xref: /freebsd/sys/netinet6/frag6.c (revision 6829dae12bb055451fa467da4589c43bd03b1e64)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	$KAME: frag6.c,v 1.33 2002/01/07 11:34:48 kjc Exp $
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_rss.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/hash.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/domain.h>
45 #include <sys/eventhandler.h>
46 #include <sys/protosw.h>
47 #include <sys/socket.h>
48 #include <sys/errno.h>
49 #include <sys/time.h>
50 #include <sys/kernel.h>
51 #include <sys/syslog.h>
52 
53 #include <machine/atomic.h>
54 
55 #include <net/if.h>
56 #include <net/if_var.h>
57 #include <net/netisr.h>
58 #include <net/route.h>
59 #include <net/vnet.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/in_var.h>
63 #include <netinet/ip6.h>
64 #include <netinet6/ip6_var.h>
65 #include <netinet/icmp6.h>
66 #include <netinet/in_systm.h>	/* for ECN definitions */
67 #include <netinet/ip.h>		/* for ECN definitions */
68 
69 #include <security/mac/mac_framework.h>
70 
71 /*
72  * Reassembly headers are stored in hash buckets.
73  */
74 #define	IP6REASS_NHASH_LOG2	10
75 #define	IP6REASS_NHASH		(1 << IP6REASS_NHASH_LOG2)
76 #define	IP6REASS_HMASK		(IP6REASS_NHASH - 1)
77 
78 static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *,
79     uint32_t bucket __unused);
80 static void frag6_deq(struct ip6asfrag *, uint32_t bucket __unused);
81 static void frag6_insque_head(struct ip6q *, struct ip6q *,
82     uint32_t bucket);
83 static void frag6_remque(struct ip6q *, uint32_t bucket);
84 static void frag6_freef(struct ip6q *, uint32_t bucket);
85 
86 struct ip6qbucket {
87 	struct ip6q	ip6q;
88 	struct mtx	lock;
89 	int		count;
90 };
91 
92 VNET_DEFINE_STATIC(volatile u_int, frag6_nfragpackets);
93 volatile u_int frag6_nfrags = 0;
94 VNET_DEFINE_STATIC(struct ip6qbucket, ip6q[IP6REASS_NHASH]);
95 VNET_DEFINE_STATIC(uint32_t, ip6q_hashseed);
96 
97 #define	V_frag6_nfragpackets		VNET(frag6_nfragpackets)
98 #define	V_ip6q				VNET(ip6q)
99 #define	V_ip6q_hashseed			VNET(ip6q_hashseed)
100 
101 #define	IP6Q_LOCK(i)		mtx_lock(&V_ip6q[(i)].lock)
102 #define	IP6Q_TRYLOCK(i)		mtx_trylock(&V_ip6q[(i)].lock)
103 #define	IP6Q_LOCK_ASSERT(i)	mtx_assert(&V_ip6q[(i)].lock, MA_OWNED)
104 #define	IP6Q_UNLOCK(i)		mtx_unlock(&V_ip6q[(i)].lock)
105 #define	IP6Q_HEAD(i)		(&V_ip6q[(i)].ip6q)
106 
107 static MALLOC_DEFINE(M_FTABLE, "fragment", "fragment reassembly header");
108 
109 /*
110  * By default, limit the number of IP6 fragments across all reassembly
111  * queues to  1/32 of the total number of mbuf clusters.
112  *
113  * Limit the total number of reassembly queues per VNET to the
114  * IP6 fragment limit, but ensure the limit will not allow any bucket
115  * to grow above 100 items. (The bucket limit is
116  * IP_MAXFRAGPACKETS / (IPREASS_NHASH / 2), so the 50 is the correct
117  * multiplier to reach a 100-item limit.)
118  * The 100-item limit was chosen as brief testing seems to show that
119  * this produces "reasonable" performance on some subset of systems
120  * under DoS attack.
121  */
122 #define	IP6_MAXFRAGS		(nmbclusters / 32)
123 #define	IP6_MAXFRAGPACKETS	(imin(IP6_MAXFRAGS, IP6REASS_NHASH * 50))
124 
125 /*
126  * Initialise reassembly queue and fragment identifier.
