xref: /xnu-11417.140.69/bsd/netinet6/frag6.c (revision 43a90889846e00bfb5cf1d255cdc0a701a1e05a4)
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/droptap.h>
77 #include <net/if.h>
78 #include <net/route.h>
79 
80 #include <netinet/in.h>
81 #include <netinet/in_var.h>
82 #include <netinet/ip.h>
83 #include <netinet/ip_var.h>
84 #include <netinet/ip6.h>
85 #include <netinet6/ip6_var.h>
86 #include <netinet/icmp6.h>
87 
88 #include <net/net_osdep.h>
89 #include <dev/random/randomdev.h>
90 
91 /*
92  * Define it to get a correct behavior on per-interface statistics.
93  */
94 #define IN6_IFSTAT_STRICT
95 struct  ip6asfrag {
96 	struct ip6asfrag *ip6af_down;
97 	struct ip6asfrag *ip6af_up;
98 	struct mbuf     *ip6af_m;
99 	int             ip6af_offset;   /* offset in ip6af_m to next header */
100 	int             ip6af_frglen;   /* fragmentable part length */
101 	int             ip6af_off;      /* fragment offset */
102 	u_int16_t       ip6af_mff;      /* more fragment bit in frag off */
103 };
104 
105 #define IP6_REASS_MBUF(ip6af) ((ip6af)->ip6af_m)
106 
107 MBUFQ_HEAD(fq6_head);
108 
109 static void frag6_save_context(struct mbuf *, uintptr_t);
110 static void frag6_scrub_context(struct mbuf *);
111 static int frag6_restore_context(struct mbuf *);
112 
113 static void frag6_icmp6_paramprob_error(struct fq6_head *);
114 static void frag6_icmp6_timeex_error(struct fq6_head *);
115 
116 static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *);
117 static void frag6_deq(struct ip6asfrag *);
118 static void frag6_insque(struct ip6q *, struct ip6q *);
119 static void frag6_remque(struct ip6q *);
120 static void frag6_purgef(struct ip6q *, struct fq6_head *, struct fq6_head *);
121 static void frag6_freef(struct ip6q *, struct fq6_head *, struct fq6_head *);
122 
123 static int frag6_timeout_run;           /* frag6 timer is scheduled to run */
124 static void frag6_timeout(void *);
125 static void frag6_sched_timeout(void);
126 
127 static struct ip6q *ip6q_alloc(void);
128 static void ip6q_free(struct ip6q *);
129 static void ip6q_updateparams(void);
130 static struct ip6asfrag *ip6af_alloc(void);
131 static void ip6af_free(struct ip6asfrag *);
132 
133 static LCK_GRP_DECLARE(ip6qlock_grp, "ip6qlock");
134 static LCK_MTX_DECLARE(ip6qlock, &ip6qlock_grp);
135 
136 /* IPv6 fragment reassembly queues (protected by ip6qlock) */
137 static struct ip6q ip6q;                /* ip6 reassembly queues */
138 static int ip6_maxfragpackets;          /* max packets in reass queues */
139 static u_int32_t frag6_nfragpackets;    /* # of packets in reass queues */
140 static int ip6_maxfrags;                /* max fragments in reass queues */
141 static u_int32_t frag6_nfrags;          /* # of fragments in reass queues */
142 static u_int32_t ip6q_limit;            /* ip6q allocation limit */
143 static u_int32_t ip6q_count;            /* current # of allocated ip6q's */
144 static u_int32_t ip6af_limit;           /* ip6asfrag allocation limit */
145 static u_int32_t ip6af_count;           /* current # of allocated ip6asfrag's */
146 
147 static int sysctl_maxfragpackets SYSCTL_HANDLER_ARGS;
148 static int sysctl_maxfrags SYSCTL_HANDLER_ARGS;
149 
150 SYSCTL_DECL(_net_inet6_ip6);
151 
152 SYSCTL_PROC(_net_inet6_ip6, IPV6CTL_MAXFRAGPACKETS, maxfragpackets,
153     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &ip6_maxfragpackets, 0,
154     sysctl_maxfragpackets, "I",
155     "Maximum number of IPv6 fragment reassembly queue entries");
156 
157 SYSCTL_UINT(_net_inet6_ip6, OID_AUTO, fragpackets,
158     CTLFLAG_RD | CTLFLAG_LOCKED, &frag6_nfragpackets, 0,
159     "Current number of IPv6 fragment reassembly queue entries");
160 
161 SYSCTL_PROC(_net_inet6_ip6, IPV6CTL_MAXFRAGS, maxfrags,
162     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &ip6_maxfrags, 0,
163     sysctl_maxfrags, "I", "Maximum number of IPv6 fragments allowed");
164 
165 /*
166  * Initialise reassembly queue and fragment identifier.
167  */
168 void
frag6_init(void)169 frag6_init(void)
170 {
171 	lck_mtx_lock(&ip6qlock);
172 	/* Initialize IPv6 reassembly queue. */
173 	ip6q.ip6q_next = ip6q.ip6q_prev = &ip6q;
174 
175 	/* same limits as IPv4 */
176 	ip6_maxfragpackets = nmbclusters / 32;
177 	ip6_maxfrags = ip6_maxfragpackets * 2;
178 	ip6q_updateparams();
179 	lck_mtx_unlock(&ip6qlock);
180 }
181 
182 static void
frag6_save_context(struct mbuf * m,uintptr_t val)183 frag6_save_context(struct mbuf *m, uintptr_t val)
184 {
185 	m->m_pkthdr.pkt_hdr = __unsafe_forge_single(void *, val);
186 }
187 
188 static void
frag6_scrub_context(struct mbuf * m)189 frag6_scrub_context(struct mbuf *m)
190 {
191 	m->m_pkthdr.pkt_hdr = NULL;
192 }
193 
194 static int
frag6_restore_context(struct mbuf * m)195 frag6_restore_context(struct mbuf *m)
196 {
197 	return (int)m->m_pkthdr.pkt_hdr;
198 }
199 
200 /*
201  * Send any deferred ICMP param problem error messages; caller must not be
202  * holding ip6qlock and is expected to have saved the per-packet parameter
203  * value via frag6_save_context().
