xref: /xnu-8796.141.3/bsd/netinet6/frag6.c (revision 1b191cb58250d0705d8a51287127505aa4bc0789) !
1 /*
2  * Copyright (c) 2000-2023 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 
29 /*	$FreeBSD: src/sys/netinet6/frag6.c,v 1.2.2.5 2001/07/03 11:01:50 ume Exp $	*/
30 /*	$KAME: frag6.c,v 1.31 2001/05/17 13:45:34 jinmei Exp $	*/
31 
32 /*
33  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
34  * All rights reserved.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. Neither the name of the project nor the names of its contributors
45  *    may be used to endorse or promote products derived from this software
46  *    without specific prior written permission.
47  *
48  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  */
60 
61 #include <sys/param.h>
62 #include <sys/systm.h>
63 #include <sys/malloc.h>
64 #include <sys/mcache.h>
65 #include <sys/mbuf.h>
66 #include <sys/domain.h>
67 #include <sys/protosw.h>
68 #include <sys/socket.h>
69 #include <sys/errno.h>
70 #include <sys/time.h>
71 #include <sys/kernel.h>
72 #include <sys/syslog.h>
73 #include <kern/queue.h>
74 #include <kern/locks.h>
75 
76 #include <net/if.h>
77 #include <net/route.h>
78 
79 #include <netinet/in.h>
80 #include <netinet/in_var.h>
81 #include <netinet/ip.h>
82 #include <netinet/ip_var.h>
83 #include <netinet/ip6.h>
84 #include <netinet6/ip6_var.h>
85 #include <netinet/icmp6.h>
86 
87 #include <net/net_osdep.h>
88 #include <dev/random/randomdev.h>
89 
90 /*
91  * Define it to get a correct behavior on per-interface statistics.
92  */
93 #define IN6_IFSTAT_STRICT
94 struct  ip6asfrag {
95 	struct ip6asfrag *ip6af_down;
96 	struct ip6asfrag *ip6af_up;
97 	struct mbuf     *ip6af_m;
98 	int             ip6af_offset;   /* offset in ip6af_m to next header */
99 	int             ip6af_frglen;   /* fragmentable part length */
100 	int             ip6af_off;      /* fragment offset */
101 	u_int16_t       ip6af_mff;      /* more fragment bit in frag off */
102 };
103 
104 #define IP6_REASS_MBUF(ip6af) ((ip6af)->ip6af_m)
105 
106 MBUFQ_HEAD(fq6_head);
107 
108 static void frag6_save_context(struct mbuf *, int);
109 static void frag6_scrub_context(struct mbuf *);
110 static int frag6_restore_context(struct mbuf *);
111 
112 static void frag6_icmp6_paramprob_error(struct fq6_head *);
113 static void frag6_icmp6_timeex_error(struct fq6_head *);
114 
115 static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *);
116 static void frag6_deq(struct ip6asfrag *);
117 static void frag6_insque(struct ip6q *, struct ip6q *);
118 static void frag6_remque(struct ip6q *);
119 static void frag6_purgef(struct ip6q *, struct fq6_head *, struct fq6_head *);
120 static void frag6_freef(struct ip6q *, struct fq6_head *, struct fq6_head *);
121 
122 static int frag6_timeout_run;           /* frag6 timer is scheduled to run */
123 static void frag6_timeout(void *);
124 static void frag6_sched_timeout(void);
125 
126 static struct ip6q *ip6q_alloc(int);
127 static void ip6q_free(struct ip6q *);
128 static void ip6q_updateparams(void);
129 static struct ip6asfrag *ip6af_alloc(int);
130 static void ip6af_free(struct ip6asfrag *);
131 
132 static LCK_GRP_DECLARE(ip6qlock_grp, "ip6qlock");
133 static LCK_MTX_DECLARE(ip6qlock, &ip6qlock_grp);
134 
135 /* IPv6 fragment reassembly queues (protected by ip6qlock) */
136 static struct ip6q ip6q;                /* ip6 reassembly queues */
137 static int ip6_maxfragpackets;          /* max packets in reass queues */
138 static u_int32_t frag6_nfragpackets;    /* # of packets in reass queues */
139 static int ip6_maxfrags;                /* max fragments in reass queues */
140 static u_int32_t frag6_nfrags;          /* # of fragments in reass queues */
141 static u_int32_t ip6q_limit;            /* ip6q allocation limit */
142 static u_int32_t ip6q_count;            /* current # of allocated ip6q's */
143 static u_int32_t ip6af_limit;           /* ip6asfrag allocation limit */
144 static u_int32_t ip6af_count;           /* current # of allocated ip6asfrag's */
145 
146 static int sysctl_maxfragpackets SYSCTL_HANDLER_ARGS;
147 static int sysctl_maxfrags SYSCTL_HANDLER_ARGS;
148 
149 SYSCTL_DECL(_net_inet6_ip6);
150 
151 SYSCTL_PROC(_net_inet6_ip6, IPV6CTL_MAXFRAGPACKETS, maxfragpackets,
152     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &ip6_maxfragpackets, 0,
153     sysctl_maxfragpackets, "I",
154     "Maximum number of IPv6 fragment reassembly queue entries");
155 
156 SYSCTL_UINT(_net_inet6_ip6, OID_AUTO, fragpackets,
157     CTLFLAG_RD | CTLFLAG_LOCKED, &frag6_nfragpackets, 0,
158     "Current number of IPv6 fragment reassembly queue entries");
159 
160 SYSCTL_PROC(_net_inet6_ip6, IPV6CTL_MAXFRAGS, maxfrags,
161     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &ip6_maxfrags, 0,
162     sysctl_maxfrags, "I", "Maximum number of IPv6 fragments allowed");
163 
164 /*
165  * Initialise reassembly queue and fragment identifier.
