rtentry *rt;
    159 	struct mbuf *m = *mp, *t;
    160 	struct ip6_hdr *ip6;
    161 	struct ip6_frag *ip6f;
    162 	struct ip6q *q6;
    163 	struct ip6asfrag *af6, *ip6af, *af6dwn;
    164 	int offset = *offp, nxt, i, next;
    165 	int first_frag = 0;
    166 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
    167 	struct ifnet *dstifp;
    168 	static struct route ro;
    169 	union {
    170 		struct sockaddr		dst;
    171 		struct sockaddr_in6	dst6;
    172 	} u;
    173 
    174 	ip6 = mtod(m, struct ip6_hdr *);
    175 	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
    176 	if (ip6f == NULL)
    177 		return IPPROTO_DONE;
    178 
    179 	dstifp = NULL;
    180 	/* find the destination interface of the packet. */
    181 	sockaddr_in6_init(&u.dst6, &ip6->ip6_dst, 0, 0, 0);
    182 	if ((rt = rtcache_lookup(&ro, &u.dst)) != NULL && rt->rt_ifa != NULL)
    183 		dstifp = ((struct in6_ifaddr *)rt->rt_ifa)->ia_ifp;
    184 
    185 	/* jumbo payload can't contain a fragment header */
    186 	if (ip6->ip6_plen == 0) {
    187 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
    188 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
    189 		goto done;
    190 	}
    191 
    192 	/*
    193 	 * Check whether fragment packet's fragment length is non-zero and
    194 	 * multiple of 8 octets.
    195 	 * sizeof(struct ip6_frag) == 8
    196 	 * sizeof(struct ip6_hdr) = 40
    197 	 */
    198 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
    199 	    (((ntohs(ip6->ip6_plen) - offset) == 0) ||
    200 	     ((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
    201 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    202 		    offsetof(struct ip6_hdr, ip6_plen));
    203 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
    204 		goto done;
    205 	}
    206 
    207 	IP6_STATINC(IP6_STAT_FRAGMENTS);
    208 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
    209 
    210 	/* offset now points to data portion */
    211 	offset += sizeof(struct ip6_frag);
    212 
    213 	/*
    214 	 * RFC6946: A host that receives an IPv6 packet which includes
    215 	 * a Fragment Header with the "Fragment Offset" equal to 0 and
    216 	 * the "M" bit equal to 0 MUST process such packet in isolation
    217 	 * from any other packets/fragments.
    218 	 *
    219 	 * XXX: Would be better to remove this fragment header entirely,
    220 	 * for us not to get confused later when looking back at the
    221 	 * previous headers in the chain.
    222 	 */
    223 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
    224 	if (fragoff == 0 && !(ip6f->ip6f_offlg & IP6F_MORE_FRAG)) {
    225 		IP6_STATINC(IP6_STAT_REASSEMBLED);
    226 		in6_ifstat_inc(dstifp, ifs6_reass_ok);
    227 		*offp = offset;
    228 		rtcache_unref(rt, &ro);
    229 		return ip6f->ip6f_nxt;
    230 	}
    231 
    232 	mutex_enter(&frag6_lock);
    233 
    234 	/*
    235 	 * Enforce upper bound on number of fragments.
    236 	 * If maxfrag is 0, never accept fragments.
    237 	 * If maxfrag is -1, accept all fragments without limitation.
    238 	 */
    239 	if (ip6_maxfrags < 0)
    240 		;
    241 	else if (frag6_nfrags >= (u_int)ip6_maxfrags)
    242 		goto dropfrag;
    243 
    244 	for (q6 = ip6q.ip6q_next; q6 != &ip6q; q6 = q6->ip6q_next)
    245 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
    246 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
    247 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst))
    248 			break;
    249 
    250 	if (q6 == &ip6q) {
    251 		/*
    252 		 * the first fragment to arrive, create a reassembly queue.
    253 		 */
    254 		first_frag = 1;
    255 
    256 		/*
    257 		 * Enforce upper bound on number of fragmented packets
    258 		 * for which we attempt reassembly;
    259 		 * If maxfragpackets is 0, never accept fragments.
    260 		 * If maxfragpackets is -1, accept all fragments without
    261 		 * limitation.
