xref: /freebsd/sys/netinet6/frag6.c (revision 0bf48626aaa33768078f5872b922b1487b3a9296)
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/domain.h>
42 #include <sys/eventhandler.h>
43 #include <sys/hash.h>
44 #include <sys/kernel.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/protosw.h>
48 #include <sys/socket.h>
49 #include <sys/sysctl.h>
50 #include <sys/syslog.h>
51 
52 #include <net/if.h>
53 #include <net/if_var.h>
54 #include <net/netisr.h>
55 #include <net/route.h>
56 #include <net/vnet.h>
57 
58 #include <netinet/in.h>
59 #include <netinet/in_var.h>
60 #include <netinet/ip6.h>
61 #include <netinet6/ip6_var.h>
62 #include <netinet/icmp6.h>
63 #include <netinet/in_systm.h>	/* for ECN definitions */
64 #include <netinet/ip.h>		/* for ECN definitions */
65 
66 #ifdef MAC
67 #include <security/mac/mac_framework.h>
68 #endif
69 
70 /*
71  * Reassembly headers are stored in hash buckets.
72  */
73 #define	IP6REASS_NHASH_LOG2	10
74 #define	IP6REASS_NHASH		(1 << IP6REASS_NHASH_LOG2)
75 #define	IP6REASS_HMASK		(IP6REASS_NHASH - 1)
76 
77 static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *,
78     uint32_t bucket __unused);
79 static void frag6_deq(struct ip6asfrag *, uint32_t bucket __unused);
80 static void frag6_insque_head(struct ip6q *, struct ip6q *,
81     uint32_t bucket);
82 static void frag6_remque(struct ip6q *, uint32_t bucket);
83 static void frag6_freef(struct ip6q *, uint32_t bucket);
84 
85 struct ip6qbucket {
86 	struct ip6q	ip6q;
87 	struct mtx	lock;
88 	int		count;
89 };
90 
91 static MALLOC_DEFINE(M_FRAG6, "frag6", "IPv6 fragment reassembly header");
92 
93 /* System wide (global) maximum and count of packets in reassembly queues. */
94 static int ip6_maxfrags;
95 static volatile u_int frag6_nfrags = 0;
96 
97 /* Maximum and current packets in per-VNET reassembly queue. */
98 VNET_DEFINE_STATIC(int,			ip6_maxfragpackets);
99 VNET_DEFINE_STATIC(volatile u_int,	frag6_nfragpackets);
100 #define	V_ip6_maxfragpackets		VNET(ip6_maxfragpackets)
101 #define	V_frag6_nfragpackets		VNET(frag6_nfragpackets)
102 
103 /* Maximum per-VNET reassembly queues per bucket and fragments per packet. */
104 VNET_DEFINE_STATIC(int,			ip6_maxfragbucketsize);
105 VNET_DEFINE_STATIC(int,			ip6_maxfragsperpacket);
106 #define	V_ip6_maxfragbucketsize		VNET(ip6_maxfragbucketsize)
107 #define	V_ip6_maxfragsperpacket		VNET(ip6_maxfragsperpacket)
108 
109 /* Per-VNET reassembly queue buckets. */
110 VNET_DEFINE_STATIC(struct ip6qbucket,	ip6qb[IP6REASS_NHASH]);
111 VNET_DEFINE_STATIC(uint32_t,		ip6qb_hashseed);
112 #define	V_ip6qb				VNET(ip6qb)
113 #define	V_ip6qb_hashseed		VNET(ip6qb_hashseed)
114 
115 #define	IP6QB_LOCK(_b)		mtx_lock(&V_ip6qb[(_b)].lock)
116 #define	IP6QB_TRYLOCK(_b)	mtx_trylock(&V_ip6qb[(_b)].lock)
117 #define	IP6QB_LOCK_ASSERT(_b)	mtx_assert(&V_ip6qb[(_b)].lock, MA_OWNED)
118 #define	IP6QB_UNLOCK(_b)	mtx_unlock(&V_ip6qb[(_b)].lock)
119 #define	IP6QB_HEAD(_b)		(&V_ip6qb[(_b)].ip6q)
120 
121 /*
122  * By default, limit the number of IP6 fragments across all reassembly
123  * queues to  1/32 of the total number of mbuf clusters.