204  */
205 static void
frag6_icmp6_paramprob_error(struct fq6_head * diq6)206 frag6_icmp6_paramprob_error(struct fq6_head *diq6)
207 {
208 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_NOTOWNED);
209 
210 	if (!MBUFQ_EMPTY(diq6)) {
211 		mbuf_ref_t merr, merr_tmp;
212 		int param;
213 		MBUFQ_FOREACH_SAFE(merr, diq6, merr_tmp) {
214 			MBUFQ_REMOVE(diq6, merr);
215 			MBUFQ_NEXT(merr) = NULL;
216 			param = frag6_restore_context(merr);
217 			frag6_scrub_context(merr);
218 			icmp6_error(merr, ICMP6_PARAM_PROB,
219 			    ICMP6_PARAMPROB_HEADER, param);
220 		}
221 	}
222 }
223 
224 /*
225  * Send any deferred ICMP time exceeded error messages;
226  * caller must not be holding ip6qlock.
227  */
228 static void
frag6_icmp6_timeex_error(struct fq6_head * diq6)229 frag6_icmp6_timeex_error(struct fq6_head *diq6)
230 {
231 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_NOTOWNED);
232 
233 	if (!MBUFQ_EMPTY(diq6)) {
234 		mbuf_ref_t m, m_tmp;
235 		MBUFQ_FOREACH_SAFE(m, diq6, m_tmp) {
236 			MBUFQ_REMOVE(diq6, m);
237 			MBUFQ_NEXT(m) = NULL;
238 			icmp6_error_flag(m, ICMP6_TIME_EXCEEDED,
239 			    ICMP6_TIME_EXCEED_REASSEMBLY, 0, 0);
240 		}
241 	}
242 }
243 
244 /*
245  * In RFC2460, fragment and reassembly rule do not agree with each other,
246  * in terms of next header field handling in fragment header.
247  * While the sender will use the same value for all of the fragmented packets,
248  * receiver is suggested not to check the consistency.
249  *
250  * fragment rule (p20):
251  *	(2) A Fragment header containing:
252  *	The Next Header value that identifies the first header of
253  *	the Fragmentable Part of the original packet.
254  *		-> next header field is same for all fragments
255  *
256  * reassembly rule (p21):
257  *	The Next Header field of the last header of the Unfragmentable
258  *	Part is obtained from the Next Header field of the first
259  *	fragment's Fragment header.
260  *		-> should grab it from the first fragment only
261  *
262  * The following note also contradicts with fragment rule - noone is going to
263  * send different fragment with different next header field.
264  *
265  * additional note (p22):
266  *	The Next Header values in the Fragment headers of different
267  *	fragments of the same original packet may differ.  Only the value
268  *	from the Offset zero fragment packet is used for reassembly.
269  *		-> should grab it from the first fragment only
270  *
271  * There is no explicit reason given in the RFC.  Historical reason maybe?
272  */
273 /*
274  * Fragment input
275  */
276 int
frag6_input(struct mbuf ** mp,int * offp,int proto)277 frag6_input(struct mbuf **mp, int *offp, int proto)
278 {
279 #pragma unused(proto)
280 	mbuf_ref_t m = *mp, t = NULL;
281 	struct ip6_hdr *ip6 = NULL;
282 	struct ip6_frag *__single ip6f = NULL;
283 	struct ip6q *__single q6 = NULL;
284 	struct ip6asfrag *__single af6 = NULL, *__single ip6af = NULL, *__single af6dwn = NULL;
285 	int offset = *offp, i = 0, next = 0;
286 	u_int8_t nxt = 0;
287 	int first_frag = 0;
288 	int fragoff = 0, frgpartlen = 0;        /* must be larger than u_int16_t */
289 	ifnet_ref_t dstifp = NULL;
290 	u_int8_t ecn = 0, ecn0 = 0;
291 	uint32_t csum = 0, csum_flags = 0;
292 	struct fq6_head diq6 = {};
293 	int locked = 0;
294 	boolean_t drop_fragq = FALSE;
295 	int local_ip6q_unfrglen;
296 	u_int8_t local_ip6q_nxt;
297 	drop_reason_t drop_reason = DROP_REASON_UNSPECIFIED;
298 
299 	VERIFY(m->m_flags & M_PKTHDR);
300 
301 	MBUFQ_INIT(&diq6);      /* for deferred ICMP param problem errors */
302 
303 	/* Expect 32-bit aligned data pointer on strict-align platforms */
304 	MBUF_STRICT_DATA_ALIGNMENT_CHECK_32(m);
305 
306 	IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), goto done);
307 	ip6 = mtod(m, struct ip6_hdr *);
308 	ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset);
309 
310 #ifdef IN6_IFSTAT_STRICT
311 	/* find the destination interface of the packet. */
312 	if (m->m_pkthdr.pkt_flags & PKTF_IFAINFO) {
313 		uint32_t idx;
314 
315 		if (ip6_getdstifaddr_info(m, &idx, NULL) == 0) {
316 			if (idx > 0 && idx <= if_index) {
317 				ifnet_head_lock_shared();
318 				dstifp = ifindex2ifnet[idx];
319 				ifnet_head_done();
320 			}
321 		}
322 	}
323 #endif /* IN6_IFSTAT_STRICT */
324 
325 	/* we are violating the spec, this may not be the dst interface */
326 	if (dstifp == NULL) {
327 		dstifp = m->m_pkthdr.rcvif;
328 	}
329 
330 	/* jumbo payload can't contain a fragment header */
331 	if (ip6->ip6_plen == 0) {
332 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
333 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
334 		m = NULL;
335 		goto 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 		m = NULL;
350 		goto done;
351 	}
352 
353 	/* If ip6_maxfragpackets or ip6_maxfrags is 0, never accept fragments */
354 	if (ip6_maxfragpackets == 0 || ip6_maxfrags == 0) {
355 		ip6stat.ip6s_fragments++;
356 		ip6stat.ip6s_fragdropped++;
357 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
358 		m_drop(m, DROPTAP_FLAG_DIR_IN | DROPTAP_FLAG_L2_MISSING, DROP_REASON_IP_FRAG_NOT_ACCEPTED,
359 		    NULL, 0);
360 		m = NULL;
361 		goto done;
362 	}
363 
364 	/* offset now points to data portion */
365 	offset += sizeof(struct ip6_frag);
366 
367 	/*
368 	 * RFC 6946: Handle "atomic" fragments (offset and m bit set to 0)
369 	 * upfront, unrelated to any reassembly.  Just skip the fragment header.