    262 		 */
    263 		if (ip6_maxfragpackets < 0)
    264 			;
    265 		else if (frag6_nfragpackets >= (u_int)ip6_maxfragpackets)
    266 			goto dropfrag;
    267 		frag6_nfragpackets++;
    268 
    269 		q6 = kmem_intr_zalloc(sizeof(struct ip6q), KM_NOSLEEP);
    270 		if (q6 == NULL) {
    271 			goto dropfrag;
    272 		}
    273 		frag6_insque(q6, &ip6q);
    274 
    275 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
    276 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
    277 		q6->ip6q_ident	= ip6f->ip6f_ident;
    278 		q6->ip6q_ttl 	= IPV6_FRAGTTL;
    279 		q6->ip6q_src	= ip6->ip6_src;
    280 		q6->ip6q_dst	= ip6->ip6_dst;
    281 		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
    282 
    283 		q6->ip6q_nfrag = 0;
    284 	}
    285 
    286 	/*
    287 	 * If it's the 1st fragment, record the length of the
    288 	 * unfragmentable part and the next header of the fragment header.
    289 	 */
    290 	if (fragoff == 0) {
    291 		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
    292 		    sizeof(struct ip6_frag);
    293 		q6->ip6q_nxt = ip6f->ip6f_nxt;
    294 	}
    295 
    296 	/*
    297 	 * Check that the reassembled packet would not exceed 65535 bytes
    298 	 * in size. If it would exceed, discard the fragment and return an
    299 	 * ICMP error.
    300 	 */
    301 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
    302 	if (q6->ip6q_unfrglen >= 0) {
    303 		/* The 1st fragment has already arrived. */
    304 		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
    305 			mutex_exit(&frag6_lock);
    306 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    307 			    offset - sizeof(struct ip6_frag) +
    308 			    offsetof(struct ip6_frag, ip6f_offlg));
    309 			goto done;
    310 		}
    311 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
    312 		mutex_exit(&frag6_lock);
    313 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
    314 		    offset - sizeof(struct ip6_frag) +
    315 		    offsetof(struct ip6_frag, ip6f_offlg));
    316 		goto done;
    317 	}
    318 
    319 	/*
    320 	 * If it's the first fragment, do the above check for each
    321 	 * fragment already stored in the reassembly queue.
    322 	 */
    323 	if (fragoff == 0) {
    324 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    325 		     af6 = af6dwn) {
    326 			af6dwn = af6->ip6af_down;
    327 
    328 			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
    329 			    IPV6_MAXPACKET) {
    330 				struct mbuf *merr = af6->ip6af_m;
    331 				struct ip6_hdr *ip6err;
    332 				int erroff = af6->ip6af_offset;
    333 
    334 				/* dequeue the fragment. */
    335 				frag6_deq(af6);
    336 				kmem_intr_free(af6, sizeof(struct ip6asfrag));
    337 
    338 				/* adjust pointer. */
    339 				ip6err = mtod(merr, struct ip6_hdr *);
    340 
    341 				/*
    342 				 * Restore source and destination addresses
    343 				 * in the erroneous IPv6 header.
    344 				 */
    345 				ip6err->ip6_src = q6->ip6q_src;
    346 				ip6err->ip6_dst = q6->ip6q_dst;
    347 
    348 				icmp6_error(merr, ICMP6_PARAM_PROB,
    349 				    ICMP6_PARAMPROB_HEADER,
    350 				    erroff - sizeof(struct ip6_frag) +
    351 				    offsetof(struct ip6_frag, ip6f_offlg));
    352 			}
    353 		}
    354 	}
    355 
    356 	ip6af = kmem_intr_zalloc(sizeof(struct ip6asfrag), KM_NOSLEEP);
    357 	if (ip6af == NULL) {
    358 		goto dropfrag;
    359 	}
    360 	ip6af->ip6af_head = ip6->ip6_flow;
    361 	ip6af->ip6af_len = ip6->ip6_plen;
    362 	ip6af->ip6af_nxt = ip6->ip6_nxt;
    363 	ip6af->ip6af_hlim = ip6->ip6_hlim;
    364 	ip6af->ip6af_mff = (ip6f->ip6f_offlg & IP6F_MORE_FRAG) != 0;
    365 	ip6af->ip6af_off = fragoff;
    366 	ip6af->ip6af_frglen = frgpartlen;
    367 	ip6af->ip6af_offset = offset;
    368 	ip6af->ip6af_m = m;
    369 
    370 	if (first_frag) {
    371 		af6 = (struct ip6asfrag *)q6;
    372 		goto insert;
    373 	}
    374 
    375 	/*
    376 	 * Find a segment which begins after this one does.