124  *
125  * Limit the total number of reassembly queues per VNET to the
126  * IP6 fragment limit, but ensure the limit will not allow any bucket
127  * to grow above 100 items. (The bucket limit is
128  * IP_MAXFRAGPACKETS / (IPREASS_NHASH / 2), so the 50 is the correct
129  * multiplier to reach a 100-item limit.)
130  * The 100-item limit was chosen as brief testing seems to show that
131  * this produces "reasonable" performance on some subset of systems
132  * under DoS attack.
133  */
134 #define	IP6_MAXFRAGS		(nmbclusters / 32)
135 #define	IP6_MAXFRAGPACKETS	(imin(IP6_MAXFRAGS, IP6REASS_NHASH * 50))
136 
137 
138 /*
139  * Sysctls and helper function.
140  */
141 SYSCTL_DECL(_net_inet6_ip6);
142 
143 static void
144 frag6_set_bucketsize(void)
145 {
146 	int i;
147 
148 	if ((i = V_ip6_maxfragpackets) > 0)
149 		V_ip6_maxfragbucketsize = imax(i / (IP6REASS_NHASH / 2), 1);
150 }
151 
152 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGS, maxfrags,
153 	CTLFLAG_RW, &ip6_maxfrags, 0,
154 	"Maximum allowed number of outstanding IPv6 packet fragments. "
155 	"A value of 0 means no fragmented packets will be accepted, while a "
156 	"a value of -1 means no limit");
157 
158 static int
159 sysctl_ip6_maxfragpackets(SYSCTL_HANDLER_ARGS)
160 {
161 	int error, val;
162 
163 	val = V_ip6_maxfragpackets;
164 	error = sysctl_handle_int(oidp, &val, 0, req);
165 	if (error != 0 || !req->newptr)
166 		return (error);
167 	V_ip6_maxfragpackets = val;
168 	frag6_set_bucketsize();
169 	return (0);
170 }
171 SYSCTL_PROC(_net_inet6_ip6, IPV6CTL_MAXFRAGPACKETS, maxfragpackets,
172 	CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, NULL, 0,
173 	sysctl_ip6_maxfragpackets, "I",
174 	"Default maximum number of outstanding fragmented IPv6 packets. "
175 	"A value of 0 means no fragmented packets will be accepted, while a "
176 	"a value of -1 means no limit");
177 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGSPERPACKET, maxfragsperpacket,
178 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragsperpacket), 0,
179 	"Maximum allowed number of fragments per packet");
180 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGBUCKETSIZE, maxfragbucketsize,
181 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragbucketsize), 0,
182 	"Maximum number of reassembly queues per hash bucket");
183 
184 
185 /*
186  * Remove the IPv6 fragmentation header from the mbuf.
187  */
188 int
189 ip6_deletefraghdr(struct mbuf *m, int offset, int wait)
190 {
191 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
192 	struct mbuf *t;
193 
194 	/* Delete frag6 header. */
195 	if (m->m_len >= offset + sizeof(struct ip6_frag)) {
196 		/* This is the only possible case with !PULLDOWN_TEST. */
197 		bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag),
198 		    offset);
199 		m->m_data += sizeof(struct ip6_frag);
200 		m->m_len -= sizeof(struct ip6_frag);
201 	} else {
202 		/* This comes with no copy if the boundary is on cluster. */
203 		if ((t = m_split(m, offset, wait)) == NULL)
204 			return (ENOMEM);
205 		m_adj(t, sizeof(struct ip6_frag));
206 		m_cat(m, t);
207 	}
208 
209 	m->m_flags |= M_FRAGMENTED;
210 	return (0);
211 }
212 
213 /*
214  * Free a fragment reassembly header and all
215  * associated datagrams.
216  */
217 static void
218 frag6_freef(struct ip6q *q6, uint32_t bucket)
219 {
220 	struct ip6asfrag *af6, *down6;
221 
222 	IP6QB_LOCK_ASSERT(bucket);
223 
224 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
225 	     af6 = down6) {
226 		struct mbuf *m = IP6_REASS_MBUF(af6);
227 
228 		down6 = af6->ip6af_down;
229 		frag6_deq(af6, bucket);
230 
231 		/*
232 		 * Return ICMP time exceeded error for the 1st fragment.
233 		 * Just free other fragments.