166  */
167 void
frag6_init(void)168 frag6_init(void)
169 {
170 	/* ip6q_alloc() uses mbufs for IPv6 fragment queue structures */
171 	_CASSERT(sizeof(struct ip6q) <= _MLEN);
172 	/* ip6af_alloc() uses mbufs for IPv6 fragment queue structures */
173 	_CASSERT(sizeof(struct ip6asfrag) <= _MLEN);
174 
175 	lck_mtx_lock(&ip6qlock);
176 	/* Initialize IPv6 reassembly queue. */
177 	ip6q.ip6q_next = ip6q.ip6q_prev = &ip6q;
178 
179 	/* same limits as IPv4 */
180 	ip6_maxfragpackets = nmbclusters / 32;
181 	ip6_maxfrags = ip6_maxfragpackets * 2;
182 	ip6q_updateparams();
183 	lck_mtx_unlock(&ip6qlock);
184 }
185 
186 static void
frag6_save_context(struct mbuf * m,int val)187 frag6_save_context(struct mbuf *m, int val)
188 {
189 	m->m_pkthdr.pkt_hdr = (void *)(uintptr_t)val;
190 }
191 
192 static void
frag6_scrub_context(struct mbuf * m)193 frag6_scrub_context(struct mbuf *m)
194 {
195 	m->m_pkthdr.pkt_hdr = NULL;
196 }
197 
198 static int
frag6_restore_context(struct mbuf * m)199 frag6_restore_context(struct mbuf *m)
200 {
201 	return (int)m->m_pkthdr.pkt_hdr;
202 }
203 
204 /*
205  * Send any deferred ICMP param problem error messages; caller must not be
206  * holding ip6qlock and is expected to have saved the per-packet parameter
207  * value via frag6_save_context().
208  */
209 static void
frag6_icmp6_paramprob_error(struct fq6_head * diq6)210 frag6_icmp6_paramprob_error(struct fq6_head *diq6)
211 {
212 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_NOTOWNED);
213 
214 	if (!MBUFQ_EMPTY(diq6)) {
215 		struct mbuf *merr, *merr_tmp;
216 		int param;
217 		MBUFQ_FOREACH_SAFE(merr, diq6, merr_tmp) {
218 			MBUFQ_REMOVE(diq6, merr);
219 			MBUFQ_NEXT(merr) = NULL;
220 			param = frag6_restore_context(merr);
221 			frag6_scrub_context(merr);
222 			icmp6_error(merr, ICMP6_PARAM_PROB,
223 			    ICMP6_PARAMPROB_HEADER, param);
224 		}
225 	}
226 }
227 
228 /*
229  * Send any deferred ICMP time exceeded error messages;
230  * caller must not be holding ip6qlock.
231  */
232 static void
frag6_icmp6_timeex_error(struct fq6_head * diq6)233 frag6_icmp6_timeex_error(struct fq6_head *diq6)
234 {
235 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_NOTOWNED);
236 
237 	if (!MBUFQ_EMPTY(diq6)) {
238 		struct mbuf *m, *m_tmp;
239 		MBUFQ_FOREACH_SAFE(m, diq6, m_tmp) {
240 			MBUFQ_REMOVE(diq6, m);
241 			MBUFQ_NEXT(m) = NULL;
242 			icmp6_error_flag(m, ICMP6_TIME_EXCEEDED,
243 			    ICMP6_TIME_EXCEED_REASSEMBLY, 0, 0);
244 		}
245 	}
246 }
247 
248 /*
249  * In RFC2460, fragment and reassembly rule do not agree with each other,
250  * in terms of next header field handling in fragment header.
251  * While the sender will use the same value for all of the fragmented packets,
252  * receiver is suggested not to check the consistency.
253  *
254  * fragment rule (p20):
255  *	(2) A Fragment header containing:
256  *	The Next Header value that identifies the first header of
257  *	the Fragmentable Part of the original packet.
258  *		-> next header field is same for all fragments
259  *
260  * reassembly rule (p21):
261  *	The Next Header field of the last header of the Unfragmentable
262  *	Part is obtained from the Next Header field of the first
263  *	fragment's Fragment header.
264  *		-> should grab it from the first fragment only
265  *
266  * The following note also contradicts with fragment rule - noone is going to
267  * send different fragment with different next header field.
268  *
269  * additional note (p22):
270  *	The Next Header values in the Fragment headers of different
271  *	fragments of the same original packet may differ.  Only the value
272  *	from the Offset zero fragment packet is used for reassembly.
273  *		-> should grab it from the first fragment only
274  *
275  * There is no explicit reason given in the RFC.  Historical reason maybe?
276  */
277 /*
278  * Fragment input
279  */
280 int
frag6_input(struct mbuf ** mp,int * offp,int proto)281 frag6_input(struct mbuf **mp, int *offp, int proto)
282 {
283 #pragma unused(proto)
284 	struct mbuf *m = *mp, *t = NULL;
285 	struct ip6_hdr *ip6 = NULL;
286 	struct ip6_frag *ip6f = NULL;
287 	struct ip6q *q6 = NULL;
288 	struct ip6asfrag *af6 = NULL, *ip6af = NULL, *af6dwn = NULL;
289 	int offset = *offp, i = 0, next = 0;
290 	u_int8_t nxt = 0;
291 	int first_frag = 0;
292 	int fragoff = 0, frgpartlen = 0;        /* must be larger than u_int16_t */
293 	struct ifnet *dstifp = NULL;
294 	u_int8_t ecn = 0, ecn0 = 0;
295 	uint32_t csum = 0, csum_flags = 0;
296 	struct fq6_head diq6 = {};
297 	int locked = 0;
298 	boolean_t drop_fragq = FALSE;
299 	int local_ip6q_unfrglen;
300 	u_int8_t local_ip6q_nxt;
301 
302 	VERIFY(m->m_flags & M_PKTHDR);
303 
304 	MBUFQ_INIT(&diq6);      /* for deferred ICMP param problem errors */
305 
306 	/* Expect 32-bit aligned data pointer on strict-align platforms */
307 	MBUF_STRICT_DATA_ALIGNMENT_CHECK_32(m);
308 
309 	IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), goto done);
310 	ip6 = mtod(m, struct ip6_hdr *);
311 	ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset);
312 
313 #ifdef IN6_IFSTAT_STRICT
314 	/* find the destination interface of the packet. */
315 	if (m->m_pkthdr.pkt_flags & PKTF_IFAINFO) {
316 		uint32_t idx;
317 
318 		if (ip6_getdstifaddr_info(m, &idx, NULL) == 0) {
319 			if (idx > 0 && idx <= if_index) {
320 				ifnet_head_lock_shared();
321 				dstifp = ifindex2ifnet[idx];
322 				ifnet_head_done();
323 			}
324 		}
325 	}
326 #endif /* IN6_IFSTAT_STRICT */
327 
328 	/* we are violating the spec, this may not be the dst interface */
329 	if (dstifp == NULL) {
330 		dstifp = m->m_pkthdr.rcvif;
331 	}
332 
333 	/* jumbo payload can't contain a fragment header */
334 	if (ip6->ip6_plen == 0) {
335 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
336 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
337 		m = NULL;
338 		goto done;
339 	}
340 
341 	/*
342 	 * check whether fragment packet's fragment length is
343 	 * multiple of 8 octets.