    377 	 */
    378 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    379 	     af6 = af6->ip6af_down)
    380 		if (af6->ip6af_off > ip6af->ip6af_off)
    381 			break;
    382 
    383 	/*
    384 	 * If the incoming fragment overlaps some existing fragments in
    385 	 * the reassembly queue - drop it as per RFC 5722.
    386 	 */
    387 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
    388 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
    389 			- ip6af->ip6af_off;
    390 		if (i > 0) {
    391 			kmem_intr_free(ip6af, sizeof(struct ip6asfrag));
    392 			goto dropfrag;
    393 		}
    394 	}
    395 	if (af6 != (struct ip6asfrag *)q6) {
    396 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
    397 		if (i > 0) {
    398 			kmem_intr_free(ip6af, sizeof(struct ip6asfrag));
    399 			goto dropfrag;
    400 		}
    401 	}
    402 
    403 insert:
    404 	/*
    405 	 * Stick new segment in its place.
    406 	 */
    407 	frag6_enq(ip6af, af6->ip6af_up);
    408 	frag6_nfrags++;
    409 	q6->ip6q_nfrag++;
    410 
    411 	/*
    412 	 * Check for complete reassembly.
    413 	 */
    414 	next = 0;
    415 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    416 	     af6 = af6->ip6af_down) {
    417 		if (af6->ip6af_off != next) {
    418 			mutex_exit(&frag6_lock);
    419 			goto done;
    420 		}
    421 		next += af6->ip6af_frglen;
    422 	}
    423 	if (af6->ip6af_up->ip6af_mff) {
    424 		mutex_exit(&frag6_lock);
    425 		goto done;
    426 	}
    427 
    428 	/*
    429 	 * Reassembly is complete; concatenate fragments.
    430 	 */
    431 	ip6af = q6->ip6q_down;
    432 	t = m = ip6af->ip6af_m;
    433 	af6 = ip6af->ip6af_down;
    434 	frag6_deq(ip6af);
    435 	while (af6 != (struct ip6asfrag *)q6) {
    436 		af6dwn = af6->ip6af_down;
    437 		frag6_deq(af6);
    438 		while (t->m_next)
    439 			t = t->m_next;
    440 		t->m_next = af6->ip6af_m;
    441 		m_adj(t->m_next, af6->ip6af_offset);
    442 		m_remove_pkthdr(t->m_next);
    443 		kmem_intr_free(af6, sizeof(struct ip6asfrag));
    444 		af6 = af6dwn;
    445 	}
    446 
    447 	/* adjust offset to point where the original next header starts */
    448 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
    449 	kmem_intr_free(ip6af, sizeof(struct ip6asfrag));
    450 	ip6 = mtod(m, struct ip6_hdr *);
    451 	ip6->ip6_plen = htons(next + offset - sizeof(struct ip6_hdr));
    452 	ip6->ip6_src = q6->ip6q_src;
    453 	ip6->ip6_dst = q6->ip6q_dst;
    454 	nxt = q6->ip6q_nxt;
    455 
    456 	/*
    457 	 * Delete frag6 header.