370 	 */
371 	if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) {
372 		/*
373 		 * Mark packet as reassembled.
374 		 * In ICMPv6 processing, we drop certain
375 		 * NDP messages that are not expected to
376 		 * have fragment header based on recommendations
377 		 * against security vulnerability as described in
378 		 * RFC 6980.
379 		 * Treat atomic fragments as re-assembled packets as well.
380 		 */
381 		m->m_pkthdr.pkt_flags |= PKTF_REASSEMBLED;
382 		ip6stat.ip6s_atmfrag_rcvd++;
383 		in6_ifstat_inc(dstifp, ifs6_atmfrag_rcvd);
384 		*mp = m;
385 		*offp = offset;
386 		return ip6f->ip6f_nxt;
387 	}
388 
389 	/*
390 	 * Leverage partial checksum offload for simple UDP/IP fragments,
391 	 * as that is the most common case.
392 	 *
393 	 * Perform 1's complement adjustment of octets that got included/
394 	 * excluded in the hardware-calculated checksum value.  Also take
395 	 * care of any trailing bytes and subtract out their partial sum.
396 	 */
397 	if (ip6f->ip6f_nxt == IPPROTO_UDP &&
398 	    offset == (sizeof(*ip6) + sizeof(*ip6f)) &&
399 	    (m->m_pkthdr.csum_flags &
400 	    (CSUM_DATA_VALID | CSUM_PARTIAL | CSUM_PSEUDO_HDR)) ==
401 	    (CSUM_DATA_VALID | CSUM_PARTIAL)) {
402 		uint32_t start = m->m_pkthdr.csum_rx_start;
403 		uint32_t ip_len = (sizeof(*ip6) + ntohs(ip6->ip6_plen));
404 		int32_t trailer = (m_pktlen(m) - ip_len);
405 		uint32_t swbytes = (uint32_t)trailer;
406 
407 		csum = m->m_pkthdr.csum_rx_val;
408 
409 		ASSERT(trailer >= 0);
410 		if (start != offset || trailer != 0) {
411 			uint16_t s = 0, d = 0;
412 
413 			if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) {
414 				s = ip6->ip6_src.s6_addr16[1];
415 				ip6->ip6_src.s6_addr16[1] = 0;
416 			}
417 			if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) {
418 				d = ip6->ip6_dst.s6_addr16[1];
419 				ip6->ip6_dst.s6_addr16[1] = 0;
420 			}
421 
422 			/* callee folds in sum */
423 			csum = m_adj_sum16(m, start, offset,
424 			    (ip_len - offset), csum);
425 			if (offset > start) {
426 				swbytes += (offset - start);
427 			} else {
428 				swbytes += (start - offset);
429 			}
430 
431 			if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) {
432 				ip6->ip6_src.s6_addr16[1] = s;
433 			}
434 			if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) {
435 				ip6->ip6_dst.s6_addr16[1] = d;
436 			}
437 		}
438 		csum_flags = m->m_pkthdr.csum_flags;
439 
440 		if (swbytes != 0) {
441 			udp_in6_cksum_stats(swbytes);
442 		}
443 		if (trailer != 0) {
444 			m_adj(m, -trailer);
445 		}
446 	} else {
447 		csum = 0;
448 		csum_flags = 0;
449 	}
450 
451 	/* Invalidate checksum */
452 	m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
453 
454 	ip6stat.ip6s_fragments++;
455 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
456 
457 	lck_mtx_lock(&ip6qlock);
458 	locked = 1;
459 
460 	for (q6 = ip6q.ip6q_next; q6 != &ip6q; q6 = q6->ip6q_next) {
461 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
462 		    in6_are_addr_equal_scoped(&ip6->ip6_src, &q6->ip6q_src, ip6_input_getsrcifscope(m), q6->ip6q_src_ifscope) &&
463 		    in6_are_addr_equal_scoped(&ip6->ip6_dst, &q6->ip6q_dst, ip6_input_getdstifscope(m), q6->ip6q_dst_ifscope)) {
464 			break;
465 		}
466 	}
467 
468 	if (q6 == &ip6q) {
469 		/*
470 		 * Create a reassembly queue as this is the first fragment to
471 		 * arrive.
472 		 * By first frag, we don't mean the one with offset 0, but
473 		 * any of the fragments of the fragmented packet that has
474 		 * reached us first.
475 		 */
476 		first_frag = 1;
477 
478 		q6 = ip6q_alloc();
479 		if (q6 == NULL) {
480 			drop_reason = DROP_REASON_IP_FRAG_TOO_MANY;
481 			goto dropfrag;
482 		}
483 
484 		frag6_insque(q6, &ip6q);
485 		frag6_nfragpackets++;
486 
487 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
488 		q6->ip6q_down   = q6->ip6q_up = (struct ip6asfrag *)q6;
489 #ifdef notyet
490 		q6->ip6q_nxtp   = (u_char *)nxtp;
491 #endif
492 		q6->ip6q_ident  = ip6f->ip6f_ident;
493 		q6->ip6q_ttl    = IPV6_FRAGTTL;
494 		q6->ip6q_src    = ip6->ip6_src;
495 		q6->ip6q_dst    = ip6->ip6_dst;
496 		q6->ip6q_dst_ifscope = IN6_IS_SCOPE_EMBED(&q6->ip6q_dst) ? ip6_input_getdstifscope(m) : IFSCOPE_NONE;
497 		q6->ip6q_src_ifscope = IN6_IS_SCOPE_EMBED(&q6->ip6q_src) ? ip6_input_getsrcifscope(m) : IFSCOPE_NONE;
498 		q6->ip6q_ecn    =
499 		    (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
500 		q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */
501 
502 		q6->ip6q_nfrag = 0;
503 		q6->ip6q_flags = 0;
504 
505 		/*
506 		 * If the first fragment has valid checksum offload
507 		 * info, the rest of fragments are eligible as well.
