xref: /freebsd/sys/kern/uipc_mbuf.c (revision 6e660824a82f590542932de52f128db584029893)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)uipc_mbuf.c	8.2 (Berkeley) 1/4/94
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_param.h"
36 #include "opt_mbuf_stress_test.h"
37 #include "opt_mbuf_profiling.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #include <sys/limits.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/sysctl.h>
47 #include <sys/domain.h>
48 #include <sys/protosw.h>
49 #include <sys/uio.h>
50 
51 int	max_linkhdr;
52 int	max_protohdr;
53 int	max_hdr;
54 int	max_datalen;
55 #ifdef MBUF_STRESS_TEST
56 int	m_defragpackets;
57 int	m_defragbytes;
58 int	m_defraguseless;
59 int	m_defragfailure;
60 int	m_defragrandomfailures;
61 #endif
62 
63 /*
64  * sysctl(8) exported objects
65  */
66 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD,
67 	   &max_linkhdr, 0, "Size of largest link layer header");
68 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD,
69 	   &max_protohdr, 0, "Size of largest protocol layer header");
70 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD,
71 	   &max_hdr, 0, "Size of largest link plus protocol header");
72 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RD,
73 	   &max_datalen, 0, "Minimum space left in mbuf after max_hdr");
74 #ifdef MBUF_STRESS_TEST
75 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
76 	   &m_defragpackets, 0, "");
77 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
78 	   &m_defragbytes, 0, "");
79 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
80 	   &m_defraguseless, 0, "");
81 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
82 	   &m_defragfailure, 0, "");
83 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
84 	   &m_defragrandomfailures, 0, "");
85 #endif
86 
87 /*
88  * m_get2() allocates minimum mbuf that would fit "size" argument.
89  */
90 struct mbuf *
91 m_get2(int size, int how, short type, int flags)
92 {
93 	struct mb_args args;
94 	struct mbuf *m, *n;
95 
96 	args.flags = flags;
97 	args.type = type;
98 
99 	if (size <= MHLEN || (size <= MLEN && (flags & M_PKTHDR) == 0))
100 		return (uma_zalloc_arg(zone_mbuf, &args, how));
101 	if (size <= MCLBYTES)
102 		return (uma_zalloc_arg(zone_pack, &args, how));
103 
104 	if (size > MJUMPAGESIZE)
105 		return (NULL);
106 
107 	m = uma_zalloc_arg(zone_mbuf, &args, how);
108 	if (m == NULL)
109 		return (NULL);
110 
111 	n = uma_zalloc_arg(zone_jumbop, m, how);
112 	if (n == NULL) {
113 		uma_zfree(zone_mbuf, m);
114 		return (NULL);
115 	}
116 
117 	return (m);
118 }
119 
120 /*
121  * m_getjcl() returns an mbuf with a cluster of the specified size attached.
122  * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES.
123  */
124 struct mbuf *
125 m_getjcl(int how, short type, int flags, int size)
126 {
127 	struct mb_args args;
128 	struct mbuf *m, *n;
129 	uma_zone_t zone;
130 
131 	if (size == MCLBYTES)
132 		return m_getcl(how, type, flags);
133 
134 	args.flags = flags;
135 	args.type = type;
136 
137 	m = uma_zalloc_arg(zone_mbuf, &args, how);
138 	if (m == NULL)
139 		return (NULL);
140 
141 	zone = m_getzone(size);
142 	n = uma_zalloc_arg(zone, m, how);
143 	if (n == NULL) {
144 		uma_zfree(zone_mbuf, m);
145 		return (NULL);
146 	}
147 	return (m);
148 }
149 
150 /*
151  * Allocate a given length worth of mbufs and/or clusters (whatever fits
152  * best) and return a pointer to the top of the allocated chain.  If an
153  * existing mbuf chain is provided, then we will append the new chain
154  * to the existing one but still return the top of the newly allocated
155  * chain.