    458 	 */
    459 	if (m->m_len >= offset + sizeof(struct ip6_frag)) {
    460 		memmove((char *)ip6 + sizeof(struct ip6_frag), ip6, offset);
    461 		m->m_data += sizeof(struct ip6_frag);
    462 		m->m_len -= sizeof(struct ip6_frag);
    463 	} else {
    464 		/* this comes with no copy if the boundary is on cluster */
    465 		if ((t = m_split(m, offset, M_DONTWAIT)) == NULL) {
    466 			frag6_remque(q6);
    467 			frag6_nfrags -= q6->ip6q_nfrag;
    468 			kmem_intr_free(q6, sizeof(struct ip6q));
    469 			frag6_nfragpackets--;
    470 			goto dropfrag;
    471 		}
    472 		m_adj(t, sizeof(struct ip6_frag));
    473 		m_cat(m, t);
    474 	}
    475 
    476 	frag6_remque(q6);
    477 	frag6_nfrags -= q6->ip6q_nfrag;
    478 	kmem_intr_free(q6, sizeof(struct ip6q));
    479 	frag6_nfragpackets--;
    480 
    481 	{
    482 		KASSERT(m->m_flags & M_PKTHDR);
    483 		int plen = 0;
    484 		for (t = m; t; t = t->m_next) {
    485 			plen += t->m_len;
    486 		}
    487 		m->m_pkthdr.len = plen;
    488 		/* XXX XXX: clear csum_flags? */
    489 	}
    490 
    491 	/*
    492 	 * Restore NXT to the original.
    493 	 */
    494 	{
    495 		const int prvnxt = ip6_get_prevhdr(m, offset);
    496 		uint8_t *prvnxtp;
    497 
    498 		IP6_EXTHDR_GET(prvnxtp, uint8_t *, m, prvnxt,
    499 		    sizeof(*prvnxtp));
    500 		if (prvnxtp == NULL) {
    501 			goto dropfrag;
    502 		}
    503 		*prvnxtp = nxt;
    504 	}
    505 
    506 	IP6_STATINC(IP6_STAT_REASSEMBLED);
    507 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
    508 	rtcache_unref(rt, &ro);
    509 	mutex_exit(&frag6_lock);
    510 
    511 	/*
    512 	 * Tell launch routine the next header.
    513 	 */
    514 	*mp = m;
    515 	*offp = offset;
    516 	return nxt;
    517 
    518  dropfrag:
    519 	mutex_exit(&frag6_lock);
    520 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
    521 	IP6_STATINC(IP6_STAT_FRAGDROPPED);
    522 	m_freem(m);
    523  done:
    524 	rtcache_unref(rt, &ro);
    525 	return IPPROTO_DONE;
    526 }
    527 
    528 int
    529 ip6_reass_packet(struct mbuf **mp, int offset)
    530 {
    531 
    532 	if (frag6_input(mp, &offset, IPPROTO_IPV6) == IPPROTO_DONE) {
    533 		*mp = NULL;
    534 		return EINVAL;
    535 	}
    536 	return 0;
    537 }
    538 
    539 /*
    540  * Free a fragment reassembly header and all
    541  * associated datagrams.
    542  */
    543 static void
    544 frag6_freef(struct ip6q *q6)
    545 {
    546 	struct ip6asfrag *af6, *down6;
    547 
    548 	KASSERT(mutex_owned(&frag6_lock));
    549 
    550 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
    551 	     af6 = down6) {
    552 		struct mbuf *m = af6->ip6af_m;
    553 
    554 		down6 = af6->ip6af_down;
    555 		frag6_deq(af6);
    556 
    557 		/*
    558 		 * Return ICMP time exceeded error for the 1st fragment.
    559 		 * Just free other fragments.
    560 		 */
    561 		if (af6->ip6af_off == 0) {
    562 			struct ip6_hdr *ip6;
    563 
    564 			/* adjust pointer */
    565 			ip6 = mtod(m, struct ip6_hdr *);
    566 
    567 			/* restore source and destination addresses */
    568 			ip6->ip6_src = q6->ip6q_src;
    569 			ip6->ip6_dst = q6->ip6q_dst;
    570 
    571 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
    572 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
    573 		} else {
    574 			m_freem(m);
    575 		}
    576 		kmem_intr_free(af6, sizeof(struct ip6asfrag));
    577 	}
    578 
    579 	frag6_remque(q6);
    580 	frag6_nfrags -= q6->ip6q_nfrag;
    581 	kmem_intr_free(q6, sizeof(struct ip6q));
    582 	frag6_nfragpackets--;
    583 }
    584 
    585 /*
    586  * Put an ip fragment on a reassembly chain.
    587  * Like insque, but pointers in middle of structure.