234 		 */
235 		if (af6->ip6af_off == 0) {
236 			struct ip6_hdr *ip6;
237 
238 			/* adjust pointer */
239 			ip6 = mtod(m, struct ip6_hdr *);
240 
241 			/* restore source and destination addresses */
242 			ip6->ip6_src = q6->ip6q_src;
243 			ip6->ip6_dst = q6->ip6q_dst;
244 
245 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
246 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
247 		} else
248 			m_freem(m);
249 		free(af6, M_FRAG6);
250 	}
251 	frag6_remque(q6, bucket);
252 	atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag);
253 #ifdef MAC
254 	mac_ip6q_destroy(q6);
255 #endif
256 	free(q6, M_FRAG6);
257 	atomic_subtract_int(&V_frag6_nfragpackets, 1);
258 }
259 
260 /*
261  * In RFC2460, fragment and reassembly rule do not agree with each other,
262  * in terms of next header field handling in fragment header.
263  * While the sender will use the same value for all of the fragmented packets,
264  * receiver is suggested not to check the consistency.
265  *
266  * fragment rule (p20):
267  *	(2) A Fragment header containing:
268  *	The Next Header value that identifies the first header of
269  *	the Fragmentable Part of the original packet.
270  *		-> next header field is same for all fragments
271  *
272  * reassembly rule (p21):
273  *	The Next Header field of the last header of the Unfragmentable
274  *	Part is obtained from the Next Header field of the first
275  *	fragment's Fragment header.
276  *		-> should grab it from the first fragment only
277  *
278  * The following note also contradicts with fragment rule - no one is going to
279  * send different fragment with different next header field.
280  *
281  * additional note (p22):
282  *	The Next Header values in the Fragment headers of different
283  *	fragments of the same original packet may differ.  Only the value
284  *	from the Offset zero fragment packet is used for reassembly.
285  *		-> should grab it from the first fragment only
286  *
287  * There is no explicit reason given in the RFC.  Historical reason maybe?
288  */
289 /*
290  * Fragment input
291  */
292 int
293 frag6_input(struct mbuf **mp, int *offp, int proto)
294 {
295 	struct mbuf *m = *mp, *t;
296 	struct ip6_hdr *ip6;
297 	struct ip6_frag *ip6f;
298 	struct ip6q *head, *q6;
299 	struct ip6asfrag *af6, *ip6af, *af6dwn;
300 	struct in6_ifaddr *ia;
301 	int offset = *offp, nxt, i, next;
302 	int first_frag = 0;
303 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
304 	uint32_t hashkey[(sizeof(struct in6_addr) * 2 +
305 		    sizeof(ip6f->ip6f_ident)) / sizeof(uint32_t)];
306 	uint32_t bucket, *hashkeyp;
307 	struct ifnet *dstifp;
308 	u_int8_t ecn, ecn0;
309 #ifdef RSS
310 	struct m_tag *mtag;
311 	struct ip6_direct_ctx *ip6dc;
312 #endif
313 
314 	ip6 = mtod(m, struct ip6_hdr *);
315 #ifndef PULLDOWN_TEST
316 	IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE);
317 	ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset);
318 #else
319 	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
320 	if (ip6f == NULL)
321 		return (IPPROTO_DONE);
322 #endif
323 
324 	dstifp = NULL;
325 	/* find the destination interface of the packet. */
326 	ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
327 	if (ia != NULL) {
328 		dstifp = ia->ia_ifp;
329 		ifa_free(&ia->ia_ifa);
330 	}
331 	/* jumbo payload can't contain a fragment header */
332 	if (ip6->ip6_plen == 0) {
333 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
334 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
335 		return IPPROTO_DONE;
336 	}
337 
338 	/*
339 	 * check whether fragment packet's fragment length is
340 	 * multiple of 8 octets.
341 	 * sizeof(struct ip6_frag) == 8
342 	 * sizeof(struct ip6_hdr) = 40
343 	 */
344 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
345 	    (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
346 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
347 		    offsetof(struct ip6_hdr, ip6_plen));
348 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
349 		return IPPROTO_DONE;
350 	}
351 
352 	IP6STAT_INC(ip6s_fragments);
353 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
354 
355 	/* offset now points to data portion */
356 	offset += sizeof(struct ip6_frag);
357 
358 	/*
359 	 * Handle "atomic" fragments (offset and m bit set to 0) upfront,
360 	 * unrelated to any reassembly (see RFC 6946 and section 4.5 of RFC
361 	 * 8200).  Just skip the fragment header.