127  */
128 void
129 frag6_set_bucketsize()
130 {
131 	int i;
132 
133 	if ((i = V_ip6_maxfragpackets) > 0)
134 		V_ip6_maxfragbucketsize = imax(i / (IP6REASS_NHASH / 2), 1);
135 }
136 
137 static void
138 frag6_change(void *tag)
139 {
140 	VNET_ITERATOR_DECL(vnet_iter);
141 
142 	ip6_maxfrags = IP6_MAXFRAGS;
143 	VNET_LIST_RLOCK_NOSLEEP();
144 	VNET_FOREACH(vnet_iter) {
145 		CURVNET_SET(vnet_iter);
146 		V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS;
147 		frag6_set_bucketsize();
148 		CURVNET_RESTORE();
149 	}
150 	VNET_LIST_RUNLOCK_NOSLEEP();
151 }
152 
153 void
154 frag6_init(void)
155 {
156 	struct ip6q *q6;
157 	int i;
158 
159 	V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS;
160 	frag6_set_bucketsize();
161 	for (i = 0; i < IP6REASS_NHASH; i++) {
162 		q6 = IP6Q_HEAD(i);
163 		q6->ip6q_next = q6->ip6q_prev = q6;
164 		mtx_init(&V_ip6q[i].lock, "ip6qlock", NULL, MTX_DEF);
165 		V_ip6q[i].count = 0;
166 	}
167 	V_ip6q_hashseed = arc4random();
168 	V_ip6_maxfragsperpacket = 64;
169 	if (!IS_DEFAULT_VNET(curvnet))
170 		return;
171 
172 	ip6_maxfrags = IP6_MAXFRAGS;
173 	EVENTHANDLER_REGISTER(nmbclusters_change,
174 	    frag6_change, NULL, EVENTHANDLER_PRI_ANY);
175 }
176 
177 /*
178  * In RFC2460, fragment and reassembly rule do not agree with each other,
179  * in terms of next header field handling in fragment header.
180  * While the sender will use the same value for all of the fragmented packets,
181  * receiver is suggested not to check the consistency.
182  *
183  * fragment rule (p20):
184  *	(2) A Fragment header containing:
185  *	The Next Header value that identifies the first header of
186  *	the Fragmentable Part of the original packet.
187  *		-> next header field is same for all fragments
188  *
189  * reassembly rule (p21):
190  *	The Next Header field of the last header of the Unfragmentable
191  *	Part is obtained from the Next Header field of the first
192  *	fragment's Fragment header.
193  *		-> should grab it from the first fragment only
194  *
195  * The following note also contradicts with fragment rule - no one is going to
196  * send different fragment with different next header field.
197  *
198  * additional note (p22):
199  *	The Next Header values in the Fragment headers of different
200  *	fragments of the same original packet may differ.  Only the value
201  *	from the Offset zero fragment packet is used for reassembly.
202  *		-> should grab it from the first fragment only
203  *
204  * There is no explicit reason given in the RFC.  Historical reason maybe?
205  */
206 /*
207  * Fragment input
208  */
209 int
210 frag6_input(struct mbuf **mp, int *offp, int proto)
211 {
212 	struct mbuf *m = *mp, *t;
213 	struct ip6_hdr *ip6;
214 	struct ip6_frag *ip6f;
215 	struct ip6q *head, *q6;
216 	struct ip6asfrag *af6, *ip6af, *af6dwn;
217 	struct in6_ifaddr *ia;
218 	int offset = *offp, nxt, i, next;
219 	int first_frag = 0;
220 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
221 	uint32_t hashkey[(sizeof(struct in6_addr) * 2 +
222 		    sizeof(ip6f->ip6f_ident)) / sizeof(uint32_t)];
223 	uint32_t hash, *hashkeyp;
224 	struct ifnet *dstifp;
225 	u_int8_t ecn, ecn0;
226 #ifdef RSS
227 	struct m_tag *mtag;
228 	struct ip6_direct_ctx *ip6dc;
229 #endif
230 
231 #if 0
232 	char ip6buf[INET6_ADDRSTRLEN];
233 #endif
234 
235 	ip6 = mtod(m, struct ip6_hdr *);
236 #ifndef PULLDOWN_TEST
237 	IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE);
238 	ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset);
239 #else
240 	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
241 	if (ip6f == NULL)
242 		return (IPPROTO_DONE);
243 #endif
244 
245 	dstifp = NULL;
246 	/* find the destination interface of the packet. */
247 	ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
248 	if (ia != NULL) {
249 		dstifp = ia->ia_ifp;
250 		ifa_free(&ia->ia_ifa);
251 	}
252 	/* jumbo payload can't contain a fragment header */
253 	if (ip6->ip6_plen == 0) {
254 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
255 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
256 		return IPPROTO_DONE;
257 	}
258 
259 	/*
260 	 * check whether fragment packet's fragment length is
261 	 * multiple of 8 octets.