344 	 * sizeof(struct ip6_frag) == 8
345 	 * sizeof(struct ip6_hdr) = 40
346 	 */
347 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
348 	    (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
349 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
350 		    offsetof(struct ip6_hdr, ip6_plen));
351 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
352 		m = NULL;
353 		goto done;
354 	}
355 
356 	/* If ip6_maxfragpackets or ip6_maxfrags is 0, never accept fragments */
357 	if (ip6_maxfragpackets == 0 || ip6_maxfrags == 0) {
358 		ip6stat.ip6s_fragments++;
359 		ip6stat.ip6s_fragdropped++;
360 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
361 		m_freem(m);
362 		m = NULL;
363 		goto done;
364 	}
365 
366 	/* offset now points to data portion */
367 	offset += sizeof(struct ip6_frag);
368 
369 	/*
370 	 * RFC 6946: Handle "atomic" fragments (offset and m bit set to 0)
371 	 * upfront, unrelated to any reassembly.  Just skip the fragment header.
372 	 */
373 	if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) {
374 		/*
375 		 * Mark packet as reassembled.
376 		 * In ICMPv6 processing, we drop certain
377 		 * NDP messages that are not expected to
378 		 * have fragment header based on recommendations
379 		 * against security vulnerability as described in
380 		 * RFC 6980.
381 		 * Treat atomic fragments as re-assembled packets as well.
382 		 */
383 		m->m_pkthdr.pkt_flags |= PKTF_REASSEMBLED;
384 		ip6stat.ip6s_atmfrag_rcvd++;
385 		in6_ifstat_inc(dstifp, ifs6_atmfrag_rcvd);
386 		*mp = m;
387 		*offp = offset;
388 		return ip6f->ip6f_nxt;
389 	}
390 
391 	/*
392 	 * Leverage partial checksum offload for simple UDP/IP fragments,
393 	 * as that is the most common case.
394 	 *
395 	 * Perform 1's complement adjustment of octets that got included/
396 	 * excluded in the hardware-calculated checksum value.  Also take
397 	 * care of any trailing bytes and subtract out their partial sum.
398 	 */
399 	if (ip6f->ip6f_nxt == IPPROTO_UDP &&
400 	    offset == (sizeof(*ip6) + sizeof(*ip6f)) &&
401 	    (m->m_pkthdr.csum_flags &
402 	    (CSUM_DATA_VALID | CSUM_PARTIAL | CSUM_PSEUDO_HDR)) ==
403 	    (CSUM_DATA_VALID | CSUM_PARTIAL)) {
404 		uint32_t start = m->m_pkthdr.csum_rx_start;
405 		uint32_t ip_len = (sizeof(*ip6) + ntohs(ip6->ip6_plen));
406 		int32_t trailer = (m_pktlen(m) - ip_len);
407 		uint32_t swbytes = (uint32_t)trailer;
408 
409 		csum = m->m_pkthdr.csum_rx_val;
410 
411 		ASSERT(trailer >= 0);
412 		if (start != offset || trailer != 0) {
413 			uint16_t s = 0, d = 0;
414 
415 			if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) {
416 				s = ip6->ip6_src.s6_addr16[1];
417 				ip6->ip6_src.s6_addr16[1] = 0;
418 			}
419 			if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) {
420 				d = ip6->ip6_dst.s6_addr16[1];
421 				ip6->ip6_dst.s6_addr16[1] = 0;
422 			}
423 
424 			/* callee folds in sum */
425 			csum = m_adj_sum16(m, start, offset,
426 			    (ip_len - offset), csum);
427 			if (offset > start) {
428 				swbytes += (offset - start);
429 			} else {
430 				swbytes += (start - offset);
431 			}
432 
433 			if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) {
434 				ip6->ip6_src.s6_addr16[1] = s;
435 			}
436 			if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) {
437 				ip6->ip6_dst.s6_addr16[1] = d;
438 			}
439 		}
440 		csum_flags = m->m_pkthdr.csum_flags;
441 
442 		if (swbytes != 0) {
443 			udp_in6_cksum_stats(swbytes);
444 		}
445 		if (trailer != 0) {
446 			m_adj(m, -trailer);
447 		}
448 	} else {
449 		csum = 0;
450 		csum_flags = 0;
451 	}
452 
453 	/* Invalidate checksum */
454 	m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
455 
456 	ip6stat.ip6s_fragments++;
457 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
458 
459 	lck_mtx_lock(&ip6qlock);
460 	locked = 1;
461 
462 	for (q6 = ip6q.ip6q_next; q6 != &ip6q; q6 = q6->ip6q_next) {
463 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
464 		    in6_are_addr_equal_scoped(&ip6->ip6_src, &q6->ip6q_src, ip6_input_getsrcifscope(m), q6->ip6q_src_ifscope) &&
465 		    in6_are_addr_equal_scoped(&ip6->ip6_dst, &q6->ip6q_dst, ip6_input_getdstifscope(m), q6->ip6q_dst_ifscope)) {
466 			break;
467 		}
468 	}
469 
470 	if (q6 == &ip6q) {
471 		/*
472 		 * Create a reassembly queue as this is the first fragment to
473 		 * arrive.
474 		 * By first frag, we don't mean the one with offset 0, but
475 		 * any of the fragments of the fragmented packet that has
476 		 * reached us first.
477 		 */
478 		first_frag = 1;
479 
480 		q6 = ip6q_alloc(M_DONTWAIT);
481 		if (q6 == NULL) {
482 			goto dropfrag;
483 		}
484 
485 		frag6_insque(q6, &ip6q);
486 		frag6_nfragpackets++;
487 
488 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
489 		q6->ip6q_down   = q6->ip6q_up = (struct ip6asfrag *)q6;
490 #ifdef notyet
491 		q6->ip6q_nxtp   = (u_char *)nxtp;
492 #endif
493 		q6->ip6q_ident  = ip6f->ip6f_ident;
494 		q6->ip6q_ttl    = IPV6_FRAGTTL;
495 		q6->ip6q_src    = ip6->ip6_src;
496 		q6->ip6q_dst    = ip6->ip6_dst;
497 		q6->ip6q_dst_ifscope = IN6_IS_SCOPE_EMBED(&q6->ip6q_dst) ? ip6_input_getdstifscope(m) : IFSCOPE_NONE;
498 		q6->ip6q_src_ifscope = IN6_IS_SCOPE_EMBED(&q6->ip6q_src) ? ip6_input_getsrcifscope(m) : IFSCOPE_NONE;
499 		q6->ip6q_ecn    =
500 		    (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
501 		q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */
502 
503 		q6->ip6q_nfrag = 0;
504 		q6->ip6q_flags = 0;
505 
506 		/*
507 		 * If the first fragment has valid checksum offload
508 		 * info, the rest of fragments are eligible as well.