362 	 */
363 	if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) {
364 		IP6STAT_INC(ip6s_atomicfrags);
365 		in6_ifstat_inc(dstifp, ifs6_reass_ok);
366 		*offp = offset;
367 		m->m_flags |= M_FRAGMENTED;
368 		return (ip6f->ip6f_nxt);
369 	}
370 
371 	/* Get fragment length and discard 0-byte fragments. */
372 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
373 	if (frgpartlen == 0) {
374 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
375 		    offsetof(struct ip6_hdr, ip6_plen));
376 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
377 		IP6STAT_INC(ip6s_fragdropped);
378 		return IPPROTO_DONE;
379 	}
380 
381 	hashkeyp = hashkey;
382 	memcpy(hashkeyp, &ip6->ip6_src, sizeof(struct in6_addr));
383 	hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp);
384 	memcpy(hashkeyp, &ip6->ip6_dst, sizeof(struct in6_addr));
385 	hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp);
386 	*hashkeyp = ip6f->ip6f_ident;
387 	bucket = jenkins_hash32(hashkey, nitems(hashkey), V_ip6qb_hashseed);
388 	bucket &= IP6REASS_HMASK;
389 	head = IP6QB_HEAD(bucket);
390 	IP6QB_LOCK(bucket);
391 
392 	/*
393 	 * Enforce upper bound on number of fragments.
394 	 * If maxfrag is 0, never accept fragments.
395 	 * If maxfrag is -1, accept all fragments without limitation.
396 	 */
397 	if (ip6_maxfrags < 0)
398 		;
399 	else if (atomic_load_int(&frag6_nfrags) >= (u_int)ip6_maxfrags)
400 		goto dropfrag;
401 
402 	for (q6 = head->ip6q_next; q6 != head; q6 = q6->ip6q_next)
403 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
404 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
405 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst)
406 #ifdef MAC
407 		    && mac_ip6q_match(m, q6)
408 #endif
409 		    )
410 			break;
411 
412 	if (q6 == head) {
413 		/*
414 		 * the first fragment to arrive, create a reassembly queue.
415 		 */
416 		first_frag = 1;
417 
418 		/*
419 		 * Enforce upper bound on number of fragmented packets
420 		 * for which we attempt reassembly;
421 		 * If maxfragpackets is 0, never accept fragments.
422 		 * If maxfragpackets is -1, accept all fragments without
423 		 * limitation.
424 		 */
425 		if (V_ip6_maxfragpackets < 0)
426 			;
427 		else if (V_ip6qb[bucket].count >= V_ip6_maxfragbucketsize ||
428 		    atomic_load_int(&V_frag6_nfragpackets) >=
429 		    (u_int)V_ip6_maxfragpackets)
430 			goto dropfrag;
431 		atomic_add_int(&V_frag6_nfragpackets, 1);
432 		q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FRAG6,
433 		    M_NOWAIT | M_ZERO);
434 		if (q6 == NULL)
435 			goto dropfrag;
436 #ifdef MAC
437 		if (mac_ip6q_init(q6, M_NOWAIT) != 0) {
438 			free(q6, M_FRAG6);
439 			goto dropfrag;
440 		}
441 		mac_ip6q_create(m, q6);
442 #endif
443 		frag6_insque_head(q6, head, bucket);
444 
445 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
446 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
447 #ifdef notyet
448 		q6->ip6q_nxtp	= (u_char *)nxtp;
449 #endif
450 		q6->ip6q_ident	= ip6f->ip6f_ident;
451 		q6->ip6q_ttl	= IPV6_FRAGTTL;
452 		q6->ip6q_src	= ip6->ip6_src;
453 		q6->ip6q_dst	= ip6->ip6_dst;
454 		q6->ip6q_ecn	=
455 		    (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
456 		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
457 
458 		q6->ip6q_nfrag = 0;
459 	}
460 
461 	/*
462 	 * If it's the 1st fragment, record the length of the
463 	 * unfragmentable part and the next header of the fragment header.
464 	 */
465 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
466 	if (fragoff == 0) {
467 		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
468 		    sizeof(struct ip6_frag);
469 		q6->ip6q_nxt = ip6f->ip6f_nxt;
470 	}
471 
472 	/*
473 	 * Check that the reassembled packet would not exceed 65535 bytes
474 	 * in size.