262 	 * sizeof(struct ip6_frag) == 8
263 	 * sizeof(struct ip6_hdr) = 40
264 	 */
265 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
266 	    (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
267 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
268 		    offsetof(struct ip6_hdr, ip6_plen));
269 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
270 		return IPPROTO_DONE;
271 	}
272 
273 	IP6STAT_INC(ip6s_fragments);
274 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
275 
276 	/* offset now points to data portion */
277 	offset += sizeof(struct ip6_frag);
278 
279 	/*
280 	 * RFC 6946: Handle "atomic" fragments (offset and m bit set to 0)
281 	 * upfront, unrelated to any reassembly.  Just skip the fragment header.
282 	 */
283 	if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) {
284 		/* XXX-BZ we want dedicated counters for this. */
285 		IP6STAT_INC(ip6s_reassembled);
286 		in6_ifstat_inc(dstifp, ifs6_reass_ok);
287 		*offp = offset;
288 		m->m_flags |= M_FRAGMENTED;
289 		return (ip6f->ip6f_nxt);
290 	}
291 
292 	/* Get fragment length and discard 0-byte fragments. */
293 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
294 	if (frgpartlen == 0) {
295 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
296 		    offsetof(struct ip6_hdr, ip6_plen));
297 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
298 		IP6STAT_INC(ip6s_fragdropped);
299 		return IPPROTO_DONE;
300 	}
301 
302 	hashkeyp = hashkey;
303 	memcpy(hashkeyp, &ip6->ip6_src, sizeof(struct in6_addr));
304 	hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp);
305 	memcpy(hashkeyp, &ip6->ip6_dst, sizeof(struct in6_addr));
306 	hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp);
307 	*hashkeyp = ip6f->ip6f_ident;
308 	hash = jenkins_hash32(hashkey, nitems(hashkey), V_ip6q_hashseed);
309 	hash &= IP6REASS_HMASK;
310 	head = IP6Q_HEAD(hash);
311 	IP6Q_LOCK(hash);
312 
313 	/*
314 	 * Enforce upper bound on number of fragments.
315 	 * If maxfrag is 0, never accept fragments.
316 	 * If maxfrag is -1, accept all fragments without limitation.
317 	 */
318 	if (ip6_maxfrags < 0)
319 		;
320 	else if (atomic_load_int(&frag6_nfrags) >= (u_int)ip6_maxfrags)
321 		goto dropfrag;
322 
323 	for (q6 = head->ip6q_next; q6 != head; q6 = q6->ip6q_next)
324 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
325 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
326 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst)
327 #ifdef MAC
328 		    && mac_ip6q_match(m, q6)
329 #endif
330 		    )
331 			break;
332 
333 	if (q6 == head) {
334 		/*
335 		 * the first fragment to arrive, create a reassembly queue.
336 		 */
337 		first_frag = 1;
338 
339 		/*
340 		 * Enforce upper bound on number of fragmented packets
341 		 * for which we attempt reassembly;
342 		 * If maxfragpackets is 0, never accept fragments.
343 		 * If maxfragpackets is -1, accept all fragments without
344 		 * limitation.