156  */
157 struct mbuf *
158 m_getm2(struct mbuf *m, int len, int how, short type, int flags)
159 {
160 	struct mbuf *mb, *nm = NULL, *mtail = NULL;
161 
162 	KASSERT(len >= 0, ("%s: len is < 0", __func__));
163 
164 	/* Validate flags. */
165 	flags &= (M_PKTHDR | M_EOR);
166 
167 	/* Packet header mbuf must be first in chain. */
168 	if ((flags & M_PKTHDR) && m != NULL)
169 		flags &= ~M_PKTHDR;
170 
171 	/* Loop and append maximum sized mbufs to the chain tail. */
172 	while (len > 0) {
173 		if (len > MCLBYTES)
174 			mb = m_getjcl(how, type, (flags & M_PKTHDR),
175 			    MJUMPAGESIZE);
176 		else if (len >= MINCLSIZE)
177 			mb = m_getcl(how, type, (flags & M_PKTHDR));
178 		else if (flags & M_PKTHDR)
179 			mb = m_gethdr(how, type);
180 		else
181 			mb = m_get(how, type);
182 
183 		/* Fail the whole operation if one mbuf can't be allocated. */
184 		if (mb == NULL) {
185 			if (nm != NULL)
186 				m_freem(nm);
187 			return (NULL);
188 		}
189 
190 		/* Book keeping. */
191 		len -= (mb->m_flags & M_EXT) ? mb->m_ext.ext_size :
192 			((mb->m_flags & M_PKTHDR) ? MHLEN : MLEN);
193 		if (mtail != NULL)
194 			mtail->m_next = mb;
195 		else
196 			nm = mb;
197 		mtail = mb;
198 		flags &= ~M_PKTHDR;	/* Only valid on the first mbuf. */
199 	}
200 	if (flags & M_EOR)
201 		mtail->m_flags |= M_EOR;  /* Only valid on the last mbuf. */
202 
203 	/* If mbuf was supplied, append new chain to the end of it. */
204 	if (m != NULL) {
205 		for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next)
206 			;
207 		mtail->m_next = nm;
208 		mtail->m_flags &= ~M_EOR;
209 	} else
210 		m = nm;
211 
212 	return (m);
213 }
214 
215 /*
216  * Free an entire chain of mbufs and associated external buffers, if
217  * applicable.
218  */
219 void
220 m_freem(struct mbuf *mb)
221 {
222 
223 	while (mb != NULL)
224 		mb = m_free(mb);
225 }
226 
227 /*-
228  * Configure a provided mbuf to refer to the provided external storage
229  * buffer and setup a reference count for said buffer.  If the setting
230  * up of the reference count fails, the M_EXT bit will not be set.  If
231  * successfull, the M_EXT bit is set in the mbuf's flags.
232  *
233  * Arguments:
234  *    mb     The existing mbuf to which to attach the provided buffer.
235  *    buf    The address of the provided external storage buffer.
236  *    size   The size of the provided buffer.
237  *    freef  A pointer to a routine that is responsible for freeing the
238  *           provided external storage buffer.
239  *    args   A pointer to an argument structure (of any type) to be passed
240  *           to the provided freef routine (may be NULL).
241  *    flags  Any other flags to be passed to the provided mbuf.
242  *    type   The type that the external storage buffer should be
243  *           labeled with.
244  *
245  * Returns:
246  *    Nothing.
247  */
248 int
249 m_extadd(struct mbuf *mb, caddr_t buf, u_int size,
250     void (*freef)(void *, void *), void *arg1, void *arg2, int flags, int type,
251     int wait)
252 {
253 	KASSERT(type != EXT_CLUSTER, ("%s: EXT_CLUSTER not allowed", __func__));
254 
255 	if (type != EXT_EXTREF)
256 		mb->m_ext.ref_cnt = uma_zalloc(zone_ext_refcnt, wait);
257 
258 	if (mb->m_ext.ref_cnt == NULL)
259 		return (ENOMEM);
260 
261 	*(mb->m_ext.ref_cnt) = 1;
262 	mb->m_flags |= (M_EXT | flags);
263 	mb->m_ext.ext_buf = buf;
264 	mb->m_data = mb->m_ext.ext_buf;
265 	mb->m_ext.ext_size = size;
266 	mb->m_ext.ext_free = freef;
267 	mb->m_ext.ext_arg1 = arg1;
268 	mb->m_ext.ext_arg2 = arg2;
269 	mb->m_ext.ext_type = type;
270 
271 	return (0);
272 }
273 
274 /*
275  * Non-directly-exported function to clean up after mbufs with M_EXT
276  * storage attached to them if the reference count hits 1.