508 		 */
509 		if (csum_flags != 0) {
510 			q6->ip6q_csum = csum;
511 			q6->ip6q_csum_flags = csum_flags;
512 		}
513 	}
514 
515 	if (q6->ip6q_flags & IP6QF_DIRTY) {
516 		drop_reason = DROP_REASON_IP6_FRAG_OVERLAPPING;
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 				mbuf_ref_t merr = IP6_REASS_MBUF(af6);
592 				struct ip6_hdr *__single 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();
620 	if (ip6af == NULL) {
621 		drop_reason = DROP_REASON_IP_FRAG_TOO_MANY;
622 		goto dropfrag;
623 	}
624 
625 	ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
626 	ip6af->ip6af_off = fragoff;
627 	ip6af->ip6af_frglen = frgpartlen;
628 	ip6af->ip6af_offset = offset;
629 	IP6_REASS_MBUF(ip6af) = m;
630 
631 	if (first_frag) {
632 		af6 = (struct ip6asfrag *)q6;
633 		goto insert;
634 	}
635 
636 	/*
637 	 * Handle ECN by comparing this segment with the first one;
638 	 * if CE is set, do not lose CE.
639 	 * drop if CE and not-ECT are mixed for the same packet.
640 	 */
641 	ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
642 	ecn0 = q6->ip6q_ecn;
643 	if (ecn == IPTOS_ECN_CE) {
644 		if (ecn0 == IPTOS_ECN_NOTECT) {
645 			ip6af_free(ip6af);
646 			drop_reason = DROP_REASON_IP6_FRAG_MIXED_CE;
647 			goto dropfrag;
648 		}
649 		if (ecn0 != IPTOS_ECN_CE) {
650 			q6->ip6q_ecn = IPTOS_ECN_CE;
651 		}
652 	}
653 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
654 		ip6af_free(ip6af);
655 		drop_reason = DROP_REASON_IP6_FRAG_MIXED_CE;
656 		goto dropfrag;
657 	}
658 
659 	/*
660 	 * Find a segment which begins after this one does.
661 	 */
662 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
663 	    af6 = af6->ip6af_down) {
664 		if (af6->ip6af_off > ip6af->ip6af_off) {
665 			break;
666 		}
667 	}
668 
669 	/*
670 	 * As per RFC 8200 reassembly rules, we MUST drop the entire
671 	 * chain of fragments for a packet to be assembled, if we receive
672 	 * any overlapping fragments.
673 	 * https://tools.ietf.org/html/rfc8200#page-20
674 	 *
675 	 * To avoid more conditional code, just reuse frag6_freef and defer
676 	 * its call to post fragment insertion in the queue.
677 	 */
678 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
679 		if (af6->ip6af_up->ip6af_off == ip6af->ip6af_off) {
680 			if (af6->ip6af_up->ip6af_frglen != ip6af->ip6af_frglen) {
681 				drop_fragq = TRUE;
682 			} else {
683 				/*
684 				 * XXX Ideally we should be comparing the entire
685 				 * packet here but for now just use off and fraglen
686 				 * to ignore a duplicate fragment.
687 				 */
688 				ip6af_free(ip6af);
689 				drop_reason = DROP_REASON_IP6_FRAG_OVERLAPPING;
690 				goto dropfrag;
691 			}
692 		} else {
693 			i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
694 			    - ip6af->ip6af_off;
695 			if (i > 0) {
696 				drop_fragq = TRUE;
697 			}
698 		}
699 	}
700 
701 	if (af6 != (struct ip6asfrag *)q6) {
702 		/*
703 		 * Given that we break when af6->ip6af_off > ip6af->ip6af_off,
704 		 * we shouldn't need a check for duplicate fragment here.
705 		 * For now just assert.
706 		 */
707 		VERIFY(af6->ip6af_off != ip6af->ip6af_off);
708 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
709 		if (i > 0) {
710 			drop_fragq = TRUE;
711 		}
712 	}
713 
714 	/*
715 	 * If this fragment contains similar checksum offload info
716 	 * as that of the existing ones, accumulate checksum.  Otherwise,
717 	 * invalidate checksum offload info for the entire datagram.
718 	 */
719 	if (csum_flags != 0 && csum_flags == q6->ip6q_csum_flags) {
720 		q6->ip6q_csum += csum;
721 	} else if (q6->ip6q_csum_flags != 0) {
722 		q6->ip6q_csum_flags = 0;
723 	}
724 
725 insert:
726 	/*
727 	 * Stick new segment in its place;
728 	 * check for complete reassembly.
729 	 * Move to front of packet queue, as we are
730 	 * the most recently active fragmented packet.
731 	 */
732 	frag6_enq(ip6af, af6->ip6af_up);
733 	frag6_nfrags++;
734 	q6->ip6q_nfrag++;
735 
736 	/*
737 	 * This holds true, when we receive overlapping fragments.
738 	 * We must silently drop all the fragments we have received
739 	 * so far.
740 	 * Also mark q6 as dirty, so as to not add any new fragments to it.