345 		 */
346 		if (V_ip6_maxfragpackets < 0)
347 			;
348 		else if (V_ip6q[hash].count >= V_ip6_maxfragbucketsize ||
349 		    atomic_load_int(&V_frag6_nfragpackets) >=
350 		    (u_int)V_ip6_maxfragpackets)
351 			goto dropfrag;
352 		atomic_add_int(&V_frag6_nfragpackets, 1);
353 		q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FTABLE,
354 		    M_NOWAIT);
355 		if (q6 == NULL)
356 			goto dropfrag;
357 		bzero(q6, sizeof(*q6));
358 #ifdef MAC
359 		if (mac_ip6q_init(q6, M_NOWAIT) != 0) {
360 			free(q6, M_FTABLE);
361 			goto dropfrag;
362 		}
363 		mac_ip6q_create(m, q6);
364 #endif
365 		frag6_insque_head(q6, head, hash);
366 
367 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
368 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
369 #ifdef notyet
370 		q6->ip6q_nxtp	= (u_char *)nxtp;
371 #endif
372 		q6->ip6q_ident	= ip6f->ip6f_ident;
373 		q6->ip6q_ttl	= IPV6_FRAGTTL;
374 		q6->ip6q_src	= ip6->ip6_src;
375 		q6->ip6q_dst	= ip6->ip6_dst;
376 		q6->ip6q_ecn	=
377 		    (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
378 		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
379 
380 		q6->ip6q_nfrag = 0;
381 	}
382 
383 	/*
384 	 * If it's the 1st fragment, record the length of the
385 	 * unfragmentable part and the next header of the fragment header.
386 	 */
387 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
388 	if (fragoff == 0) {
389 		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
390 		    sizeof(struct ip6_frag);
391 		q6->ip6q_nxt = ip6f->ip6f_nxt;
392 	}
393 
394 	/*
395 	 * Check that the reassembled packet would not exceed 65535 bytes
396 	 * in size.
397 	 * If it would exceed, discard the fragment and return an ICMP error.
398 	 */
399 	if (q6->ip6q_unfrglen >= 0) {
400 		/* The 1st fragment has already arrived. */
401 		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
402 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
403 			    offset - sizeof(struct ip6_frag) +
404 			    offsetof(struct ip6_frag, ip6f_offlg));
405 			IP6Q_UNLOCK(hash);
406 			return (IPPROTO_DONE);
407 		}
408 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
409 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
410 		    offset - sizeof(struct ip6_frag) +
411 		    offsetof(struct ip6_frag, ip6f_offlg));
412 		IP6Q_UNLOCK(hash);
413 		return (IPPROTO_DONE);
414 	}
415 	/*
416 	 * If it's the first fragment, do the above check for each
417 	 * fragment already stored in the reassembly queue.
418 	 */
419 	if (fragoff == 0) {
420 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
421 		     af6 = af6dwn) {
422 			af6dwn = af6->ip6af_down;
423 
424 			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
425 			    IPV6_MAXPACKET) {
426 				struct mbuf *merr = IP6_REASS_MBUF(af6);
427 				struct ip6_hdr *ip6err;
428 				int erroff = af6->ip6af_offset;
429 
430 				/* dequeue the fragment. */
431 				frag6_deq(af6, hash);
432 				free(af6, M_FTABLE);
433 
434 				/* adjust pointer. */
435 				ip6err = mtod(merr, struct ip6_hdr *);
436 
437 				/*
438 				 * Restore source and destination addresses
439 				 * in the erroneous IPv6 header.
440 				 */
441 				ip6err->ip6_src = q6->ip6q_src;
442 				ip6err->ip6_dst = q6->ip6q_dst;
443 
444 				icmp6_error(merr, ICMP6_PARAM_PROB,
445 				    ICMP6_PARAMPROB_HEADER,
446 				    erroff - sizeof(struct ip6_frag) +
447 				    offsetof(struct ip6_frag, ip6f_offlg));
448 			}
449 		}
450 	}
451 
452 	ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FTABLE,
453 	    M_NOWAIT);
454 	if (ip6af == NULL)
455 		goto dropfrag;
456 	bzero(ip6af, sizeof(*ip6af));
457 	ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
458 	ip6af->ip6af_off = fragoff;
459 	ip6af->ip6af_frglen = frgpartlen;
460 	ip6af->ip6af_offset = offset;
461 	IP6_REASS_MBUF(ip6af) = m;
462 
463 	if (first_frag) {
464 		af6 = (struct ip6asfrag *)q6;
465 		goto insert;
466 	}
467 
468 	/*
469 	 * Handle ECN by comparing this segment with the first one;
470 	 * if CE is set, do not lose CE.
471 	 * drop if CE and not-ECT are mixed for the same packet.
