xref: /netbsd-src/sys/netinet/ip_reass.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: ip_reass.c,v 1.18 2018/07/10 15:46:58 maxv Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1988, 1993
5  *	The Regents of the University of California.  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 University 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 REGENTS 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 REGENTS 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  *	@(#)ip_input.c	8.2 (Berkeley) 1/4/94
32  */
33 
34 /*
35  * IP reassembly.
36  *
37  * Additive-Increase/Multiplicative-Decrease (AIMD) strategy for IP
38  * reassembly queue buffer managment.
39  *
40  * We keep a count of total IP fragments (NB: not fragmented packets),
41  * awaiting reassembly (ip_nfrags) and a limit (ip_maxfrags) on fragments.
42  * If ip_nfrags exceeds ip_maxfrags the limit, we drop half the total
43  * fragments in reassembly queues.  This AIMD policy avoids repeatedly
44  * deleting single packets under heavy fragmentation load (e.g., from lossy
45  * NFS peers).
46  */
47 
48 #include <sys/cdefs.h>
49 __KERNEL_RCSID(0, "$NetBSD: ip_reass.c,v 1.18 2018/07/10 15:46:58 maxv Exp $");
50 
51 #include <sys/param.h>
52 #include <sys/types.h>
53 
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/mutex.h>
57 #include <sys/pool.h>
58 #include <sys/queue.h>
59 #include <sys/sysctl.h>
60 #include <sys/systm.h>
61 
62 #include <net/if.h>
63 
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/ip.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/ip_var.h>
69 #include <netinet/ip_private.h>
70 #include <netinet/in_var.h>
71 
72 /*
73  * IP reassembly queue structures.  Each fragment being reassembled is
74  * attached to one of these structures.  They are timed out after TTL
75  * drops to 0, and may also be reclaimed if memory becomes tight.
76  */
77 
78 typedef struct ipfr_qent {
79 	TAILQ_ENTRY(ipfr_qent)	ipqe_q;
80 	struct ip *		ipqe_ip;
81 	struct mbuf *		ipqe_m;
82 	bool			ipqe_mff;
83 } ipfr_qent_t;
84 
85 TAILQ_HEAD(ipfr_qent_head, ipfr_qent);
86 
87 typedef struct ipfr_queue {
88 	LIST_ENTRY(ipfr_queue)	ipq_q;		/* to other reass headers */
89 	struct ipfr_qent_head	ipq_fragq;	/* queue of fragment entries */
90 	uint8_t			ipq_ttl;	/* time for reass q to live */
91 	uint8_t			ipq_p;		/* protocol of this fragment */
92 	uint16_t		ipq_id;		/* sequence id for reassembly */
93 	struct in_addr		ipq_src;
94 	struct in_addr		ipq_dst;
95 	uint16_t		ipq_nfrags;	/* frags in this queue entry */
96 	uint8_t			ipq_tos;	/* TOS of this fragment */
97 	int			ipq_ipsec;	/* IPsec flags */
98 } ipfr_queue_t;
99 
100 /*
101  * Hash table of IP reassembly queues.
102  */
103 #define	IPREASS_HASH_SHIFT	6
104 #define	IPREASS_HASH_SIZE	(1 << IPREASS_HASH_SHIFT)
105 #define	IPREASS_HASH_MASK	(IPREASS_HASH_SIZE - 1)
106 #define	IPREASS_HASH(x, y) \
107 	(((((x) & 0xf) | ((((x) >> 8) & 0xf) << 4)) ^ (y)) & IPREASS_HASH_MASK)
108 
109 static LIST_HEAD(, ipfr_queue)	ip_frags[IPREASS_HASH_SIZE];
110 static pool_cache_t	ipfren_cache;
111 static kmutex_t		ipfr_lock;
112 
113 /* Number of packets in reassembly queue and total number of fragments. */
114 static int		ip_nfragpackets;
115 static int		ip_nfrags;
116 
117 /* Limits on packet and fragments. */
118 static int		ip_maxfragpackets;
119 static int		ip_maxfrags;
120 
121 /*
122  * Cached copy of nmbclusters.  If nbclusters is different, recalculate
123  * IP parameters derived from nmbclusters.