741 	 * Make sure even q6 marked dirty is kept till timer expires for
742 	 * reassembly and when that happens, silenty get rid of q6
743 	 */
744 	if (drop_fragq) {
745 		struct fq6_head dfq6 = {0};
746 		MBUFQ_INIT(&dfq6);      /* for deferred frees */
747 		q6->ip6q_flags |= IP6QF_DIRTY;
748 		/* Purge all the fragments but do not free q6 */
749 		frag6_purgef(q6, &dfq6, NULL);
750 		af6 = NULL;
751 
752 		/* free fragments that need to be freed */
753 		if (!MBUFQ_EMPTY(&dfq6)) {
754 			MBUFQ_DROP_AND_DRAIN(&dfq6, DROPTAP_FLAG_DIR_IN, DROP_REASON_IP6_FRAG_OVERLAPPING);
755 		}
756 		VERIFY(MBUFQ_EMPTY(&dfq6));
757 		/*
758 		 * Just in case the above logic got anything added
759 		 * to diq6, drain it.
760 		 * Please note that these mbufs are not present in the
761 		 * fragment queue and are added to diq6 for sending
762 		 * ICMPv6 error.
763 		 * Given that the current fragment was an overlapping
764 		 * fragment and the RFC requires us to not send any
765 		 * ICMPv6 errors while purging the entire queue.
766 		 * Just empty it out.
767 		 */
768 		if (!MBUFQ_EMPTY(&diq6)) {
769 			MBUFQ_DROP_AND_DRAIN(&diq6, DROPTAP_FLAG_DIR_IN, DROP_REASON_IP6_FRAG_OVERLAPPING);
770 		}
771 		VERIFY(MBUFQ_EMPTY(&diq6));
772 		/*
773 		 * MBUFQ_DRAIN would have drained all the mbufs
774 		 * in the fragment queue.
775 		 * This shouldn't be needed as we are returning IPPROTO_DONE
776 		 * from here but change the passed mbuf pointer to NULL.
777 		 */
778 		*mp = NULL;
779 		lck_mtx_unlock(&ip6qlock);
780 		return IPPROTO_DONE;
781 	}
782 
783 	/*
784 	 * We're keeping the fragment.
785 	 */
786 	q6->ip6q_unfrglen = local_ip6q_unfrglen;
787 	q6->ip6q_nxt = local_ip6q_nxt;
788 
789 	next = 0;
790 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
791 	    af6 = af6->ip6af_down) {
792 		if (af6->ip6af_off != next) {
793 			lck_mtx_unlock(&ip6qlock);
794 			locked = 0;
795 			m = NULL;
796 			goto done;
797 		}
798 		next += af6->ip6af_frglen;
799 	}
800 	if (af6->ip6af_up->ip6af_mff) {
801 		lck_mtx_unlock(&ip6qlock);
802 		locked = 0;
803 		m = NULL;
804 		goto done;
805 	}
806 
807 	/*
808 	 * Reassembly is complete; concatenate fragments.
809 	 */
810 	ip6af = q6->ip6q_down;
811 	t = m = IP6_REASS_MBUF(ip6af);
812 	af6 = ip6af->ip6af_down;
813 	frag6_deq(ip6af);
814 	while (af6 != (struct ip6asfrag *)q6) {
815 		af6dwn = af6->ip6af_down;
816 		frag6_deq(af6);
817 		while (t->m_next) {
818 			t = t->m_next;
819 		}
820 		t->m_next = IP6_REASS_MBUF(af6);
821 		m_adj(t->m_next, af6->ip6af_offset);
822 		ip6af_free(af6);
823 		af6 = af6dwn;
824 	}
825 
826 	/*
827 	 * Store partial hardware checksum info from the fragment queue;
828 	 * the receive start offset is set to 40 bytes (see code at the
829 	 * top of this routine.)
830 	 */
831 	if (q6->ip6q_csum_flags != 0) {
832 		csum = q6->ip6q_csum;
833 
834 		ADDCARRY(csum);
835 
836 		m->m_pkthdr.csum_rx_val = (u_int16_t)csum;
837 		m->m_pkthdr.csum_rx_start = sizeof(struct ip6_hdr);
838 		m->m_pkthdr.csum_flags = q6->ip6q_csum_flags;
839 	} else if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
840 	    (m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
841 		/* loopback checksums are always OK */
842 		m->m_pkthdr.csum_data = 0xffff;
843 		m->m_pkthdr.csum_flags = CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
844 	}
845 
846 	/* adjust offset to point where the original next header starts */
847 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
848 	ip6af_free(ip6af);
849 	ip6 = mtod(m, struct ip6_hdr *);
850 	ip6->ip6_plen = htons((uint16_t)(next + offset - sizeof(struct ip6_hdr)));
851 	ip6->ip6_src = q6->ip6q_src;
852 	ip6->ip6_dst = q6->ip6q_dst;
853 	ip6_output_setdstifscope(m, q6->ip6q_dst_ifscope, NULL);
854 	ip6_output_setsrcifscope(m, q6->ip6q_src_ifscope, NULL);
855 	if (q6->ip6q_ecn == IPTOS_ECN_CE) {
856 		ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20);
857 	}
858 
859 	nxt = q6->ip6q_nxt;
860 #ifdef notyet
861 	*q6->ip6q_nxtp = (u_char)(nxt & 0xff);
862 #endif
863 
864 	/* Delete frag6 header */
865 	if (m->m_len >= offset + sizeof(struct ip6_frag)) {
866 		/* This is the only possible case with !PULLDOWN_TEST */
867 		ovbcopy((caddr_t)ip6, (caddr_t)ip6 + sizeof(struct ip6_frag),
868 		    offset);
869 		m->m_data += sizeof(struct ip6_frag);
870 		m->m_len -= sizeof(struct ip6_frag);
871 	} else {
872 		/* this comes with no copy if the boundary is on cluster */
873 		if ((t = m_split(m, offset, M_DONTWAIT)) == NULL) {
874 			frag6_remque(q6);
875 			frag6_nfragpackets--;
876 			frag6_nfrags -= q6->ip6q_nfrag;
877 			ip6q_free(q6);
878 			goto dropfrag;
879 		}
880 		m_adj(t, sizeof(struct ip6_frag));
881 		m_cat(m, t);
882 	}
883 
884 	/*
885 	 * Store NXT to the original.