472 	 */
473 	ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
474 	ecn0 = q6->ip6q_ecn;
475 	if (ecn == IPTOS_ECN_CE) {
476 		if (ecn0 == IPTOS_ECN_NOTECT) {
477 			free(ip6af, M_FTABLE);
478 			goto dropfrag;
479 		}
480 		if (ecn0 != IPTOS_ECN_CE)
481 			q6->ip6q_ecn = IPTOS_ECN_CE;
482 	}
483 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
484 		free(ip6af, M_FTABLE);
485 		goto dropfrag;
486 	}
487 
488 	/*
489 	 * Find a segment which begins after this one does.
490 	 */
491 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
492 	     af6 = af6->ip6af_down)
493 		if (af6->ip6af_off > ip6af->ip6af_off)
494 			break;
495 
496 #if 0
497 	/*
498 	 * If there is a preceding segment, it may provide some of
499 	 * our data already.  If so, drop the data from the incoming
500 	 * segment.  If it provides all of our data, drop us.
501 	 */
502 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
503 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
504 			- ip6af->ip6af_off;
505 		if (i > 0) {
506 			if (i >= ip6af->ip6af_frglen)
507 				goto dropfrag;
508 			m_adj(IP6_REASS_MBUF(ip6af), i);
509 			ip6af->ip6af_off += i;
510 			ip6af->ip6af_frglen -= i;
511 		}
512 	}
513 
514 	/*
515 	 * While we overlap succeeding segments trim them or,
516 	 * if they are completely covered, dequeue them.
517 	 */
518 	while (af6 != (struct ip6asfrag *)q6 &&
519 	       ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) {
520 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
521 		if (i < af6->ip6af_frglen) {
522 			af6->ip6af_frglen -= i;
523 			af6->ip6af_off += i;
524 			m_adj(IP6_REASS_MBUF(af6), i);
525 			break;
526 		}
527 		af6 = af6->ip6af_down;
528 		m_freem(IP6_REASS_MBUF(af6->ip6af_up));
529 		frag6_deq(af6->ip6af_up, hash);
530 	}
531 #else
532 	/*
533 	 * If the incoming framgent overlaps some existing fragments in
534 	 * the reassembly queue, drop it, since it is dangerous to override
535 	 * existing fragments from a security point of view.
536 	 * We don't know which fragment is the bad guy - here we trust
537 	 * fragment that came in earlier, with no real reason.
538 	 *
539 	 * Note: due to changes after disabling this part, mbuf passed to
540 	 * m_adj() below now does not meet the requirement.
541 	 */
542 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
543 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
544 			- ip6af->ip6af_off;
545 		if (i > 0) {
546 #if 0				/* suppress the noisy log */
547 			log(LOG_ERR, "%d bytes of a fragment from %s "
548 			    "overlaps the previous fragment\n",
549 			    i, ip6_sprintf(ip6buf, &q6->ip6q_src));
550 #endif
551 			free(ip6af, M_FTABLE);
552 			goto dropfrag;
553 		}
554 	}
555 	if (af6 != (struct ip6asfrag *)q6) {
556 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
557 		if (i > 0) {
558 #if 0				/* suppress the noisy log */
559 			log(LOG_ERR, "%d bytes of a fragment from %s "
560 			    "overlaps the succeeding fragment",
561 			    i, ip6_sprintf(ip6buf, &q6->ip6q_src));
562 #endif
563 			free(ip6af, M_FTABLE);
564 			goto dropfrag;
565 		}
566 	}
567 #endif
568 
569 insert:
570 #ifdef MAC
571 	if (!first_frag)
572 		mac_ip6q_update(m, q6);
573 #endif
574 
575 	/*
576 	 * Stick new segment in its place;
577 	 * check for complete reassembly.
578 	 * If not complete, check fragment limit.
579 	 * Move to front of packet queue, as we are
580 	 * the most recently active fragmented packet.
581 	 */
582 	frag6_enq(ip6af, af6->ip6af_up, hash);
583 	atomic_add_int(&frag6_nfrags, 1);
584 	q6->ip6q_nfrag++;
585 #if 0 /* xxx */
586 	if (q6 != head->ip6q_next) {
587 		frag6_remque(q6, hash);
588 		frag6_insque_head(q6, head, hash);
589 	}
590 #endif
591 	next = 0;
592 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
593 	     af6 = af6->ip6af_down) {
594 		if (af6->ip6af_off != next) {
595 			if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) {
596 				IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag);
597 				frag6_freef(q6, hash);
598 			}
599 			IP6Q_UNLOCK(hash);
600 			return IPPROTO_DONE;
601 		}
602 		next += af6->ip6af_frglen;
603 	}
604 	if (af6->ip6af_up->ip6af_mff) {
605 		if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) {
606 			IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag);
607 			frag6_freef(q6, hash);
608 		}
609 		IP6Q_UNLOCK(hash);
610 		return IPPROTO_DONE;
611 	}
612 
613 	/*
614 	 * Reassembly is complete; concatenate fragments.