124  */
125 static int		ip_nmbclusters;
126 
127 /*
128  * IP reassembly TTL machinery for multiplicative drop.
129  */
130 static u_int		fragttl_histo[IPFRAGTTL + 1];
131 
132 static struct sysctllog *ip_reass_sysctllog;
133 
134 void			sysctl_ip_reass_setup(void);
135 static void		ip_nmbclusters_changed(void);
136 
137 static struct mbuf *	ip_reass(ipfr_qent_t *, ipfr_queue_t *, u_int);
138 static u_int		ip_reass_ttl_decr(u_int ticks);
139 static void		ip_reass_drophalf(void);
140 static void		ip_freef(ipfr_queue_t *);
141 
142 /*
143  * ip_reass_init:
144  *
145  *	Initialization of IP reassembly mechanism.
146  */
147 void
148 ip_reass_init(void)
149 {
150 	int i;
151 
152 	ipfren_cache = pool_cache_init(sizeof(ipfr_qent_t), coherency_unit,
153 	    0, 0, "ipfrenpl", NULL, IPL_NET, NULL, NULL, NULL);
154 	mutex_init(&ipfr_lock, MUTEX_DEFAULT, IPL_VM);
155 
156 	for (i = 0; i < IPREASS_HASH_SIZE; i++) {
157 		LIST_INIT(&ip_frags[i]);
158 	}
159 	ip_maxfragpackets = 200;
160 	ip_maxfrags = 0;
161 	ip_nmbclusters_changed();
162 
163 	sysctl_ip_reass_setup();
164 }
165 
166 void
167 sysctl_ip_reass_setup(void)
168 {
169 
170 	sysctl_createv(&ip_reass_sysctllog, 0, NULL, NULL,
171 		CTLFLAG_PERMANENT,
172 		CTLTYPE_NODE, "inet",
173 		SYSCTL_DESCR("PF_INET related settings"),
174 		NULL, 0, NULL, 0,
175 		CTL_NET, PF_INET, CTL_EOL);
176 	sysctl_createv(&ip_reass_sysctllog, 0, NULL, NULL,
177 		CTLFLAG_PERMANENT,
178 		CTLTYPE_NODE, "ip",
179 		SYSCTL_DESCR("IPv4 related settings"),
180 		NULL, 0, NULL, 0,
181 		CTL_NET, PF_INET, IPPROTO_IP, CTL_EOL);
182 
183 	sysctl_createv(&ip_reass_sysctllog, 0, NULL, NULL,
184 		CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
185 		CTLTYPE_INT, "maxfragpackets",
186 		SYSCTL_DESCR("Maximum number of fragments to retain for "
187 			     "possible reassembly"),
188 		NULL, 0, &ip_maxfragpackets, 0,
189 		CTL_NET, PF_INET, IPPROTO_IP, IPCTL_MAXFRAGPACKETS, CTL_EOL);
190 }
191 
192 #define CHECK_NMBCLUSTER_PARAMS()				\
193 do {								\
194 	if (__predict_false(ip_nmbclusters != nmbclusters))	\
195 		ip_nmbclusters_changed();			\
196 } while (/*CONSTCOND*/0)
197 
198 /*
199  * Compute IP limits derived from the value of nmbclusters.
200  */
201 static void
202 ip_nmbclusters_changed(void)
203 {
204 	ip_maxfrags = nmbclusters / 4;
205 	ip_nmbclusters = nmbclusters;
206 }
207 
208 /*
209  * ip_reass:
210  *
211  *	Take incoming datagram fragment and try to reassemble it into whole
212  *	datagram.  If a chain for reassembly of this datagram already exists,
213  *	then it is given as 'fp'; otherwise have to make a chain.