509 		 */
510 		if (csum_flags != 0) {
511 			q6->ip6q_csum = csum;
512 			q6->ip6q_csum_flags = csum_flags;
513 		}
514 	}
515 
516 	if (q6->ip6q_flags & IP6QF_DIRTY) {
517 		goto dropfrag;
518 	}
519 
520 	local_ip6q_unfrglen = q6->ip6q_unfrglen;
521 	local_ip6q_nxt = q6->ip6q_nxt;
522 
523 	/*
524 	 * If it's the 1st fragment, record the length of the
525 	 * unfragmentable part and the next header of the fragment header.
526 	 * Assume the first fragement to arrive will be correct.
527 	 * We do not have any duplicate checks here yet so another packet
528 	 * with fragoff == 0 could come and overwrite the ip6q_unfrglen
529 	 * and worse, the next header, at any time.
530 	 */
531 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
532 	if (fragoff == 0 && local_ip6q_unfrglen == -1) {
533 		local_ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
534 		    sizeof(struct ip6_frag);
535 		local_ip6q_nxt = ip6f->ip6f_nxt;
536 		/* XXX ECN? */
537 	}
538 
539 	/*
540 	 * Check that the reassembled packet would not exceed 65535 bytes
541 	 * in size.
542 	 * If it would exceed, discard the fragment and return an ICMP error.
543 	 */
544 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
545 	if (local_ip6q_unfrglen >= 0) {
546 		/* The 1st fragment has already arrived. */
547 		if (local_ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
548 			lck_mtx_unlock(&ip6qlock);
549 			locked = 0;
550 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
551 			    offset - sizeof(struct ip6_frag) +
552 			    offsetof(struct ip6_frag, ip6f_offlg));
553 			m = NULL;
554 			goto done;
555 		}
556 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
557 		lck_mtx_unlock(&ip6qlock);
558 		locked = 0;
559 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
560 		    offset - sizeof(struct ip6_frag) +
561 		    offsetof(struct ip6_frag, ip6f_offlg));
562 		m = NULL;
563 		goto done;
564 	}
565 	/*
566 	 * If it's the first fragment, do the above check for each
567 	 * fragment already stored in the reassembly queue.
568 	 */
569 	if (fragoff == 0) {
570 		/*
571 		 * https://tools.ietf.org/html/rfc8200#page-20
572 		 * If the first fragment does not include all headers through an
573 		 * Upper-Layer header, then that fragment should be discarded and
574 		 * an ICMP Parameter Problem, Code 3, message should be sent to
575 		 * the source of the fragment, with the Pointer field set to zero.
576 		 */
577 		if (!ip6_pkt_has_ulp(m)) {
578 			lck_mtx_unlock(&ip6qlock);
579 			locked = 0;
580 			icmp6_error(m, ICMP6_PARAM_PROB,
581 			    ICMP6_PARAMPROB_FIRSTFRAG_INCOMP_HDR, 0);
582 			m = NULL;
583 			goto done;
584 		}
585 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
586 		    af6 = af6dwn) {
587 			af6dwn = af6->ip6af_down;
588 
589 			if (local_ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
590 			    IPV6_MAXPACKET) {
591 				struct mbuf *merr = IP6_REASS_MBUF(af6);
592 				struct ip6_hdr *ip6err;
593 				int erroff = af6->ip6af_offset;
594 
595 				/* dequeue the fragment. */
596 				frag6_deq(af6);
597 				ip6af_free(af6);
598 
599 				/* adjust pointer. */
600 				ip6err = mtod(merr, struct ip6_hdr *);
601 
602 				/*
603 				 * Restore source and destination addresses
604 				 * in the erroneous IPv6 header.
605 				 */
606 				ip6err->ip6_src = q6->ip6q_src;
607 				ip6err->ip6_dst = q6->ip6q_dst;
608 				ip6_output_setdstifscope(m, q6->ip6q_dst_ifscope, NULL);
609 				ip6_output_setsrcifscope(m, q6->ip6q_src_ifscope, NULL);
610 				frag6_save_context(merr,
611 				    erroff - sizeof(struct ip6_frag) +
612 				    offsetof(struct ip6_frag, ip6f_offlg));
613 
614 				MBUFQ_ENQUEUE(&diq6, merr);
615 			}
616 		}
617 	}
618 
619 	ip6af = ip6af_alloc(M_DONTWAIT);
620 	if (ip6af == NULL) {
621 		goto dropfrag;
622 	}
623 
624 	ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
625 	ip6af->ip6af_off = fragoff;
626 	ip6af->ip6af_frglen = frgpartlen;
627 	ip6af->ip6af_offset = offset;
628 	IP6_REASS_MBUF(ip6af) = m;
629 
630 	if (first_frag) {
631 		af6 = (struct ip6asfrag *)q6;
632 		goto insert;
633 	}
634 
635 	/*
636 	 * Handle ECN by comparing this segment with the first one;
637 	 * if CE is set, do not lose CE.
638 	 * drop if CE and not-ECT are mixed for the same packet.
639 	 */
640 	ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
641 	ecn0 = q6->ip6q_ecn;
642 	if (ecn == IPTOS_ECN_CE) {
643 		if (ecn0 == IPTOS_ECN_NOTECT) {
644 			ip6af_free(ip6af);
645 			goto dropfrag;
646 		}
647 		if (ecn0 != IPTOS_ECN_CE) {
648 			q6->ip6q_ecn = IPTOS_ECN_CE;
649 		}
650 	}
651 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
652 		ip6af_free(ip6af);
653 		goto dropfrag;
654 	}
655 
656 	/*
657 	 * Find a segment which begins after this one does.
658 	 */
659 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
660 	    af6 = af6->ip6af_down) {
661 		if (af6->ip6af_off > ip6af->ip6af_off) {
662 			break;
663 		}
664 	}
665 
666 	/*
667 	 * As per RFC 8200 reassembly rules, we MUST drop the entire
668 	 * chain of fragments for a packet to be assembled, if we receive
669 	 * any overlapping fragments.
670 	 * https://tools.ietf.org/html/rfc8200#page-20
671 	 *
672 	 * To avoid more conditional code, just reuse frag6_freef and defer
673 	 * its call to post fragment insertion in the queue.