886 	 */
887 	{
888 		char *prvnxtp = ip6_get_prevhdr(m, offset); /* XXX */
889 		*prvnxtp = nxt;
890 	}
891 
892 	frag6_remque(q6);
893 	frag6_nfragpackets--;
894 	frag6_nfrags -= q6->ip6q_nfrag;
895 	ip6q_free(q6);
896 
897 	if (m->m_flags & M_PKTHDR) {    /* Isn't it always true? */
898 		m_fixhdr(m);
899 		/*
900 		 * Mark packet as reassembled
901 		 * In ICMPv6 processing, we drop certain
902 		 * NDP messages that are not expected to
903 		 * have fragment header based on recommendations
904 		 * against security vulnerability as described in
905 		 * RFC 6980.
906 		 */
907 		m->m_pkthdr.pkt_flags |= PKTF_REASSEMBLED;
908 	}
909 	ip6stat.ip6s_reassembled++;
910 
911 	/*
912 	 * Tell launch routine the next header
913 	 */
914 	*mp = m;
915 	*offp = offset;
916 
917 	/* arm the purge timer if not already and if there's work to do */
918 	frag6_sched_timeout();
919 	lck_mtx_unlock(&ip6qlock);
920 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
921 	frag6_icmp6_paramprob_error(&diq6);
922 	VERIFY(MBUFQ_EMPTY(&diq6));
923 	return nxt;
924 
925 done:
926 	VERIFY(m == NULL);
927 	*mp = m;
928 	if (!locked) {
929 		if (frag6_nfragpackets == 0) {
930 			frag6_icmp6_paramprob_error(&diq6);
931 			VERIFY(MBUFQ_EMPTY(&diq6));
932 			return IPPROTO_DONE;
933 		}
934 		lck_mtx_lock(&ip6qlock);
935 	}
936 	/* arm the purge timer if not already and if there's work to do */
937 	frag6_sched_timeout();
938 	lck_mtx_unlock(&ip6qlock);
939 	frag6_icmp6_paramprob_error(&diq6);
940 	VERIFY(MBUFQ_EMPTY(&diq6));
941 	return IPPROTO_DONE;
942 
943 dropfrag:
944 	ip6stat.ip6s_fragdropped++;
945 	/* arm the purge timer if not already and if there's work to do */
946 	frag6_sched_timeout();
947 	lck_mtx_unlock(&ip6qlock);
948 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
949 	m_drop(m, DROPTAP_FLAG_DIR_IN | DROPTAP_FLAG_L2_MISSING, drop_reason, NULL, 0);
950 	*mp = NULL;
951 	frag6_icmp6_paramprob_error(&diq6);
952 	VERIFY(MBUFQ_EMPTY(&diq6));
953 	return IPPROTO_DONE;
954 }
955 
956 /*
957  * This routine removes the enqueued frames from the passed fragment
958  * header and enqueues those to dfq6 which is an out-arg for the dequeued
959  * fragments.
960  * If the caller also provides diq6, this routine also enqueues the 0 offset
961  * fragment to that list as it potentially gets used by the caller
962  * to prepare the relevant ICMPv6 error message (time exceeded or
963  * param problem).
964  * It leaves the fragment header object (q6) intact.
965  */
966 static void
frag6_purgef(struct ip6q * q6,struct fq6_head * dfq6,struct fq6_head * diq6)967 frag6_purgef(struct ip6q *q6, struct fq6_head *dfq6, struct fq6_head *diq6)
968 {
969 	struct ip6asfrag *__single af6 = NULL;
970 	struct ip6asfrag *__single down6 = NULL;
971 
972 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_OWNED);
973 
974 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
975 	    af6 = down6) {
976 		mbuf_ref_t m = IP6_REASS_MBUF(af6);
977 
978 		down6 = af6->ip6af_down;
979 		frag6_deq(af6);
980 
981 		/*
982 		 * If caller wants to generate ICMP time-exceeded,
983 		 * as indicated by the argument diq6, return it for
984 		 * the first fragment and add others to the fragment
985 		 * free queue.
986 		 */
987 		if (af6->ip6af_off == 0 && diq6 != NULL) {
988 			struct ip6_hdr *__single ip6;
989 
990 			/* adjust pointer */
991 			ip6 = mtod(m, struct ip6_hdr *);
992 
993 			/* restore source and destination addresses */
994 			ip6->ip6_src = q6->ip6q_src;
995 			ip6->ip6_dst = q6->ip6q_dst;
996 			ip6_output_setdstifscope(m, q6->ip6q_dst_ifscope, NULL);
997 			ip6_output_setsrcifscope(m, q6->ip6q_src_ifscope, NULL);
998 			MBUFQ_ENQUEUE(diq6, m);
999 		} else {
1000 			MBUFQ_ENQUEUE(dfq6, m);
1001 		}
1002 		ip6af_free(af6);
1003 	}
1004 }
1005 
1006 /*
1007  * This routine removes the enqueued frames from the passed fragment
1008  * header and enqueues those to dfq6 which is an out-arg for the dequeued
1009  * fragments.
1010  * If the caller also provides diq6, this routine also enqueues the 0 offset
1011  * fragment to that list as it potentially gets used by the caller
1012  * to prepare the relevant ICMPv6 error message (time exceeded or
1013  * param problem).
1014  * It also remove the fragment header object from the queue and frees it.
1015  */
1016 static void
frag6_freef(struct ip6q * q6,struct fq6_head * dfq6,struct fq6_head * diq6)1017 frag6_freef(struct ip6q *q6, struct fq6_head *dfq6, struct fq6_head *diq6)
1018 {
1019 	frag6_purgef(q6, dfq6, diq6);
1020 	frag6_remque(q6);
1021 	frag6_nfragpackets--;
1022 	frag6_nfrags -= q6->ip6q_nfrag;
1023 	ip6q_free(q6);
1024 }
1025 
1026 /*
1027  * Put an ip fragment on a reassembly chain.
1028  * Like insque, but pointers in middle of structure.