475 	 * If it would exceed, discard the fragment and return an ICMP error.
476 	 */
477 	if (q6->ip6q_unfrglen >= 0) {
478 		/* The 1st fragment has already arrived. */
479 		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
480 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
481 			    offset - sizeof(struct ip6_frag) +
482 			    offsetof(struct ip6_frag, ip6f_offlg));
483 			IP6QB_UNLOCK(bucket);
484 			return (IPPROTO_DONE);
485 		}
486 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
487 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
488 		    offset - sizeof(struct ip6_frag) +
489 		    offsetof(struct ip6_frag, ip6f_offlg));
490 		IP6QB_UNLOCK(bucket);
491 		return (IPPROTO_DONE);
492 	}
493 	/*
494 	 * If it's the first fragment, do the above check for each
495 	 * fragment already stored in the reassembly queue.
496 	 */
497 	if (fragoff == 0) {
498 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
499 		     af6 = af6dwn) {
500 			af6dwn = af6->ip6af_down;
501 
502 			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
503 			    IPV6_MAXPACKET) {
504 				struct mbuf *merr = IP6_REASS_MBUF(af6);
505 				struct ip6_hdr *ip6err;
506 				int erroff = af6->ip6af_offset;
507 
508 				/* dequeue the fragment. */
509 				frag6_deq(af6, bucket);
510 				free(af6, M_FRAG6);
511 
512 				/* adjust pointer. */
513 				ip6err = mtod(merr, struct ip6_hdr *);
514 
515 				/*
516 				 * Restore source and destination addresses
517 				 * in the erroneous IPv6 header.
518 				 */
519 				ip6err->ip6_src = q6->ip6q_src;
520 				ip6err->ip6_dst = q6->ip6q_dst;
521 
522 				icmp6_error(merr, ICMP6_PARAM_PROB,
523 				    ICMP6_PARAMPROB_HEADER,
524 				    erroff - sizeof(struct ip6_frag) +
525 				    offsetof(struct ip6_frag, ip6f_offlg));
526 			}
527 		}
528 	}
529 
530 	ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FRAG6,
531 	    M_NOWAIT | M_ZERO);
532 	if (ip6af == NULL)
533 		goto dropfrag;
534 	ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
535 	ip6af->ip6af_off = fragoff;
536 	ip6af->ip6af_frglen = frgpartlen;
537 	ip6af->ip6af_offset = offset;
538 	IP6_REASS_MBUF(ip6af) = m;
539 
540 	if (first_frag) {
541 		af6 = (struct ip6asfrag *)q6;
542 		goto insert;
543 	}
544 
545 	/*
546 	 * Handle ECN by comparing this segment with the first one;
547 	 * if CE is set, do not lose CE.
548 	 * drop if CE and not-ECT are mixed for the same packet.
549 	 */
550 	ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
551 	ecn0 = q6->ip6q_ecn;
552 	if (ecn == IPTOS_ECN_CE) {
553 		if (ecn0 == IPTOS_ECN_NOTECT) {
554 			free(ip6af, M_FRAG6);
555 			goto dropfrag;
556 		}
557 		if (ecn0 != IPTOS_ECN_CE)
558 			q6->ip6q_ecn = IPTOS_ECN_CE;
559 	}
560 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
561 		free(ip6af, M_FRAG6);
562 		goto dropfrag;
563 	}
564 
565 	/*
566 	 * Find a segment which begins after this one does.
567 	 */
568 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
569 	     af6 = af6->ip6af_down)
570 		if (af6->ip6af_off > ip6af->ip6af_off)
571 			break;
572 
573 	/*
574 	 * If the incoming framgent overlaps some existing fragments in
575 	 * the reassembly queue, drop it, since it is dangerous to override
576 	 * existing fragments from a security point of view.
577 	 * We don't know which fragment is the bad guy - here we trust
578 	 * fragment that came in earlier, with no real reason.
579 	 *
580 	 * Note: due to changes after disabling this part, mbuf passed to
581 	 * m_adj() below now does not meet the requirement.