277  */
278 void
279 mb_free_ext(struct mbuf *m)
280 {
281 	int skipmbuf;
282 
283 	KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
284 	KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
285 
286 
287 	/*
288 	 * check if the header is embedded in the cluster
289 	 */
290 	skipmbuf = (m->m_flags & M_NOFREE);
291 
292 	/* Free attached storage if this mbuf is the only reference to it. */
293 	if (*(m->m_ext.ref_cnt) == 1 ||
294 	    atomic_fetchadd_int(m->m_ext.ref_cnt, -1) == 1) {
295 		switch (m->m_ext.ext_type) {
296 		case EXT_PACKET:	/* The packet zone is special. */
297 			if (*(m->m_ext.ref_cnt) == 0)
298 				*(m->m_ext.ref_cnt) = 1;
299 			uma_zfree(zone_pack, m);
300 			return;		/* Job done. */
301 		case EXT_CLUSTER:
302 			uma_zfree(zone_clust, m->m_ext.ext_buf);
303 			break;
304 		case EXT_JUMBOP:
305 			uma_zfree(zone_jumbop, m->m_ext.ext_buf);
306 			break;
307 		case EXT_JUMBO9:
308 			uma_zfree(zone_jumbo9, m->m_ext.ext_buf);
309 			break;
310 		case EXT_JUMBO16:
311 			uma_zfree(zone_jumbo16, m->m_ext.ext_buf);
312 			break;
313 		case EXT_SFBUF:
314 		case EXT_NET_DRV:
315 		case EXT_MOD_TYPE:
316 		case EXT_DISPOSABLE:
317 			*(m->m_ext.ref_cnt) = 0;
318 			uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *,
319 				m->m_ext.ref_cnt));
320 			/* FALLTHROUGH */
321 		case EXT_EXTREF:
322 			KASSERT(m->m_ext.ext_free != NULL,
323 				("%s: ext_free not set", __func__));
324 			(*(m->m_ext.ext_free))(m->m_ext.ext_arg1,
325 			    m->m_ext.ext_arg2);
326 			break;
327 		default:
328 			KASSERT(m->m_ext.ext_type == 0,
329 				("%s: unknown ext_type", __func__));
330 		}
331 	}
332 	if (skipmbuf)
333 		return;
334 
335 	/*
336 	 * Free this mbuf back to the mbuf zone with all m_ext
337 	 * information purged.
338 	 */
339 	m->m_ext.ext_buf = NULL;
340 	m->m_ext.ext_free = NULL;
341 	m->m_ext.ext_arg1 = NULL;
342 	m->m_ext.ext_arg2 = NULL;
343 	m->m_ext.ref_cnt = NULL;
344 	m->m_ext.ext_size = 0;
345 	m->m_ext.ext_type = 0;
346 	m->m_flags &= ~M_EXT;
347 	uma_zfree(zone_mbuf, m);
348 }
349 
350 /*
351  * Attach the cluster from *m to *n, set up m_ext in *n
352  * and bump the refcount of the cluster.
353  */
354 static void
355 mb_dupcl(struct mbuf *n, struct mbuf *m)
356 {
357 	KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
358 	KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
359 	KASSERT((n->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
360 
361 	if (*(m->m_ext.ref_cnt) == 1)
362 		*(m->m_ext.ref_cnt) += 1;
363 	else
364 		atomic_add_int(m->m_ext.ref_cnt, 1);
365 	n->m_ext.ext_buf = m->m_ext.ext_buf;
366 	n->m_ext.ext_free = m->m_ext.ext_free;
367 	n->m_ext.ext_arg1 = m->m_ext.ext_arg1;
368 	n->m_ext.ext_arg2 = m->m_ext.ext_arg2;
369 	n->m_ext.ext_size = m->m_ext.ext_size;
370 	n->m_ext.ref_cnt = m->m_ext.ref_cnt;
371 	n->m_ext.ext_type = m->m_ext.ext_type;
372 	n->m_flags |= M_EXT;
373 	n->m_flags |= m->m_flags & M_RDONLY;
374 }
375 
376 /*
377  * Clean up mbuf (chain) from any tags and packet headers.
378  * If "all" is set then the first mbuf in the chain will be
379  * cleaned too.