674 	 */
675 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
676 		if (af6->ip6af_up->ip6af_off == ip6af->ip6af_off) {
677 			if (af6->ip6af_up->ip6af_frglen != ip6af->ip6af_frglen) {
678 				drop_fragq = TRUE;
679 			} else {
680 				/*
681 				 * XXX Ideally we should be comparing the entire
682 				 * packet here but for now just use off and fraglen
683 				 * to ignore a duplicate fragment.
684 				 */
685 				ip6af_free(ip6af);
686 				goto dropfrag;
687 			}
688 		} else {
689 			i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
690 			    - ip6af->ip6af_off;
691 			if (i > 0) {
692 				drop_fragq = TRUE;
693 			}
694 		}
695 	}
696 
697 	if (af6 != (struct ip6asfrag *)q6) {
698 		/*
699 		 * Given that we break when af6->ip6af_off > ip6af->ip6af_off,
700 		 * we shouldn't need a check for duplicate fragment here.
701 		 * For now just assert.
702 		 */
703 		VERIFY(af6->ip6af_off != ip6af->ip6af_off);
704 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
705 		if (i > 0) {
706 			drop_fragq = TRUE;
707 		}
708 	}
709 
710 	/*
711 	 * If this fragment contains similar checksum offload info
712 	 * as that of the existing ones, accumulate checksum.  Otherwise,
713 	 * invalidate checksum offload info for the entire datagram.
714 	 */
715 	if (csum_flags != 0 && csum_flags == q6->ip6q_csum_flags) {
716 		q6->ip6q_csum += csum;
717 	} else if (q6->ip6q_csum_flags != 0) {
718 		q6->ip6q_csum_flags = 0;
719 	}
720 
721 insert:
722 	/*
723 	 * Stick new segment in its place;
724 	 * check for complete reassembly.
725 	 * Move to front of packet queue, as we are
726 	 * the most recently active fragmented packet.
727 	 */
728 	frag6_enq(ip6af, af6->ip6af_up);
729 	frag6_nfrags++;
730 	q6->ip6q_nfrag++;
731 
732 	/*
733 	 * This holds true, when we receive overlapping fragments.
734 	 * We must silently drop all the fragments we have received
735 	 * so far.
736 	 * Also mark q6 as dirty, so as to not add any new fragments to it.
737 	 * Make sure even q6 marked dirty is kept till timer expires for
738 	 * reassembly and when that happens, silenty get rid of q6
739 	 */
740 	if (drop_fragq) {
741 		struct fq6_head dfq6 = {0};
742 		MBUFQ_INIT(&dfq6);      /* for deferred frees */
743 		q6->ip6q_flags |= IP6QF_DIRTY;
744 		/* Purge all the fragments but do not free q6 */
745 		frag6_purgef(q6, &dfq6, NULL);
746 		af6 = NULL;
747 
748 		/* free fragments that need to be freed */
749 		if (!MBUFQ_EMPTY(&dfq6)) {
750 			MBUFQ_DRAIN(&dfq6);
751 		}
752 		VERIFY(MBUFQ_EMPTY(&dfq6));
753 		/*
754 		 * Just in case the above logic got anything added
755 		 * to diq6, drain it.
756 		 * Please note that these mbufs are not present in the
757 		 * fragment queue and are added to diq6 for sending
758 		 * ICMPv6 error.
759 		 * Given that the current fragment was an overlapping
760 		 * fragment and the RFC requires us to not send any
761 		 * ICMPv6 errors while purging the entire queue.
762 		 * Just empty it out.
763 		 */
764 		if (!MBUFQ_EMPTY(&diq6)) {
765 			MBUFQ_DRAIN(&diq6);
766 		}
767 		VERIFY(MBUFQ_EMPTY(&diq6));
768 		/*
769 		 * MBUFQ_DRAIN would have drained all the mbufs
770 		 * in the fragment queue.
771 		 * This shouldn't be needed as we are returning IPPROTO_DONE
772 		 * from here but change the passed mbuf pointer to NULL.
773 		 */
774 		*mp = NULL;
775 		lck_mtx_unlock(&ip6qlock);
776 		return IPPROTO_DONE;
777 	}
778 
779 	/*
780 	 * We're keeping the fragment.
781 	 */
782 	q6->ip6q_unfrglen = local_ip6q_unfrglen;
783 	q6->ip6q_nxt = local_ip6q_nxt;
784 
785 	next = 0;
786 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
787 	    af6 = af6->ip6af_down) {
788 		if (af6->ip6af_off != next) {
789 			lck_mtx_unlock(&ip6qlock);
790 			locked = 0;
791 			m = NULL;
792 			goto done;
793 		}
794 		next += af6->ip6af_frglen;
795 	}
796 	if (af6->ip6af_up->ip6af_mff) {
797 		lck_mtx_unlock(&ip6qlock);
798 		locked = 0;
799 		m = NULL;
800 		goto done;
801 	}
802 
803 	/*
804 	 * Reassembly is complete; concatenate fragments.
805 	 */
806 	ip6af = q6->ip6q_down;
807 	t = m = IP6_REASS_MBUF(ip6af);
808 	af6 = ip6af->ip6af_down;
809 	frag6_deq(ip6af);
810 	while (af6 != (struct ip6asfrag *)q6) {
811 		af6dwn = af6->ip6af_down;
812 		frag6_deq(af6);
813 		while (t->m_next) {
814 			t = t->m_next;
815 		}
816 		t->m_next = IP6_REASS_MBUF(af6);
817 		m_adj(t->m_next, af6->ip6af_offset);
818 		ip6af_free(af6);
819 		af6 = af6dwn;
820 	}
821 
822 	/*
823 	 * Store partial hardware checksum info from the fragment queue;
824 	 * the receive start offset is set to 40 bytes (see code at the
825 	 * top of this routine.)