615 	 */
616 	ip6af = q6->ip6q_down;
617 	t = m = IP6_REASS_MBUF(ip6af);
618 	af6 = ip6af->ip6af_down;
619 	frag6_deq(ip6af, hash);
620 	while (af6 != (struct ip6asfrag *)q6) {
621 		m->m_pkthdr.csum_flags &=
622 		    IP6_REASS_MBUF(af6)->m_pkthdr.csum_flags;
623 		m->m_pkthdr.csum_data +=
624 		    IP6_REASS_MBUF(af6)->m_pkthdr.csum_data;
625 
626 		af6dwn = af6->ip6af_down;
627 		frag6_deq(af6, hash);
628 		while (t->m_next)
629 			t = t->m_next;
630 		m_adj(IP6_REASS_MBUF(af6), af6->ip6af_offset);
631 		m_demote_pkthdr(IP6_REASS_MBUF(af6));
632 		m_cat(t, IP6_REASS_MBUF(af6));
633 		free(af6, M_FTABLE);
634 		af6 = af6dwn;
635 	}
636 
637 	while (m->m_pkthdr.csum_data & 0xffff0000)
638 		m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
639 		    (m->m_pkthdr.csum_data >> 16);
640 
641 	/* adjust offset to point where the original next header starts */
642 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
643 	free(ip6af, M_FTABLE);
644 	ip6 = mtod(m, struct ip6_hdr *);
645 	ip6->ip6_plen = htons((u_short)next + offset - sizeof(struct ip6_hdr));
646 	if (q6->ip6q_ecn == IPTOS_ECN_CE)
647 		ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20);
648 	nxt = q6->ip6q_nxt;
649 #ifdef notyet
650 	*q6->ip6q_nxtp = (u_char)(nxt & 0xff);
651 #endif
652 
653 	if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) {
654 		frag6_remque(q6, hash);
655 		atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag);
656 #ifdef MAC
657 		mac_ip6q_destroy(q6);
658 #endif
659 		free(q6, M_FTABLE);
660 		atomic_subtract_int(&V_frag6_nfragpackets, 1);
661 
662 		goto dropfrag;
663 	}
664 
665 	/*
666 	 * Store NXT to the original.
667 	 */
668 	m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t),
669 	    (caddr_t)&nxt);
670 
671 	frag6_remque(q6, hash);
672 	atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag);
673 #ifdef MAC
674 	mac_ip6q_reassemble(q6, m);
675 	mac_ip6q_destroy(q6);
676 #endif
677 	free(q6, M_FTABLE);
678 	atomic_subtract_int(&V_frag6_nfragpackets, 1);
679 
680 	if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
681 		int plen = 0;
682 		for (t = m; t; t = t->m_next)
683 			plen += t->m_len;
684 		m->m_pkthdr.len = plen;
685 	}
686 
687 #ifdef RSS
688 	mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc),
689 	    M_NOWAIT);
690 	if (mtag == NULL)
691 		goto dropfrag;
692 
693 	ip6dc = (struct ip6_direct_ctx *)(mtag + 1);
694 	ip6dc->ip6dc_nxt = nxt;
695 	ip6dc->ip6dc_off = offset;
696 
697 	m_tag_prepend(m, mtag);
698 #endif
699 
700 	IP6Q_UNLOCK(hash);
701 	IP6STAT_INC(ip6s_reassembled);
702 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
703 
704 #ifdef RSS
705 	/*
706 	 * Queue/dispatch for reprocessing.