214  */
215 static struct mbuf *
216 ip_reass(ipfr_qent_t *ipqe, ipfr_queue_t *fp, const u_int hash)
217 {
218 	struct ip *ip = ipqe->ipqe_ip, *qip;
219 	const int hlen = ip->ip_hl << 2;
220 	struct mbuf *m = ipqe->ipqe_m, *t;
221 	int ipsecflags = m->m_flags & (M_DECRYPTED|M_AUTHIPHDR);
222 	ipfr_qent_t *nq, *p, *q;
223 	int i, next;
224 
225 	KASSERT(mutex_owned(&ipfr_lock));
226 
227 	/*
228 	 * Presence of header sizes in mbufs would confuse code below.
229 	 */
230 	m->m_data += hlen;
231 	m->m_len -= hlen;
232 
233 #ifdef	notyet
234 	/* Make sure fragment limit is up-to-date. */
235 	CHECK_NMBCLUSTER_PARAMS();
236 
237 	/* If we have too many fragments, drop the older half. */
238 	if (ip_nfrags >= ip_maxfrags) {
239 		ip_reass_drophalf(void);
240 	}
241 #endif
242 
243 	/*
244 	 * We are about to add a fragment; increment frag count.
245 	 */
246 	ip_nfrags++;
247 
248 	/*
249 	 * If first fragment to arrive, create a reassembly queue.
250 	 */
251 	if (fp == NULL) {
252 		/*
253 		 * Enforce upper bound on number of fragmented packets
254 		 * for which we attempt reassembly:  a) if maxfrag is 0,
255 		 * never accept fragments  b) if maxfrag is -1, accept
256 		 * all fragments without limitation.
257 		 */
258 		if (ip_maxfragpackets < 0)
259 			;
260 		else if (ip_nfragpackets >= ip_maxfragpackets) {
261 			goto dropfrag;
262 		}
263 		fp = malloc(sizeof(ipfr_queue_t), M_FTABLE, M_NOWAIT);
264 		if (fp == NULL) {
265 			goto dropfrag;
266 		}
267 		ip_nfragpackets++;
268 		TAILQ_INIT(&fp->ipq_fragq);
269 		fp->ipq_nfrags = 1;
270 		fp->ipq_ttl = IPFRAGTTL;
271 		fp->ipq_p = ip->ip_p;
272 		fp->ipq_id = ip->ip_id;
273 		fp->ipq_tos = ip->ip_tos;
274 		fp->ipq_ipsec = ipsecflags;
275 		fp->ipq_src = ip->ip_src;
276 		fp->ipq_dst = ip->ip_dst;
277 		LIST_INSERT_HEAD(&ip_frags[hash], fp, ipq_q);
278 		p = NULL;
279 		goto insert;
280 	} else {
281 		fp->ipq_nfrags++;
282 	}
283 
284 	/*
285 	 * Find a segment which begins after this one does.
286 	 */
287 	TAILQ_FOREACH(q, &fp->ipq_fragq, ipqe_q) {
288 		if (ntohs(q->ipqe_ip->ip_off) > ntohs(ip->ip_off))
289 			break;
290 	}
291 	if (q != NULL) {
292 		p = TAILQ_PREV(q, ipfr_qent_head, ipqe_q);
293 	} else {
294 		p = TAILQ_LAST(&fp->ipq_fragq, ipfr_qent_head);
295 	}
296 
297 	/*
298 	 * If there is a preceding segment, it may provide some of our
299 	 * data already.  If so, drop the data from the incoming segment.
300 	 * If it provides all of our data, drop us.