826 	 */
827 	if (q6->ip6q_csum_flags != 0) {
828 		csum = q6->ip6q_csum;
829 
830 		ADDCARRY(csum);
831 
832 		m->m_pkthdr.csum_rx_val = (u_int16_t)csum;
833 		m->m_pkthdr.csum_rx_start = sizeof(struct ip6_hdr);
834 		m->m_pkthdr.csum_flags = q6->ip6q_csum_flags;
835 	} else if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
836 	    (m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
837 		/* loopback checksums are always OK */
838 		m->m_pkthdr.csum_data = 0xffff;
839 		m->m_pkthdr.csum_flags = CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
840 	}
841 
842 	/* adjust offset to point where the original next header starts */
843 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
844 	ip6af_free(ip6af);
845 	ip6 = mtod(m, struct ip6_hdr *);
846 	ip6->ip6_plen = htons((uint16_t)(next + offset - sizeof(struct ip6_hdr)));
847 	ip6->ip6_src = q6->ip6q_src;
848 	ip6->ip6_dst = q6->ip6q_dst;
849 	ip6_output_setdstifscope(m, q6->ip6q_dst_ifscope, NULL);
850 	ip6_output_setsrcifscope(m, q6->ip6q_src_ifscope, NULL);
851 	if (q6->ip6q_ecn == IPTOS_ECN_CE) {
852 		ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20);
853 	}
854 
855 	nxt = q6->ip6q_nxt;
856 #ifdef notyet
857 	*q6->ip6q_nxtp = (u_char)(nxt & 0xff);
858 #endif
859 
860 	/* Delete frag6 header */
861 	if (m->m_len >= offset + sizeof(struct ip6_frag)) {
862 		/* This is the only possible case with !PULLDOWN_TEST */
863 		ovbcopy((caddr_t)ip6, (caddr_t)ip6 + sizeof(struct ip6_frag),
864 		    offset);
865 		m->m_data += sizeof(struct ip6_frag);
866 		m->m_len -= sizeof(struct ip6_frag);
867 	} else {
868 		/* this comes with no copy if the boundary is on cluster */
869 		if ((t = m_split(m, offset, M_DONTWAIT)) == NULL) {
870 			frag6_remque(q6);
871 			frag6_nfragpackets--;
872 			frag6_nfrags -= q6->ip6q_nfrag;
873 			ip6q_free(q6);
874 			goto dropfrag;
875 		}
876 		m_adj(t, sizeof(struct ip6_frag));
877 		m_cat(m, t);
878 	}
879 
880 	/*
881 	 * Store NXT to the original.
882 	 */
883 	{
884 		char *prvnxtp = ip6_get_prevhdr(m, offset); /* XXX */
885 		*prvnxtp = nxt;
886 	}
887 
888 	frag6_remque(q6);
889 	frag6_nfragpackets--;
890 	frag6_nfrags -= q6->ip6q_nfrag;
891 	ip6q_free(q6);
892 
893 	if (m->m_flags & M_PKTHDR) {    /* Isn't it always true? */
894 		m_fixhdr(m);
895 		/*
896 		 * Mark packet as reassembled
897 		 * In ICMPv6 processing, we drop certain
898 		 * NDP messages that are not expected to
899 		 * have fragment header based on recommendations
900 		 * against security vulnerability as described in
901 		 * RFC 6980.
902 		 */
903 		m->m_pkthdr.pkt_flags |= PKTF_REASSEMBLED;
904 	}
905 	ip6stat.ip6s_reassembled++;
906 
907 	/*
908 	 * Tell launch routine the next header
909 	 */
910 	*mp = m;
911 	*offp = offset;
912 
913 	/* arm the purge timer if not already and if there's work to do */
914 	frag6_sched_timeout();
915 	lck_mtx_unlock(&ip6qlock);
916 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
917 	frag6_icmp6_paramprob_error(&diq6);
918 	VERIFY(MBUFQ_EMPTY(&diq6));
919 	return nxt;
920 
921 done:
922 	VERIFY(m == NULL);
923 	*mp = m;
924 	if (!locked) {
925 		if (frag6_nfragpackets == 0) {
926 			frag6_icmp6_paramprob_error(&diq6);
927 			VERIFY(MBUFQ_EMPTY(&diq6));
928 			return IPPROTO_DONE;
929 		}
930 		lck_mtx_lock(&ip6qlock);
931 	}
932 	/* arm the purge timer if not already and if there's work to do */
933 	frag6_sched_timeout();
934 	lck_mtx_unlock(&ip6qlock);
935 	frag6_icmp6_paramprob_error(&diq6);
936 	VERIFY(MBUFQ_EMPTY(&diq6));
937 	return IPPROTO_DONE;
938 
939 dropfrag:
940 	ip6stat.ip6s_fragdropped++;
941 	/* arm the purge timer if not already and if there's work to do */
942 	frag6_sched_timeout();
943 	lck_mtx_unlock(&ip6qlock);
944 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
945 	m_freem(m);
946 	*mp = NULL;
947 	frag6_icmp6_paramprob_error(&diq6);
948 	VERIFY(MBUFQ_EMPTY(&diq6));
949 	return IPPROTO_DONE;
950 }
951 
952 /*
953  * This routine removes the enqueued frames from the passed fragment
954  * header and enqueues those to dfq6 which is an out-arg for the dequeued
955  * fragments.
956  * If the caller also provides diq6, this routine also enqueues the 0 offset
957  * fragment to that list as it potentially gets used by the caller
958  * to prepare the relevant ICMPv6 error message (time exceeded or
959  * param problem).
960  * It leaves the fragment header object (q6) intact.
961  */
962 static void
frag6_purgef(struct ip6q * q6,struct fq6_head * dfq6,struct fq6_head * diq6)963 frag6_purgef(struct ip6q *q6, struct fq6_head *dfq6, struct fq6_head *diq6)
964 {
965 	struct ip6asfrag *af6 = NULL;
966 	struct ip6asfrag *down6 = NULL;
967 
968 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_OWNED);
969 
970 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
971 	    af6 = down6) {
972 		struct mbuf *m = IP6_REASS_MBUF(af6);
973 
974 		down6 = af6->ip6af_down;
975 		frag6_deq(af6);
976 
977 		/*
978 		 * If caller wants to generate ICMP time-exceeded,
979 		 * as indicated by the argument diq6, return it for
980 		 * the first fragment and add others to the fragment
981 		 * free queue.
982 		 */
983 		if (af6->ip6af_off == 0 && diq6 != NULL) {
984 			struct ip6_hdr *ip6;
985 
986 			/* adjust pointer */
987 			ip6 = mtod(m, struct ip6_hdr *);
988 
989 			/* restore source and destination addresses */
990 			ip6->ip6_src = q6->ip6q_src;
991 			ip6->ip6_dst = q6->ip6q_dst;
992 			ip6_output_setdstifscope(m, q6->ip6q_dst_ifscope, NULL);
993 			ip6_output_setsrcifscope(m, q6->ip6q_src_ifscope, NULL);
994 			MBUFQ_ENQUEUE(diq6, m);
995 		} else {
996 			MBUFQ_ENQUEUE(dfq6, m);
997 		}
998 		ip6af_free(af6);
999 	}
1000 }
1001 
1002 /*
1003  * This routine removes the enqueued frames from the passed fragment
1004  * header and enqueues those to dfq6 which is an out-arg for the dequeued
1005  * fragments.