707 	 */
708 	netisr_dispatch(NETISR_IPV6_DIRECT, m);
709 	return IPPROTO_DONE;
710 #endif
711 
712 	/*
713 	 * Tell launch routine the next header
714 	 */
715 
716 	*mp = m;
717 	*offp = offset;
718 
719 	return nxt;
720 
721  dropfrag:
722 	IP6Q_UNLOCK(hash);
723 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
724 	IP6STAT_INC(ip6s_fragdropped);
725 	m_freem(m);
726 	return IPPROTO_DONE;
727 }
728 
729 /*
730  * Free a fragment reassembly header and all
731  * associated datagrams.
732  */
733 static void
734 frag6_freef(struct ip6q *q6, uint32_t bucket)
735 {
736 	struct ip6asfrag *af6, *down6;
737 
738 	IP6Q_LOCK_ASSERT(bucket);
739 
740 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
741 	     af6 = down6) {
742 		struct mbuf *m = IP6_REASS_MBUF(af6);
743 
744 		down6 = af6->ip6af_down;
745 		frag6_deq(af6, bucket);
746 
747 		/*
748 		 * Return ICMP time exceeded error for the 1st fragment.
749 		 * Just free other fragments.
750 		 */
751 		if (af6->ip6af_off == 0) {
752 			struct ip6_hdr *ip6;
753 
754 			/* adjust pointer */
755 			ip6 = mtod(m, struct ip6_hdr *);
756 
757 			/* restore source and destination addresses */
758 			ip6->ip6_src = q6->ip6q_src;
759 			ip6->ip6_dst = q6->ip6q_dst;
760 
761 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
762 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
763 		} else
764 			m_freem(m);
765 		free(af6, M_FTABLE);
766 	}
767 	frag6_remque(q6, bucket);
768 	atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag);
769 #ifdef MAC
770 	mac_ip6q_destroy(q6);
771 #endif
772 	free(q6, M_FTABLE);
773 	atomic_subtract_int(&V_frag6_nfragpackets, 1);
774 }
775 
776 /*
777  * Put an ip fragment on a reassembly chain.
778  * Like insque, but pointers in middle of structure.
779  */
780 static void
781 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6,
782     uint32_t bucket __unused)
783 {
784 
785 	IP6Q_LOCK_ASSERT(bucket);
786 
787 	af6->ip6af_up = up6;
788 	af6->ip6af_down = up6->ip6af_down;
789 	up6->ip6af_down->ip6af_up = af6;
790 	up6->ip6af_down = af6;
791 }
792 
793 /*
794  * To frag6_enq as remque is to insque.
795  */
796 static void
797 frag6_deq(struct ip6asfrag *af6, uint32_t bucket __unused)
798 {
799 
800 	IP6Q_LOCK_ASSERT(bucket);
801 
802 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
803 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
804 }
805 
806 static void
807 frag6_insque_head(struct ip6q *new, struct ip6q *old, uint32_t bucket)
808 {
809 
810 	IP6Q_LOCK_ASSERT(bucket);
811 	KASSERT(IP6Q_HEAD(bucket) == old,
812 	    ("%s: attempt to insert at head of wrong bucket"
813 	    " (bucket=%u, old=%p)", __func__, bucket, old));
814 
815 	new->ip6q_prev = old;
816 	new->ip6q_next = old->ip6q_next;
817 	old->ip6q_next->ip6q_prev= new;
818 	old->ip6q_next = new;
819 	V_ip6q[bucket].count++;
820 }
821 
822 static void
823 frag6_remque(struct ip6q *p6, uint32_t bucket)
824 {
825 
826 	IP6Q_LOCK_ASSERT(bucket);
827 
828 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
829 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
830 	V_ip6q[bucket].count--;
831 }
832 
833 /*
834  * IPv6 reassembling timer processing;
835  * if a timer expires on a reassembly
836  * queue, discard it.
837  */
838 void
839 frag6_slowtimo(void)
840 {
841 	VNET_ITERATOR_DECL(vnet_iter);
842 	struct ip6q *head, *q6;
843 	int i;
844 
845 	VNET_LIST_RLOCK_NOSLEEP();
846 	VNET_FOREACH(vnet_iter) {
847 		CURVNET_SET(vnet_iter);
848 		for (i = 0; i < IP6REASS_NHASH; i++) {
849 			IP6Q_LOCK(i);
850 			head = IP6Q_HEAD(i);
851 			q6 = head->ip6q_next;
852 			if (q6 == NULL) {
853 				/*
854 				 * XXXJTL: This should never happen. This
855 				 * should turn into an assertion.