301 	 */
302 	if (p != NULL) {
303 		i = ntohs(p->ipqe_ip->ip_off) + ntohs(p->ipqe_ip->ip_len) -
304 		    ntohs(ip->ip_off);
305 		if (i > 0) {
306 			if (i >= ntohs(ip->ip_len)) {
307 				goto dropfrag;
308 			}
309 			m_adj(ipqe->ipqe_m, i);
310 			ip->ip_off = htons(ntohs(ip->ip_off) + i);
311 			ip->ip_len = htons(ntohs(ip->ip_len) - i);
312 		}
313 	}
314 
315 	/*
316 	 * While we overlap succeeding segments trim them or, if they are
317 	 * completely covered, dequeue them.
318 	 */
319 	while (q != NULL) {
320 		size_t end;
321 
322 		qip = q->ipqe_ip;
323 		end = ntohs(ip->ip_off) + ntohs(ip->ip_len);
324 		if (end <= ntohs(qip->ip_off)) {
325 			break;
326 		}
327 		i = end - ntohs(qip->ip_off);
328 		if (i < ntohs(qip->ip_len)) {
329 			qip->ip_len = htons(ntohs(qip->ip_len) - i);
330 			qip->ip_off = htons(ntohs(qip->ip_off) + i);
331 			m_adj(q->ipqe_m, i);
332 			break;
333 		}
334 		nq = TAILQ_NEXT(q, ipqe_q);
335 		m_freem(q->ipqe_m);
336 		TAILQ_REMOVE(&fp->ipq_fragq, q, ipqe_q);
337 		pool_cache_put(ipfren_cache, q);
338 		fp->ipq_nfrags--;
339 		ip_nfrags--;
340 		q = nq;
341 	}
342 
343 insert:
344 	/*
345 	 * Stick new segment in its place; check for complete reassembly.
346 	 */
347 	if (p == NULL) {
348 		TAILQ_INSERT_HEAD(&fp->ipq_fragq, ipqe, ipqe_q);
349 	} else {
350 		TAILQ_INSERT_AFTER(&fp->ipq_fragq, p, ipqe, ipqe_q);
351 	}
352 	next = 0;
353 	TAILQ_FOREACH(q, &fp->ipq_fragq, ipqe_q) {
354 		qip = q->ipqe_ip;
355 		if (ntohs(qip->ip_off) != next) {
356 			mutex_exit(&ipfr_lock);
357 			return NULL;
358 		}
359 		next += ntohs(qip->ip_len);
360 	}
361 	p = TAILQ_LAST(&fp->ipq_fragq, ipfr_qent_head);
362 	if (p->ipqe_mff) {
363 		mutex_exit(&ipfr_lock);
364 		return NULL;
365 	}
366 
367 	/*
368 	 * Reassembly is complete.  Check for a bogus message size.
369 	 */
370 	q = TAILQ_FIRST(&fp->ipq_fragq);
371 	ip = q->ipqe_ip;
372 	if ((next + (ip->ip_hl << 2)) > IP_MAXPACKET) {
373 		IP_STATINC(IP_STAT_TOOLONG);
374 		ip_freef(fp);
375 		mutex_exit(&ipfr_lock);
376 		return NULL;
377 	}
378 	LIST_REMOVE(fp, ipq_q);
379 	ip_nfrags -= fp->ipq_nfrags;
380 	ip_nfragpackets--;
381 	mutex_exit(&ipfr_lock);
382 
383 	/* Concatenate all fragments. */
384 	m = q->ipqe_m;
385 	t = m->m_next;
386 	m->m_next = NULL;
387 	m_cat(m, t);
388 	nq = TAILQ_NEXT(q, ipqe_q);
389 	pool_cache_put(ipfren_cache, q);
390 
391 	for (q = nq; q != NULL; q = nq) {
392 		t = q->ipqe_m;
393 		nq = TAILQ_NEXT(q, ipqe_q);
394 		pool_cache_put(ipfren_cache, q);
395 		m_remove_pkthdr(t);
396 		m_cat(m, t);
397 	}
398 
399 	/*
400 	 * Create header for new packet by modifying header of first
401 	 * packet.  Dequeue and discard fragment reassembly header.  Make
402 	 * header visible.