380  */
381 void
382 m_demote(struct mbuf *m0, int all)
383 {
384 	struct mbuf *m;
385 
386 	for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) {
387 		if (m->m_flags & M_PKTHDR) {
388 			m_tag_delete_chain(m, NULL);
389 			m->m_flags &= ~M_PKTHDR;
390 			bzero(&m->m_pkthdr, sizeof(struct pkthdr));
391 		}
392 		if (m != m0 && m->m_nextpkt != NULL) {
393 			KASSERT(m->m_nextpkt == NULL,
394 			    ("%s: m_nextpkt not NULL", __func__));
395 			m_freem(m->m_nextpkt);
396 			m->m_nextpkt = NULL;
397 		}
398 		m->m_flags = m->m_flags & (M_EXT|M_RDONLY|M_NOFREE);
399 	}
400 }
401 
402 /*
403  * Sanity checks on mbuf (chain) for use in KASSERT() and general
404  * debugging.
405  * Returns 0 or panics when bad and 1 on all tests passed.
406  * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they
407  * blow up later.
408  */
409 int
410 m_sanity(struct mbuf *m0, int sanitize)
411 {
412 	struct mbuf *m;
413 	caddr_t a, b;
414 	int pktlen = 0;
415 
416 #ifdef INVARIANTS
417 #define	M_SANITY_ACTION(s)	panic("mbuf %p: " s, m)
418 #else
419 #define	M_SANITY_ACTION(s)	printf("mbuf %p: " s, m)
420 #endif
421 
422 	for (m = m0; m != NULL; m = m->m_next) {
423 		/*
424 		 * Basic pointer checks.  If any of these fails then some
425 		 * unrelated kernel memory before or after us is trashed.
426 		 * No way to recover from that.
427 		 */
428 		a = ((m->m_flags & M_EXT) ? m->m_ext.ext_buf :
429 			((m->m_flags & M_PKTHDR) ? (caddr_t)(&m->m_pktdat) :
430 			 (caddr_t)(&m->m_dat)) );
431 		b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size :
432 			((m->m_flags & M_PKTHDR) ? MHLEN : MLEN)));
433 		if ((caddr_t)m->m_data < a)
434 			M_SANITY_ACTION("m_data outside mbuf data range left");
435 		if ((caddr_t)m->m_data > b)
436 			M_SANITY_ACTION("m_data outside mbuf data range right");
437 		if ((caddr_t)m->m_data + m->m_len > b)
438 			M_SANITY_ACTION("m_data + m_len exeeds mbuf space");
439 		if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.header) {
440 			if ((caddr_t)m->m_pkthdr.header < a ||
441 			    (caddr_t)m->m_pkthdr.header > b)
442 				M_SANITY_ACTION("m_pkthdr.header outside mbuf data range");
443 		}
444 
445 		/* m->m_nextpkt may only be set on first mbuf in chain. */
446 		if (m != m0 && m->m_nextpkt != NULL) {
447 			if (sanitize) {
448 				m_freem(m->m_nextpkt);
449 				m->m_nextpkt = (struct mbuf *)0xDEADC0DE;
450 			} else
451 				M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf");
452 		}
453 
454 		/* packet length (not mbuf length!) calculation */
455 		if (m0->m_flags & M_PKTHDR)
456 			pktlen += m->m_len;
457 
458 		/* m_tags may only be attached to first mbuf in chain. */
459 		if (m != m0 && m->m_flags & M_PKTHDR &&
460 		    !SLIST_EMPTY(&m->m_pkthdr.tags)) {
461 			if (sanitize) {
462 				m_tag_delete_chain(m, NULL);
463 				/* put in 0xDEADC0DE perhaps? */
464 			} else
465 				M_SANITY_ACTION("m_tags on in-chain mbuf");
466 		}
467 
468 		/* M_PKTHDR may only be set on first mbuf in chain */
469 		if (m != m0 && m->m_flags & M_PKTHDR) {
470 			if (sanitize) {
471 				bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
472 				m->m_flags &= ~M_PKTHDR;
473 				/* put in 0xDEADCODE and leave hdr flag in */
474 			} else
475 				M_SANITY_ACTION("M_PKTHDR on in-chain mbuf");
476 		}
477 	}
478 	m = m0;
479 	if (pktlen && pktlen != m->m_pkthdr.len) {
480 		if (sanitize)
481 			m->m_pkthdr.len = 0;
482 		else
483 			M_SANITY_ACTION("m_pkthdr.len != mbuf chain length");
484 	}
485 	return 1;
486 
487 #undef	M_SANITY_ACTION
488 }
489 
490 
491 /*
492  * "Move" mbuf pkthdr from "from" to "to".
493  * "from" must have M_PKTHDR set, and "to" must be empty.