582 	 */
583 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
584 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
585 			- ip6af->ip6af_off;
586 		if (i > 0) {
587 			free(ip6af, M_FRAG6);
588 			goto dropfrag;
589 		}
590 	}
591 	if (af6 != (struct ip6asfrag *)q6) {
592 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
593 		if (i > 0) {
594 			free(ip6af, M_FRAG6);
595 			goto dropfrag;
596 		}
597 	}
598 
599 insert:
600 #ifdef MAC
601 	if (!first_frag)
602 		mac_ip6q_update(m, q6);
603 #endif
604 
605 	/*
606 	 * Stick new segment in its place;
607 	 * check for complete reassembly.
608 	 * If not complete, check fragment limit.
609 	 * Move to front of packet queue, as we are
610 	 * the most recently active fragmented packet.
611 	 */
612 	frag6_enq(ip6af, af6->ip6af_up, bucket);
613 	atomic_add_int(&frag6_nfrags, 1);
614 	q6->ip6q_nfrag++;
615 	next = 0;
616 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
617 	     af6 = af6->ip6af_down) {
618 		if (af6->ip6af_off != next) {
619 			if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) {
620 				IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag);
621 				frag6_freef(q6, bucket);
622 			}
623 			IP6QB_UNLOCK(bucket);
624 			return IPPROTO_DONE;
625 		}
626 		next += af6->ip6af_frglen;
627 	}
628 	if (af6->ip6af_up->ip6af_mff) {
629 		if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) {
630 			IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag);
631 			frag6_freef(q6, bucket);
632 		}
633 		IP6QB_UNLOCK(bucket);
634 		return IPPROTO_DONE;
635 	}
636 
637 	/*
638 	 * Reassembly is complete; concatenate fragments.
639 	 */
640 	ip6af = q6->ip6q_down;
641 	t = m = IP6_REASS_MBUF(ip6af);
642 	af6 = ip6af->ip6af_down;
643 	frag6_deq(ip6af, bucket);
644 	while (af6 != (struct ip6asfrag *)q6) {
645 		m->m_pkthdr.csum_flags &=
646 		    IP6_REASS_MBUF(af6)->m_pkthdr.csum_flags;
647 		m->m_pkthdr.csum_data +=
648 		    IP6_REASS_MBUF(af6)->m_pkthdr.csum_data;
649 
650 		af6dwn = af6->ip6af_down;
651 		frag6_deq(af6, bucket);
652 		while (t->m_next)
653 			t = t->m_next;
654 		m_adj(IP6_REASS_MBUF(af6), af6->ip6af_offset);
655 		m_demote_pkthdr(IP6_REASS_MBUF(af6));
656 		m_cat(t, IP6_REASS_MBUF(af6));
657 		free(af6, M_FRAG6);
658 		af6 = af6dwn;
659 	}
660 
661 	while (m->m_pkthdr.csum_data & 0xffff0000)
662 		m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
663 		    (m->m_pkthdr.csum_data >> 16);
664 
665 	/* adjust offset to point where the original next header starts */
666 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
667 	free(ip6af, M_FRAG6);
668 	ip6 = mtod(m, struct ip6_hdr *);
669 	ip6->ip6_plen = htons((u_short)next + offset - sizeof(struct ip6_hdr));
670 	if (q6->ip6q_ecn == IPTOS_ECN_CE)
671 		ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20);
672 	nxt = q6->ip6q_nxt;
673 
674 	if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) {
675 		frag6_remque(q6, bucket);
676 		atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag);
677 #ifdef MAC
678 		mac_ip6q_destroy(q6);
679 #endif
680 		free(q6, M_FRAG6);
681 		atomic_subtract_int(&V_frag6_nfragpackets, 1);
682 
683 		goto dropfrag;
684 	}
685 
686 	/*
687 	 * Store NXT to the original.
688 	 */
689 	m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t),
690 	    (caddr_t)&nxt);
691 
692 	frag6_remque(q6, bucket);
693 	atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag);
694 #ifdef MAC
695 	mac_ip6q_reassemble(q6, m);
696 	mac_ip6q_destroy(q6);
697 #endif
698 	free(q6, M_FRAG6);
699 	atomic_subtract_int(&V_frag6_nfragpackets, 1);
700 
701 	if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
702 		int plen = 0;
703 		for (t = m; t; t = t->m_next)
704 			plen += t->m_len;
705 		m->m_pkthdr.len = plen;
706 	}
707 
708 #ifdef RSS
709 	mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc),
710 	    M_NOWAIT);
711 	if (mtag == NULL)
712 		goto dropfrag;
713 
714 	ip6dc = (struct ip6_direct_ctx *)(mtag + 1);
715 	ip6dc->ip6dc_nxt = nxt;
716 	ip6dc->ip6dc_off = offset;
717 
718 	m_tag_prepend(m, mtag);
719 #endif
720 
721 	IP6QB_UNLOCK(bucket);
722 	IP6STAT_INC(ip6s_reassembled);
723 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
724 
725 #ifdef RSS
726 	/*
727 	 * Queue/dispatch for reprocessing.