403 	 */
404 	ip->ip_len = htons((ip->ip_hl << 2) + next);
405 	ip->ip_src = fp->ipq_src;
406 	ip->ip_dst = fp->ipq_dst;
407 	free(fp, M_FTABLE);
408 
409 	m->m_len += (ip->ip_hl << 2);
410 	m->m_data -= (ip->ip_hl << 2);
411 
412 	/* Fix up mbuf.  XXX This should be done elsewhere. */
413 	{
414 		KASSERT(m->m_flags & M_PKTHDR);
415 		int plen = 0;
416 		for (t = m; t; t = t->m_next) {
417 			plen += t->m_len;
418 		}
419 		m->m_pkthdr.len = plen;
420 		m->m_pkthdr.csum_flags = 0;
421 	}
422 	return m;
423 
424 dropfrag:
425 	if (fp != NULL) {
426 		fp->ipq_nfrags--;
427 	}
428 	ip_nfrags--;
429 	IP_STATINC(IP_STAT_FRAGDROPPED);
430 	mutex_exit(&ipfr_lock);
431 
432 	pool_cache_put(ipfren_cache, ipqe);
433 	m_freem(m);
434 	return NULL;
435 }
436 
437 /*
438  * ip_freef:
439  *
440  *	Free a fragment reassembly header and all associated datagrams.
441  */
442 static void
443 ip_freef(ipfr_queue_t *fp)
444 {
445 	ipfr_qent_t *q;
446 
447 	KASSERT(mutex_owned(&ipfr_lock));
448 
449 	LIST_REMOVE(fp, ipq_q);
450 	ip_nfrags -= fp->ipq_nfrags;
451 	ip_nfragpackets--;
452 
453 	while ((q = TAILQ_FIRST(&fp->ipq_fragq)) != NULL) {
454 		TAILQ_REMOVE(&fp->ipq_fragq, q, ipqe_q);
455 		m_freem(q->ipqe_m);
456 		pool_cache_put(ipfren_cache, q);
457 	}
458 	free(fp, M_FTABLE);
459 }
460 
461 /*
462  * ip_reass_ttl_decr:
463  *
464  *	Decrement TTL of all reasembly queue entries by `ticks'.  Count
465  *	number of distinct fragments (as opposed to partial, fragmented
466  *	datagrams) inthe reassembly queue.  While we  traverse the entire
467  *	reassembly queue, compute and return the median TTL over all
468  *	fragments.
469  */
470 static u_int
471 ip_reass_ttl_decr(u_int ticks)
472 {
473 	u_int nfrags, median, dropfraction, keepfraction;
474 	ipfr_queue_t *fp, *nfp;
475 	int i;
476 
477 	nfrags = 0;
478 	memset(fragttl_histo, 0, sizeof(fragttl_histo));
479 
480 	for (i = 0; i < IPREASS_HASH_SIZE; i++) {
481 		for (fp = LIST_FIRST(&ip_frags[i]); fp != NULL; fp = nfp) {
482 			fp->ipq_ttl = ((fp->ipq_ttl <= ticks) ?
483 			    0 : fp->ipq_ttl - ticks);
484 			nfp = LIST_NEXT(fp, ipq_q);
485 			if (fp->ipq_ttl == 0) {
486 				IP_STATINC(IP_STAT_FRAGTIMEOUT);
487 				ip_freef(fp);
488 			} else {
489 				nfrags += fp->ipq_nfrags;
490 				fragttl_histo[fp->ipq_ttl] += fp->ipq_nfrags;
491 			}
492 		}
493 	}
494 
495 	KASSERT(ip_nfrags == nfrags);
496 
497 	/* Find median (or other drop fraction) in histogram. */
498 	dropfraction = (ip_nfrags / 2);
499 	keepfraction = ip_nfrags - dropfraction;
500 	for (i = IPFRAGTTL, median = 0; i >= 0; i--) {
501 		median += fragttl_histo[i];
502 		if (median >= keepfraction)
503 			break;
504 	}
505 
506 	/* Return TTL of median (or other fraction). */
507 	return (u_int)i;
508 }
509 
510 static void
511 ip_reass_drophalf(void)
512 {
513 	u_int median_ticks;
514 
515 	KASSERT(mutex_owned(&ipfr_lock));
516 
517 	/*
518 	 * Compute median TTL of all fragments, and count frags
519 	 * with that TTL or lower (roughly half of all fragments).