1006  * If the caller also provides diq6, this routine also enqueues the 0 offset
1007  * fragment to that list as it potentially gets used by the caller
1008  * to prepare the relevant ICMPv6 error message (time exceeded or
1009  * param problem).
1010  * It also remove the fragment header object from the queue and frees it.
1011  */
1012 static void
frag6_freef(struct ip6q * q6,struct fq6_head * dfq6,struct fq6_head * diq6)1013 frag6_freef(struct ip6q *q6, struct fq6_head *dfq6, struct fq6_head *diq6)
1014 {
1015 	frag6_purgef(q6, dfq6, diq6);
1016 	frag6_remque(q6);
1017 	frag6_nfragpackets--;
1018 	frag6_nfrags -= q6->ip6q_nfrag;
1019 	ip6q_free(q6);
1020 }
1021 
1022 /*
1023  * Put an ip fragment on a reassembly chain.
1024  * Like insque, but pointers in middle of structure.
1025  */
1026 void
frag6_enq(struct ip6asfrag * af6,struct ip6asfrag * up6)1027 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6)
1028 {
1029 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_OWNED);
1030 
1031 	af6->ip6af_up = up6;
1032 	af6->ip6af_down = up6->ip6af_down;
1033 	up6->ip6af_down->ip6af_up = af6;
1034 	up6->ip6af_down = af6;
1035 }
1036 
1037 /*
1038  * To frag6_enq as remque is to insque.
1039  */
1040 void
frag6_deq(struct ip6asfrag * af6)1041 frag6_deq(struct ip6asfrag *af6)
1042 {
1043 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_OWNED);
1044 
1045 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
1046 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
1047 }
1048 
1049 void
frag6_insque(struct ip6q * new,struct ip6q * old)1050 frag6_insque(struct ip6q *new, struct ip6q *old)
1051 {
1052 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_OWNED);
1053 
1054 	new->ip6q_prev = old;
1055 	new->ip6q_next = old->ip6q_next;
1056 	old->ip6q_next->ip6q_prev = new;
1057 	old->ip6q_next = new;
1058 }
1059 
1060 void
frag6_remque(struct ip6q * p6)1061 frag6_remque(struct ip6q *p6)
1062 {
1063 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_OWNED);
1064 
1065 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
1066 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
1067 }
1068 
1069 /*
1070  * IPv6 reassembling timer processing;
1071  * if a timer expires on a reassembly
1072  * queue, discard it.
1073  */
1074 static void
frag6_timeout(void * arg)1075 frag6_timeout(void *arg)
1076 {
1077 #pragma unused(arg)
1078 	struct fq6_head dfq6, diq6;
1079 	struct fq6_head *diq6_tmp = NULL;
1080 	struct ip6q *q6;
1081 
1082 	MBUFQ_INIT(&dfq6);      /* for deferred frees */
1083 	MBUFQ_INIT(&diq6);      /* for deferred ICMP time exceeded errors */
1084 
1085 	/*
1086 	 * Update coarse-grained networking timestamp (in sec.); the idea
1087 	 * is to piggy-back on the timeout callout to update the counter
1088 	 * returnable via net_uptime().
1089 	 */
1090 	net_update_uptime();
1091 
1092 	lck_mtx_lock(&ip6qlock);
1093 	q6 = ip6q.ip6q_next;
1094 	if (q6) {
1095 		while (q6 != &ip6q) {
1096 			--q6->ip6q_ttl;
1097 			q6 = q6->ip6q_next;
1098 			if (q6->ip6q_prev->ip6q_ttl == 0) {
1099 				ip6stat.ip6s_fragtimeout++;
1100 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
1101 				/*
1102 				 * Avoid sending ICMPv6 Time Exceeded for fragment headers
1103 				 * that are marked dirty.
1104 				 */
1105 				diq6_tmp = (q6->ip6q_prev->ip6q_flags & IP6QF_DIRTY) ?
1106 				    NULL : &diq6;
1107 				frag6_freef(q6->ip6q_prev, &dfq6, diq6_tmp);
1108 			}
1109 		}
1110 	}
1111 	/*
1112 	 * If we are over the maximum number of fragments
1113 	 * (due to the limit being lowered), drain off
1114 	 * enough to get down to the new limit.
1115 	 */
1116 	if (ip6_maxfragpackets >= 0) {
1117 		while (frag6_nfragpackets > (unsigned)ip6_maxfragpackets &&
1118 		    ip6q.ip6q_prev) {
1119 			ip6stat.ip6s_fragoverflow++;
1120 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
1121 			/*
1122 			 * Avoid sending ICMPv6 Time Exceeded for fragment headers
1123 			 * that are marked dirty.
1124 			 */
1125 			diq6_tmp = (ip6q.ip6q_prev->ip6q_flags & IP6QF_DIRTY) ?
1126 			    NULL : &diq6;
1127 			frag6_freef(ip6q.ip6q_prev, &dfq6, diq6_tmp);
1128 		}
1129 	}
1130 	/* re-arm the purge timer if there's work to do */
1131 	frag6_timeout_run = 0;
1132 	frag6_sched_timeout();
1133 	lck_mtx_unlock(&ip6qlock);
1134 
1135 	/* free fragments that need to be freed */
1136 	if (!MBUFQ_EMPTY(&dfq6)) {
1137 		MBUFQ_DRAIN(&dfq6);
1138 	}
1139 
1140 	frag6_icmp6_timeex_error(&diq6);
1141 
1142 	VERIFY(MBUFQ_EMPTY(&dfq6));
1143 	VERIFY(MBUFQ_EMPTY(&diq6));
1144 }
1145 
1146 static void
frag6_sched_timeout(void)1147 frag6_sched_timeout(void)
1148 {
1149 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_OWNED);
1150 
1151 	if (!frag6_timeout_run && frag6_nfragpackets > 0) {
1152 		frag6_timeout_run = 1;
1153 		timeout(frag6_timeout, NULL, hz);
1154 	}
1155 }
1156 
1157 /*
1158  * Drain off all datagram fragments.
1159  */
1160 void
frag6_drain(void)1161 frag6_drain(void)
1162 {
1163 	struct fq6_head dfq6, diq6;
1164 	struct fq6_head *diq6_tmp = NULL;
1165 
1166 	MBUFQ_INIT(&dfq6);      /* for deferred frees */
1167 	MBUFQ_INIT(&diq6);      /* for deferred ICMP time exceeded errors */
1168 
1169 	lck_mtx_lock(&ip6qlock);
1170 	while (ip6q.ip6q_next != &ip6q) {
1171 		ip6stat.ip6s_fragdropped++;
1172 		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
1173 		/*
1174 		 * Avoid sending ICMPv6 Time Exceeded for fragment headers
1175 		 * that are marked dirty.