728 	 */
729 	netisr_dispatch(NETISR_IPV6_DIRECT, m);
730 	return IPPROTO_DONE;
731 #endif
732 
733 	/*
734 	 * Tell launch routine the next header
735 	 */
736 
737 	*mp = m;
738 	*offp = offset;
739 
740 	return nxt;
741 
742  dropfrag:
743 	IP6QB_UNLOCK(bucket);
744 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
745 	IP6STAT_INC(ip6s_fragdropped);
746 	m_freem(m);
747 	return IPPROTO_DONE;
748 }
749 
750 /*
751  * IPv6 reassembling timer processing;
752  * if a timer expires on a reassembly
753  * queue, discard it.
754  */
755 void
756 frag6_slowtimo(void)
757 {
758 	VNET_ITERATOR_DECL(vnet_iter);
759 	struct ip6q *head, *q6;
760 	uint32_t bucket;
761 
762 	VNET_LIST_RLOCK_NOSLEEP();
763 	VNET_FOREACH(vnet_iter) {
764 		CURVNET_SET(vnet_iter);
765 		for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
766 			IP6QB_LOCK(bucket);
767 			head = IP6QB_HEAD(bucket);
768 			q6 = head->ip6q_next;
769 			if (q6 == NULL) {
770 				/*
771 				 * XXXJTL: This should never happen. This
772 				 * should turn into an assertion.
773 				 */
774 				IP6QB_UNLOCK(bucket);
775 				continue;
776 			}
777 			while (q6 != head) {
778 				--q6->ip6q_ttl;
779 				q6 = q6->ip6q_next;
780 				if (q6->ip6q_prev->ip6q_ttl == 0) {
781 					IP6STAT_ADD(ip6s_fragtimeout,
782 						q6->ip6q_prev->ip6q_nfrag);
783 					/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
784 					frag6_freef(q6->ip6q_prev, bucket);
785 				}
786 			}
787 			/*
788 			 * If we are over the maximum number of fragments
789 			 * (due to the limit being lowered), drain off
790 			 * enough to get down to the new limit.
791 			 * Note that we drain all reassembly queues if
792 			 * maxfragpackets is 0 (fragmentation is disabled),
793 			 * and don't enforce a limit when maxfragpackets
794 			 * is negative.
795 			 */
796 			while ((V_ip6_maxfragpackets == 0 ||
797 			    (V_ip6_maxfragpackets > 0 &&
798 			    V_ip6qb[bucket].count > V_ip6_maxfragbucketsize)) &&
799 			    head->ip6q_prev != head) {
800 				IP6STAT_ADD(ip6s_fragoverflow,
801 					q6->ip6q_prev->ip6q_nfrag);
802 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
803 				frag6_freef(head->ip6q_prev, bucket);
804 			}
805 			IP6QB_UNLOCK(bucket);
806 		}
807 		/*
808 		 * If we are still over the maximum number of fragmented
809 		 * packets, drain off enough to get down to the new limit.
810 		 */
811 		bucket = 0;
812 		while (V_ip6_maxfragpackets >= 0 &&
813 		    atomic_load_int(&V_frag6_nfragpackets) >
814 		    (u_int)V_ip6_maxfragpackets) {
815 			IP6QB_LOCK(bucket);
816 			head = IP6QB_HEAD(bucket);
817 			if (head->ip6q_prev != head) {
818 				IP6STAT_ADD(ip6s_fragoverflow,
819 					q6->ip6q_prev->ip6q_nfrag);
820 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
821 				frag6_freef(head->ip6q_prev, bucket);
822 			}
823 			IP6QB_UNLOCK(bucket);
824 			bucket = (bucket + 1) % IP6REASS_NHASH;
825 		}
826 		CURVNET_RESTORE();
827 	}
828 	VNET_LIST_RUNLOCK_NOSLEEP();
829 }
830 
831 static void
832 frag6_change(void *tag)
833 {
834 	VNET_ITERATOR_DECL(vnet_iter);
835 
836 	ip6_maxfrags = IP6_MAXFRAGS;
837 	VNET_LIST_RLOCK_NOSLEEP();
838 	VNET_FOREACH(vnet_iter) {
839 		CURVNET_SET(vnet_iter);
840 		V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS;
841 		frag6_set_bucketsize();
842 		CURVNET_RESTORE();
843 	}
844 	VNET_LIST_RUNLOCK_NOSLEEP();
845 }
846 
847 /*
848  * Initialise reassembly queue and fragment identifier.