520 	 */
521 	median_ticks = ip_reass_ttl_decr(0);
522 
523 	/* Drop half. */
524 	median_ticks = ip_reass_ttl_decr(median_ticks);
525 }
526 
527 /*
528  * ip_reass_drain: drain off all datagram fragments.  Do not acquire
529  * softnet_lock as can be called from hardware interrupt context.
530  */
531 void
532 ip_reass_drain(void)
533 {
534 
535 	/*
536 	 * We may be called from a device's interrupt context.  If
537 	 * the ipq is already busy, just bail out now.
538 	 */
539 	if (mutex_tryenter(&ipfr_lock)) {
540 		/*
541 		 * Drop half the total fragments now. If more mbufs are
542 		 * needed, we will be called again soon.
543 		 */
544 		ip_reass_drophalf();
545 		mutex_exit(&ipfr_lock);
546 	}
547 }
548 
549 /*
550  * ip_reass_slowtimo:
551  *
552  *	If a timer expires on a reassembly queue, discard it.
553  */
554 void
555 ip_reass_slowtimo(void)
556 {
557 	static u_int dropscanidx = 0;
558 	u_int i, median_ttl;
559 
560 	mutex_enter(&ipfr_lock);
561 
562 	/* Age TTL of all fragments by 1 tick .*/
563 	median_ttl = ip_reass_ttl_decr(1);
564 
565 	/* Make sure fragment limit is up-to-date. */
566 	CHECK_NMBCLUSTER_PARAMS();
567 
568 	/* If we have too many fragments, drop the older half. */
569 	if (ip_nfrags > ip_maxfrags) {
570 		ip_reass_ttl_decr(median_ttl);
571 	}
572 
573 	/*
574 	 * If we are over the maximum number of fragmented packets (due to
575 	 * the limit being lowered), drain off enough to get down to the
576 	 * new limit.  Start draining from the reassembly hashqueue most
577 	 * recently drained.
578 	 */
579 	if (ip_maxfragpackets < 0)
580 		;
581 	else {
582 		int wrapped = 0;
583 
584 		i = dropscanidx;
585 		while (ip_nfragpackets > ip_maxfragpackets && wrapped == 0) {
586 			while (LIST_FIRST(&ip_frags[i]) != NULL) {
587 				ip_freef(LIST_FIRST(&ip_frags[i]));
588 			}
589 			if (++i >= IPREASS_HASH_SIZE) {
590 				i = 0;
591 			}
592 			/*
593 			 * Do not scan forever even if fragment counters are
594 			 * wrong: stop after scanning entire reassembly queue.
595 			 */
596 			if (i == dropscanidx) {
597 				wrapped = 1;
598 			}
599 		}
600 		dropscanidx = i;
601 	}
602 	mutex_exit(&ipfr_lock);
603 }
604 
605 /*
606  * ip_reass_packet: generic routine to perform IP reassembly.
607  *
608  * => Passed fragment should have IP_MF flag and/or offset set.
609  * => Fragment should not have other than IP_MF flags set.
610  *
611  * => Returns 0 on success or error otherwise.
612  * => On complete, m0 represents a constructed final packet.