1029  */
1030 void
frag6_enq(struct ip6asfrag * af6,struct ip6asfrag * up6)1031 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6)
1032 {
1033 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_OWNED);
1034 
1035 	af6->ip6af_up = up6;
1036 	af6->ip6af_down = up6->ip6af_down;
1037 	up6->ip6af_down->ip6af_up = af6;
1038 	up6->ip6af_down = af6;
1039 }
1040 
1041 /*
1042  * To frag6_enq as remque is to insque.
1043  */
1044 void
frag6_deq(struct ip6asfrag * af6)1045 frag6_deq(struct ip6asfrag *af6)
1046 {
1047 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_OWNED);
1048 
1049 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
1050 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
1051 }
1052 
1053 void
frag6_insque(struct ip6q * new,struct ip6q * old)1054 frag6_insque(struct ip6q *new, struct ip6q *old)
1055 {
1056 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_OWNED);
1057 
1058 	new->ip6q_prev = old;
1059 	new->ip6q_next = old->ip6q_next;
1060 	old->ip6q_next->ip6q_prev = new;
1061 	old->ip6q_next = new;
1062 }
1063 
1064 void
frag6_remque(struct ip6q * p6)1065 frag6_remque(struct ip6q *p6)
1066 {
1067 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_OWNED);
1068 
1069 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
1070 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
1071 }
1072 
1073 /*
1074  * IPv6 reassembling timer processing;
1075  * if a timer expires on a reassembly
1076  * queue, discard it.
1077  */
1078 static void
frag6_timeout(void * arg)1079 frag6_timeout(void *arg)
1080 {
1081 #pragma unused(arg)
1082 	struct fq6_head dfq6, diq6;
1083 	struct fq6_head *__single diq6_tmp = NULL;
1084 	struct ip6q *__single q6;
1085 
1086 	MBUFQ_INIT(&dfq6);      /* for deferred frees */
1087 	MBUFQ_INIT(&diq6);      /* for deferred ICMP time exceeded errors */
1088 
1089 	/*
1090 	 * Update coarse-grained networking timestamp (in sec.); the idea
1091 	 * is to piggy-back on the timeout callout to update the counter
1092 	 * returnable via net_uptime().
1093 	 */
1094 	net_update_uptime();
1095 
1096 	lck_mtx_lock(&ip6qlock);
1097 	q6 = ip6q.ip6q_next;
1098 	if (q6) {
1099 		while (q6 != &ip6q) {
1100 			--q6->ip6q_ttl;
1101 			q6 = q6->ip6q_next;
1102 			if (q6->ip6q_prev->ip6q_ttl == 0) {
1103 				ip6stat.ip6s_fragtimeout++;
1104 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
1105 				/*
1106 				 * Avoid sending ICMPv6 Time Exceeded for fragment headers
1107 				 * that are marked dirty.
1108 				 */
1109 				diq6_tmp = (q6->ip6q_prev->ip6q_flags & IP6QF_DIRTY) ?
1110 				    NULL : &diq6;
1111 				frag6_freef(q6->ip6q_prev, &dfq6, diq6_tmp);
1112 			}
1113 		}
1114 	}
1115 	/*
1116 	 * If we are over the maximum number of fragments
1117 	 * (due to the limit being lowered), drain off
1118 	 * enough to get down to the new limit.
1119 	 */
1120 	if (ip6_maxfragpackets >= 0) {
1121 		while (frag6_nfragpackets > (unsigned)ip6_maxfragpackets &&
1122 		    ip6q.ip6q_prev) {
1123 			ip6stat.ip6s_fragoverflow++;
1124 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
1125 			/*
1126 			 * Avoid sending ICMPv6 Time Exceeded for fragment headers
1127 			 * that are marked dirty.
1128 			 */
1129 			diq6_tmp = (ip6q.ip6q_prev->ip6q_flags & IP6QF_DIRTY) ?
1130 			    NULL : &diq6;
1131 			frag6_freef(ip6q.ip6q_prev, &dfq6, diq6_tmp);
1132 		}
1133 	}
1134 	/* re-arm the purge timer if there's work to do */
1135 	frag6_timeout_run = 0;
1136 	frag6_sched_timeout();
1137 	lck_mtx_unlock(&ip6qlock);
1138 
1139 	/* free fragments that need to be freed */
1140 	if (!MBUFQ_EMPTY(&dfq6)) {
1141 		MBUFQ_DROP_AND_DRAIN(&dfq6, DROPTAP_FLAG_DIR_IN, DROP_REASON_IP_FRAG_TIMEOUT);
1142 	}
1143 
1144 	frag6_icmp6_timeex_error(&diq6);
1145 
1146 	VERIFY(MBUFQ_EMPTY(&dfq6));
1147 	VERIFY(MBUFQ_EMPTY(&diq6));
1148 }
1149 
1150 static void
frag6_sched_timeout(void)1151 frag6_sched_timeout(void)
1152 {
1153 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_OWNED);
1154 
1155 	if (!frag6_timeout_run && frag6_nfragpackets > 0) {
1156 		frag6_timeout_run = 1;
1157 		timeout(frag6_timeout, NULL, hz);
1158 	}
1159 }
1160 
1161 /*
1162  * Drain off all datagram fragments.
1163  */
1164 void
frag6_drain(void)1165 frag6_drain(void)
1166 {
1167 	struct fq6_head dfq6, diq6;
1168 	struct fq6_head *__single diq6_tmp = NULL;
1169 
1170 	MBUFQ_INIT(&dfq6);      /* for deferred frees */
1171 	MBUFQ_INIT(&diq6);      /* for deferred ICMP time exceeded errors */
1172 
1173 	lck_mtx_lock(&ip6qlock);
1174 	while (ip6q.ip6q_next != &ip6q) {
1175 		ip6stat.ip6s_fragdropped++;
1176 		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
1177 		/*
1178 		 * Avoid sending ICMPv6 Time Exceeded for fragment headers
1179 		 * that are marked dirty.
1180 		 */
1181 		diq6_tmp = (ip6q.ip6q_next->ip6q_flags & IP6QF_DIRTY) ?