849  */
850 void
851 frag6_init(void)
852 {
853 	struct ip6q *q6;
854 	uint32_t bucket;
855 
856 	V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS;
857 	frag6_set_bucketsize();
858 	for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
859 		q6 = IP6QB_HEAD(bucket);
860 		q6->ip6q_next = q6->ip6q_prev = q6;
861 		mtx_init(&V_ip6qb[bucket].lock, "ip6qlock", NULL, MTX_DEF);
862 		V_ip6qb[bucket].count = 0;
863 	}
864 	V_ip6qb_hashseed = arc4random();
865 	V_ip6_maxfragsperpacket = 64;
866 	if (!IS_DEFAULT_VNET(curvnet))
867 		return;
868 
869 	ip6_maxfrags = IP6_MAXFRAGS;
870 	EVENTHANDLER_REGISTER(nmbclusters_change,
871 	    frag6_change, NULL, EVENTHANDLER_PRI_ANY);
872 }
873 
874 /*
875  * Drain off all datagram fragments.
876  */
877 void
878 frag6_drain(void)
879 {
880 	VNET_ITERATOR_DECL(vnet_iter);
881 	struct ip6q *head;
882 	uint32_t bucket;
883 
884 	VNET_LIST_RLOCK_NOSLEEP();
885 	VNET_FOREACH(vnet_iter) {
886 		CURVNET_SET(vnet_iter);
887 		for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
888 			if (IP6QB_TRYLOCK(bucket) == 0)
889 				continue;
890 			head = IP6QB_HEAD(bucket);
891 			while (head->ip6q_next != head) {
892 				IP6STAT_INC(ip6s_fragdropped);
893 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
894 				frag6_freef(head->ip6q_next, bucket);
895 			}
896 			IP6QB_UNLOCK(bucket);
897 		}
898 		CURVNET_RESTORE();
899 	}
900 	VNET_LIST_RUNLOCK_NOSLEEP();
901 }
902 
903 /*
904  * Put an ip fragment on a reassembly chain.
905  * Like insque, but pointers in middle of structure.
906  */
907 static void
908 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6,
909     uint32_t bucket __unused)
910 {
911 
912 	IP6QB_LOCK_ASSERT(bucket);
913 
914 	af6->ip6af_up = up6;
915 	af6->ip6af_down = up6->ip6af_down;
916 	up6->ip6af_down->ip6af_up = af6;
917 	up6->ip6af_down = af6;
918 }
919 
920 /*
921  * To frag6_enq as remque is to insque.
922  */
923 static void
924 frag6_deq(struct ip6asfrag *af6, uint32_t bucket __unused)
925 {
926 
927 	IP6QB_LOCK_ASSERT(bucket);
928 
929 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
930 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
931 }
932 
933 static void
934 frag6_insque_head(struct ip6q *new, struct ip6q *old, uint32_t bucket)
935 {
936 
937 	IP6QB_LOCK_ASSERT(bucket);
938 	KASSERT(IP6QB_HEAD(bucket) == old,
939 	    ("%s: attempt to insert at head of wrong bucket"
940 	    " (bucket=%u, old=%p)", __func__, bucket, old));
941 
942 	new->ip6q_prev = old;
943 	new->ip6q_next = old->ip6q_next;
944 	old->ip6q_next->ip6q_prev= new;
945 	old->ip6q_next = new;
946 	V_ip6qb[bucket].count++;
947 }
948 
949 static void
950 frag6_remque(struct ip6q *p6, uint32_t bucket)
951 {
952 
953 	IP6QB_LOCK_ASSERT(bucket);
954 
955 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
956 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
957 	V_ip6qb[bucket].count--;
958 }
959