613  */
614 int
615 ip_reass_packet(struct mbuf **m0)
616 {
617 	struct mbuf *m = *m0;
618 	struct ip *ip = mtod(m, struct ip *);
619 	const int hlen = ip->ip_hl << 2;
620 	const int len = ntohs(ip->ip_len);
621 	int ipsecflags = m->m_flags & (M_DECRYPTED|M_AUTHIPHDR);
622 	ipfr_queue_t *fp;
623 	ipfr_qent_t *ipqe;
624 	u_int hash, off, flen;
625 	bool mff;
626 
627 	/*
628 	 * Prevent TCP blind data attacks by not allowing non-initial
629 	 * fragments to start at less than 68 bytes (minimal fragment
630 	 * size) and making sure the first fragment is at least 68
631 	 * bytes.
632 	 */
633 	off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
634 	if ((off > 0 ? off + hlen : len) < IP_MINFRAGSIZE - 1) {
635 		IP_STATINC(IP_STAT_BADFRAGS);
636 		return EINVAL;
637 	}
638 
639 	if (off + len > IP_MAXPACKET) {
640 		IP_STATINC(IP_STAT_TOOLONG);
641 		return EINVAL;
642 	}
643 
644 	/*
645 	 * Fragment length and MF flag.  Make sure that fragments have
646 	 * a data length which is non-zero and multiple of 8 bytes.
647 	 */
648 	flen = ntohs(ip->ip_len) - hlen;
649 	mff = (ip->ip_off & htons(IP_MF)) != 0;
650 	if (mff && (flen == 0 || (flen & 0x7) != 0)) {
651 		IP_STATINC(IP_STAT_BADFRAGS);
652 		return EINVAL;
653 	}
654 
655 	/*
656 	 * Adjust total IP length to not reflect header and convert
657 	 * offset of this to bytes.  XXX: clobbers struct ip.
658 	 */
659 	ip->ip_len = htons(flen);
660 	ip->ip_off = htons(off);
661 
662 	/* Look for queue of fragments of this datagram. */
663 	mutex_enter(&ipfr_lock);
664 	hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
665 	LIST_FOREACH(fp, &ip_frags[hash], ipq_q) {
666 		if (ip->ip_id != fp->ipq_id)
667 			continue;
668 		if (!in_hosteq(ip->ip_src, fp->ipq_src))
669 			continue;
670 		if (!in_hosteq(ip->ip_dst, fp->ipq_dst))
671 			continue;
672 		if (ip->ip_p != fp->ipq_p)
673 			continue;
674 		break;
675 	}
676 
677 	if (fp) {
678 		/* All fragments must have the same IPsec flags. */
679 		if (fp->ipq_ipsec != ipsecflags) {
680 			IP_STATINC(IP_STAT_BADFRAGS);
681 			mutex_exit(&ipfr_lock);
682 			return EINVAL;
683 		}
684 
685 		/* Make sure that TOS matches previous fragments. */
686 		if (fp->ipq_tos != ip->ip_tos) {
687 			IP_STATINC(IP_STAT_BADFRAGS);
688 			mutex_exit(&ipfr_lock);
689 			return EINVAL;
690 		}
691 	}
692 
693 	/*
694 	 * Create new entry and attempt to reassembly.
695 	 */
696 	IP_STATINC(IP_STAT_FRAGMENTS);
697 	ipqe = pool_cache_get(ipfren_cache, PR_NOWAIT);
698 	if (ipqe == NULL) {
699 		IP_STATINC(IP_STAT_RCVMEMDROP);
700 		mutex_exit(&ipfr_lock);
701 		return ENOMEM;
702 	}
703 	ipqe->ipqe_mff = mff;
704 	ipqe->ipqe_m = m;
705 	ipqe->ipqe_ip = ip;
706 
707 	*m0 = ip_reass(ipqe, fp, hash);
708 	if (*m0) {
709 		/* Note that finally reassembled. */
710 		IP_STATINC(IP_STAT_REASSEMBLED);
711 	}
712 	return 0;
713 }
714