856 				 */
857 				IP6Q_UNLOCK(i);
858 				continue;
859 			}
860 			while (q6 != head) {
861 				--q6->ip6q_ttl;
862 				q6 = q6->ip6q_next;
863 				if (q6->ip6q_prev->ip6q_ttl == 0) {
864 					IP6STAT_ADD(ip6s_fragtimeout,
865 						q6->ip6q_prev->ip6q_nfrag);
866 					/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
867 					frag6_freef(q6->ip6q_prev, i);
868 				}
869 			}
870 			/*
871 			 * If we are over the maximum number of fragments
872 			 * (due to the limit being lowered), drain off
873 			 * enough to get down to the new limit.
874 			 * Note that we drain all reassembly queues if
875 			 * maxfragpackets is 0 (fragmentation is disabled),
876 			 * and don't enforce a limit when maxfragpackets
877 			 * is negative.
878 			 */
879 			while ((V_ip6_maxfragpackets == 0 ||
880 			    (V_ip6_maxfragpackets > 0 &&
881 			    V_ip6q[i].count > V_ip6_maxfragbucketsize)) &&
882 			    head->ip6q_prev != head) {
883 				IP6STAT_ADD(ip6s_fragoverflow,
884 					q6->ip6q_prev->ip6q_nfrag);
885 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
886 				frag6_freef(head->ip6q_prev, i);
887 			}
888 			IP6Q_UNLOCK(i);
889 		}
890 		/*
891 		 * If we are still over the maximum number of fragmented
892 		 * packets, drain off enough to get down to the new limit.
893 		 */
894 		i = 0;
895 		while (V_ip6_maxfragpackets >= 0 &&
896 		    atomic_load_int(&V_frag6_nfragpackets) >
897 		    (u_int)V_ip6_maxfragpackets) {
898 			IP6Q_LOCK(i);
899 			head = IP6Q_HEAD(i);
900 			if (head->ip6q_prev != head) {
901 				IP6STAT_ADD(ip6s_fragoverflow,
902 					q6->ip6q_prev->ip6q_nfrag);
903 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
904 				frag6_freef(head->ip6q_prev, i);
905 			}
906 			IP6Q_UNLOCK(i);
907 			i = (i + 1) % IP6REASS_NHASH;
908 		}
909 		CURVNET_RESTORE();
910 	}
911 	VNET_LIST_RUNLOCK_NOSLEEP();
912 }
913 
914 /*
915  * Drain off all datagram fragments.
916  */
917 void
918 frag6_drain(void)
919 {
920 	VNET_ITERATOR_DECL(vnet_iter);
921 	struct ip6q *head;
922 	int i;
923 
924 	VNET_LIST_RLOCK_NOSLEEP();
925 	VNET_FOREACH(vnet_iter) {
926 		CURVNET_SET(vnet_iter);
927 		for (i = 0; i < IP6REASS_NHASH; i++) {
928 			if (IP6Q_TRYLOCK(i) == 0)
929 				continue;
930 			head = IP6Q_HEAD(i);
931 			while (head->ip6q_next != head) {
932 				IP6STAT_INC(ip6s_fragdropped);
933 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
934 				frag6_freef(head->ip6q_next, i);
935 			}
936 			IP6Q_UNLOCK(i);
937 		}
938 		CURVNET_RESTORE();
939 	}
940 	VNET_LIST_RUNLOCK_NOSLEEP();
941 }
942 
943 int
944 ip6_deletefraghdr(struct mbuf *m, int offset, int wait)
945 {
946 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
947 	struct mbuf *t;
948 
949 	/* Delete frag6 header. */
950 	if (m->m_len >= offset + sizeof(struct ip6_frag)) {
951 		/* This is the only possible case with !PULLDOWN_TEST. */
952 		bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag),
953 		    offset);
954 		m->m_data += sizeof(struct ip6_frag);
955 		m->m_len -= sizeof(struct ip6_frag);
956 	} else {
957 		/* This comes with no copy if the boundary is on cluster. */
958 		if ((t = m_split(m, offset, wait)) == NULL)
959 			return (ENOMEM);
960 		m_adj(t, sizeof(struct ip6_frag));
961 		m_cat(m, t);
962 	}
963 
964 	m->m_flags |= M_FRAGMENTED;
965 	return (0);
966 }
967