    588  */
    589 void
    590 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6)
    591 {
    592 
    593 	KASSERT(mutex_owned(&frag6_lock));
    594 
    595 	af6->ip6af_up = up6;
    596 	af6->ip6af_down = up6->ip6af_down;
    597 	up6->ip6af_down->ip6af_up = af6;
    598 	up6->ip6af_down = af6;
    599 }
    600 
    601 /*
    602  * To frag6_enq as remque is to insque.
    603  */
    604 void
    605 frag6_deq(struct ip6asfrag *af6)
    606 {
    607 
    608 	KASSERT(mutex_owned(&frag6_lock));
    609 
    610 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
    611 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
    612 }
    613 
    614 /*
    615  * Insert newq after oldq.
    616  */
    617 void
    618 frag6_insque(struct ip6q *newq, struct ip6q *oldq)
    619 {
    620 
    621 	KASSERT(mutex_owned(&frag6_lock));
    622 
    623 	newq->ip6q_prev = oldq;
    624 	newq->ip6q_next = oldq->ip6q_next;
    625 	oldq->ip6q_next->ip6q_prev = newq;
    626 	oldq->ip6q_next = newq;
    627 }
    628 
    629 /*
    630  * Unlink p6.
    631  */
    632 void
    633 frag6_remque(struct ip6q *p6)
    634 {
    635 
    636 	KASSERT(mutex_owned(&frag6_lock));
    637 
    638 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
    639 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
    640 }
    641 
    642 void
    643 frag6_fasttimo(void)
    644 {
    645 
    646 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    647 
    648 	if (frag6_drainwanted) {
    649 		frag6_drain();
    650 		frag6_drainwanted = 0;
    651 	}
    652 
    653 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    654 }
    655 
    656 /*
    657  * IPv6 reassembling timer processing;
    658  * if a timer expires on a reassembly
    659  * queue, discard it.
    660  */
    661 void
    662 frag6_slowtimo(void)
    663 {
    664 	struct ip6q *q6;
    665 
    666 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
    667 
    668 	mutex_enter(&frag6_lock);
    669 	q6 = ip6q.ip6q_next;
    670 	if (q6) {
    671 		while (q6 != &ip6q) {
    672 			--q6->ip6q_ttl;
    673 			q6 = q6->ip6q_next;
    674 			if (q6->ip6q_prev->ip6q_ttl == 0) {
    675 				IP6_STATINC(IP6_STAT_FRAGTIMEOUT);
    676 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    677 				frag6_freef(q6->ip6q_prev);
    678 			}
    679 		}
    680 	}
    681 
    682 	/*
    683 	 * If we are over the maximum number of fragments
    684 	 * (due to the limit being lowered), drain off
    685 	 * enough to get down to the new limit.
    686 	 */
    687 	while (frag6_nfragpackets > (u_int)ip6_maxfragpackets &&
    688 	    ip6q.ip6q_prev) {
    689 		IP6_STATINC(IP6_STAT_FRAGOVERFLOW);
    690 		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    691 		frag6_freef(ip6q.ip6q_prev);
    692 	}
    693 	mutex_exit(&frag6_lock);
    694 
    695 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
    696 
    697 #if 0
    698 	/*
    699 	 * Routing changes might produce a better route than we last used;
    700 	 * make sure we notice eventually, even if forwarding only for one
    701 	 * destination and the cache is never replaced.
    702 	 */
    703 	rtcache_free(&ip6_forward_rt);
    704 	rtcache_free(&ipsrcchk_rt);
    705 #endif
    706 }
    707 
    708 void
    709 frag6_drainstub(void)
    710 {
    711 	frag6_drainwanted = 1;
    712 }
    713 
    714 /*
    715  * Drain off all datagram fragments.
    716  */
    717 void
    718 frag6_drain(void)
    719 {
    720 
    721 	if (mutex_tryenter(&frag6_lock)) {
    722 		while (ip6q.ip6q_next != &ip6q) {
    723 			IP6_STATINC(IP6_STAT_FRAGDROPPED);
    724 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
    725 			frag6_freef(ip6q.ip6q_next);
    726 		}
    727 		mutex_exit(&frag6_lock);
    728 	}
    729 }
    730