1176 		 */
1177 		diq6_tmp = (ip6q.ip6q_next->ip6q_flags & IP6QF_DIRTY) ?
1178 		    NULL : &diq6;
1179 		frag6_freef(ip6q.ip6q_next, &dfq6, diq6_tmp);
1180 	}
1181 	lck_mtx_unlock(&ip6qlock);
1182 
1183 	/* free fragments that need to be freed */
1184 	if (!MBUFQ_EMPTY(&dfq6)) {
1185 		MBUFQ_DRAIN(&dfq6);
1186 	}
1187 
1188 	frag6_icmp6_timeex_error(&diq6);
1189 
1190 	VERIFY(MBUFQ_EMPTY(&dfq6));
1191 	VERIFY(MBUFQ_EMPTY(&diq6));
1192 }
1193 
1194 static struct ip6q *
ip6q_alloc(int how)1195 ip6q_alloc(int how)
1196 {
1197 	struct mbuf *t;
1198 	struct ip6q *q6;
1199 
1200 	/*
1201 	 * See comments in ip6q_updateparams().  Keep the count separate
1202 	 * from frag6_nfragpackets since the latter represents the elements
1203 	 * already in the reassembly queues.
1204 	 */
1205 	if (ip6q_limit > 0 && ip6q_count > ip6q_limit) {
1206 		return NULL;
1207 	}
1208 
1209 	t = m_get(how, MT_FTABLE);
1210 	if (t != NULL) {
1211 		atomic_add_32(&ip6q_count, 1);
1212 		q6 = mtod(t, struct ip6q *);
1213 		bzero(q6, sizeof(*q6));
1214 	} else {
1215 		q6 = NULL;
1216 	}
1217 	return q6;
1218 }
1219 
1220 static void
ip6q_free(struct ip6q * q6)1221 ip6q_free(struct ip6q *q6)
1222 {
1223 	(void) m_free(dtom(q6));
1224 	atomic_add_32(&ip6q_count, -1);
1225 }
1226 
1227 static struct ip6asfrag *
ip6af_alloc(int how)1228 ip6af_alloc(int how)
1229 {
1230 	struct mbuf *t;
1231 	struct ip6asfrag *af6;
1232 
1233 	/*
1234 	 * See comments in ip6q_updateparams().  Keep the count separate
1235 	 * from frag6_nfrags since the latter represents the elements
1236 	 * already in the reassembly queues.
1237 	 */
1238 	if (ip6af_limit > 0 && ip6af_count > ip6af_limit) {
1239 		return NULL;
1240 	}
1241 
1242 	t = m_get(how, MT_FTABLE);
1243 	if (t != NULL) {
1244 		atomic_add_32(&ip6af_count, 1);
1245 		af6 = mtod(t, struct ip6asfrag *);
1246 		bzero(af6, sizeof(*af6));
1247 	} else {
1248 		af6 = NULL;
1249 	}
1250 	return af6;
1251 }
1252 
1253 static void
ip6af_free(struct ip6asfrag * af6)1254 ip6af_free(struct ip6asfrag *af6)
1255 {
1256 	(void) m_free(dtom(af6));
1257 	atomic_add_32(&ip6af_count, -1);
1258 }
1259 
1260 static void
ip6q_updateparams(void)1261 ip6q_updateparams(void)
1262 {
1263 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_OWNED);
1264 	/*
1265 	 * -1 for unlimited allocation.
1266 	 */
1267 	if (ip6_maxfragpackets < 0) {
1268 		ip6q_limit = 0;
1269 	}
1270 	if (ip6_maxfrags < 0) {
1271 		ip6af_limit = 0;
1272 	}
1273 	/*
1274 	 * Positive number for specific bound.
1275 	 */
1276 	if (ip6_maxfragpackets > 0) {
1277 		ip6q_limit = ip6_maxfragpackets;
1278 	}
1279 	if (ip6_maxfrags > 0) {
1280 		ip6af_limit = ip6_maxfrags;
1281 	}
1282 	/*
1283 	 * Zero specifies no further fragment queue allocation -- set the
1284 	 * bound very low, but rely on implementation elsewhere to actually
1285 	 * prevent allocation and reclaim current queues.
1286 	 */
1287 	if (ip6_maxfragpackets == 0) {
1288 		ip6q_limit = 1;
1289 	}
1290 	if (ip6_maxfrags == 0) {
1291 		ip6af_limit = 1;
1292 	}
1293 	/*
1294 	 * Arm the purge timer if not already and if there's work to do
1295 	 */
1296 	frag6_sched_timeout();
1297 }
1298 
1299 static int
1300 sysctl_maxfragpackets SYSCTL_HANDLER_ARGS
1301 {
1302 #pragma unused(arg1, arg2)
1303 	int error, i;
1304 
1305 	lck_mtx_lock(&ip6qlock);
1306 	i = ip6_maxfragpackets;
1307 	error = sysctl_handle_int(oidp, &i, 0, req);
1308 	if (error || req->newptr == USER_ADDR_NULL) {
1309 		goto done;
1310 	}
1311 	/* impose bounds */
1312 	if (i < -1 || i > (nmbclusters / 4)) {
1313 		error = EINVAL;
1314 		goto done;
1315 	}
1316 	ip6_maxfragpackets = i;
1317 	ip6q_updateparams();
1318 done:
1319 	lck_mtx_unlock(&ip6qlock);
1320 	return error;
1321 }
1322 
1323 static int
1324 sysctl_maxfrags SYSCTL_HANDLER_ARGS
1325 {
1326 #pragma unused(arg1, arg2)
1327 	int error, i;
1328 
1329 	lck_mtx_lock(&ip6qlock);
1330 	i = ip6_maxfrags;
1331 	error = sysctl_handle_int(oidp, &i, 0, req);
1332 	if (error || req->newptr == USER_ADDR_NULL) {
1333 		goto done;
1334 	}
1335 	/* impose bounds */
1336 	if (i < -1 || i > (nmbclusters / 4)) {
1337 		error = EINVAL;
1338 		goto done;
1339 	}
1340 	ip6_maxfrags = i;
1341 	ip6q_updateparams();    /* see if we need to arm timer */
1342 done:
1343 	lck_mtx_unlock(&ip6qlock);
1344 	return error;
1345 }
1346