1182 		    NULL : &diq6;
1183 		frag6_freef(ip6q.ip6q_next, &dfq6, diq6_tmp);
1184 	}
1185 	lck_mtx_unlock(&ip6qlock);
1186 
1187 	/* free fragments that need to be freed */
1188 	if (!MBUFQ_EMPTY(&dfq6)) {
1189 		MBUFQ_DROP_AND_DRAIN(&dfq6, DROPTAP_FLAG_DIR_IN, DROP_REASON_IP_FRAG_DRAINED);
1190 	}
1191 
1192 	frag6_icmp6_timeex_error(&diq6);
1193 
1194 	VERIFY(MBUFQ_EMPTY(&dfq6));
1195 	VERIFY(MBUFQ_EMPTY(&diq6));
1196 }
1197 
1198 static struct ip6q *
ip6q_alloc(void)1199 ip6q_alloc(void)
1200 {
1201 	struct ip6q *__single q6;
1202 
1203 	/*
1204 	 * See comments in ip6q_updateparams().  Keep the count separate
1205 	 * from frag6_nfragpackets since the latter represents the elements
1206 	 * already in the reassembly queues.
1207 	 */
1208 	if (ip6q_limit > 0 && ip6q_count > ip6q_limit) {
1209 		return NULL;
1210 	}
1211 
1212 	q6 = kalloc_type(struct ip6q, Z_NOWAIT | Z_ZERO);
1213 	if (q6 != NULL) {
1214 		os_atomic_inc(&ip6q_count, relaxed);
1215 	}
1216 	return q6;
1217 }
1218 
1219 static void
ip6q_free(struct ip6q * q6)1220 ip6q_free(struct ip6q *q6)
1221 {
1222 	kfree_type(struct ip6q, q6);
1223 	os_atomic_dec(&ip6q_count, relaxed);
1224 }
1225 
1226 static struct ip6asfrag *
ip6af_alloc(void)1227 ip6af_alloc(void)
1228 {
1229 	struct ip6asfrag *__single af6;
1230 
1231 	/*
1232 	 * See comments in ip6q_updateparams().  Keep the count separate
1233 	 * from frag6_nfrags since the latter represents the elements
1234 	 * already in the reassembly queues.
1235 	 */
1236 	if (ip6af_limit > 0 && ip6af_count > ip6af_limit) {
1237 		return NULL;
1238 	}
1239 
1240 	af6 = kalloc_type(struct ip6asfrag, Z_NOWAIT | Z_ZERO);
1241 	if (af6 != NULL) {
1242 		os_atomic_inc(&ip6af_count, relaxed);
1243 	}
1244 	return af6;
1245 }
1246 
1247 static void
ip6af_free(struct ip6asfrag * af6)1248 ip6af_free(struct ip6asfrag *af6)
1249 {
1250 	kfree_type(struct ip6asfrag, af6);
1251 	os_atomic_dec(&ip6af_count, relaxed);
1252 }
1253 
1254 static void
ip6q_updateparams(void)1255 ip6q_updateparams(void)
1256 {
1257 	LCK_MTX_ASSERT(&ip6qlock, LCK_MTX_ASSERT_OWNED);
1258 	/*
1259 	 * -1 for unlimited allocation.
1260 	 */
1261 	if (ip6_maxfragpackets < 0) {
1262 		ip6q_limit = 0;
1263 	}
1264 	if (ip6_maxfrags < 0) {
1265 		ip6af_limit = 0;
1266 	}
1267 	/*
1268 	 * Positive number for specific bound.
1269 	 */
1270 	if (ip6_maxfragpackets > 0) {
1271 		ip6q_limit = ip6_maxfragpackets;
1272 	}
1273 	if (ip6_maxfrags > 0) {
1274 		ip6af_limit = ip6_maxfrags;
1275 	}
1276 	/*
1277 	 * Zero specifies no further fragment queue allocation -- set the
1278 	 * bound very low, but rely on implementation elsewhere to actually
1279 	 * prevent allocation and reclaim current queues.
1280 	 */
1281 	if (ip6_maxfragpackets == 0) {
1282 		ip6q_limit = 1;
1283 	}
1284 	if (ip6_maxfrags == 0) {
1285 		ip6af_limit = 1;
1286 	}
1287 	/*
1288 	 * Arm the purge timer if not already and if there's work to do
1289 	 */
1290 	frag6_sched_timeout();
1291 }
1292 
1293 static int
1294 sysctl_maxfragpackets SYSCTL_HANDLER_ARGS
1295 {
1296 #pragma unused(arg1, arg2)
1297 	int error, i;
1298 
1299 	lck_mtx_lock(&ip6qlock);
1300 	i = ip6_maxfragpackets;
1301 	error = sysctl_handle_int(oidp, &i, 0, req);
1302 	if (error || req->newptr == USER_ADDR_NULL) {
1303 		goto done;
1304 	}
1305 	/* impose bounds */
1306 	if (i < -1 || i > (nmbclusters / 4)) {
1307 		error = EINVAL;
1308 		goto done;
1309 	}
1310 	ip6_maxfragpackets = i;
1311 	ip6q_updateparams();
1312 done:
1313 	lck_mtx_unlock(&ip6qlock);
1314 	return error;
1315 }
1316 
1317 static int
1318 sysctl_maxfrags SYSCTL_HANDLER_ARGS
1319 {
1320 #pragma unused(arg1, arg2)
1321 	int error, i;
1322 
1323 	lck_mtx_lock(&ip6qlock);
1324 	i = ip6_maxfrags;
1325 	error = sysctl_handle_int(oidp, &i, 0, req);
1326 	if (error || req->newptr == USER_ADDR_NULL) {
1327 		goto done;
1328 	}
1329 	/* impose bounds */
1330 	if (i < -1 || i > (nmbclusters / 4)) {
1331 		error = EINVAL;
1332 		goto done;
1333 	}
1334 	ip6_maxfrags = i;
1335 	ip6q_updateparams();    /* see if we need to arm timer */
1336 done:
1337 	lck_mtx_unlock(&ip6qlock);
1338 	return error;
1339 }
1340