494  */
495 void
496 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
497 {
498 
499 #if 0
500 	/* see below for why these are not enabled */
501 	M_ASSERTPKTHDR(to);
502 	/* Note: with MAC, this may not be a good assertion. */
503 	KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
504 	    ("m_move_pkthdr: to has tags"));
505 #endif
506 #ifdef MAC
507 	/*
508 	 * XXXMAC: It could be this should also occur for non-MAC?
509 	 */
510 	if (to->m_flags & M_PKTHDR)
511 		m_tag_delete_chain(to, NULL);
512 #endif
513 	to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
514 	if ((to->m_flags & M_EXT) == 0)
515 		to->m_data = to->m_pktdat;
516 	to->m_pkthdr = from->m_pkthdr;		/* especially tags */
517 	SLIST_INIT(&from->m_pkthdr.tags);	/* purge tags from src */
518 	from->m_flags &= ~M_PKTHDR;
519 }
520 
521 /*
522  * Duplicate "from"'s mbuf pkthdr in "to".
523  * "from" must have M_PKTHDR set, and "to" must be empty.
524  * In particular, this does a deep copy of the packet tags.
525  */
526 int
527 m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how)
528 {
529 
530 #if 0
531 	/*
532 	 * The mbuf allocator only initializes the pkthdr
533 	 * when the mbuf is allocated with m_gethdr(). Many users
534 	 * (e.g. m_copy*, m_prepend) use m_get() and then
535 	 * smash the pkthdr as needed causing these
536 	 * assertions to trip.  For now just disable them.
537 	 */
538 	M_ASSERTPKTHDR(to);
539 	/* Note: with MAC, this may not be a good assertion. */
540 	KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
541 #endif
542 	MBUF_CHECKSLEEP(how);
543 #ifdef MAC
544 	if (to->m_flags & M_PKTHDR)
545 		m_tag_delete_chain(to, NULL);
546 #endif
547 	to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
548 	if ((to->m_flags & M_EXT) == 0)
549 		to->m_data = to->m_pktdat;
550 	to->m_pkthdr = from->m_pkthdr;
551 	SLIST_INIT(&to->m_pkthdr.tags);
552 	return (m_tag_copy_chain(to, from, MBTOM(how)));
553 }
554 
555 /*
556  * Lesser-used path for M_PREPEND:
557  * allocate new mbuf to prepend to chain,
558  * copy junk along.
559  */
560 struct mbuf *
561 m_prepend(struct mbuf *m, int len, int how)
562 {
563 	struct mbuf *mn;
564 
565 	if (m->m_flags & M_PKTHDR)
566 		mn = m_gethdr(how, m->m_type);
567 	else
568 		mn = m_get(how, m->m_type);
569 	if (mn == NULL) {
570 		m_freem(m);
571 		return (NULL);
572 	}
573 	if (m->m_flags & M_PKTHDR)
574 		m_move_pkthdr(mn, m);
575 	mn->m_next = m;
576 	m = mn;
577 	if(m->m_flags & M_PKTHDR) {
578 		if (len < MHLEN)
579 			MH_ALIGN(m, len);
580 	} else {
581 		if (len < MLEN)
582 			M_ALIGN(m, len);
583 	}
584 	m->m_len = len;
585 	return (m);
586 }
587 
588 /*
589  * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
590  * continuing for "len" bytes.  If len is M_COPYALL, copy to end of mbuf.
591  * The wait parameter is a choice of M_WAITOK/M_NOWAIT from caller.
592  * Note that the copy is read-only, because clusters are not copied,
593  * only their reference counts are incremented.
594  */
595 struct mbuf *
596 m_copym(struct mbuf *m, int off0, int len, int wait)
597 {
598 	struct mbuf *n, **np;
599 	int off = off0;
600 	struct mbuf *top;
601 	int copyhdr = 0;
602 
603 	KASSERT(off >= 0, ("m_copym, negative off %d", off));
604 	KASSERT(len >= 0, ("m_copym, negative len %d", len));
605 	MBUF_CHECKSLEEP(wait);
606 	if (off == 0 && m->m_flags & M_PKTHDR)
607 		copyhdr = 1;
608 	while (off > 0) {
609 		KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
610 		if (off < m->m_len)
611 			break;
612 		off -= m->m_len;
613 		m = m->m_next;
614 	}
615 	np = &top;
616 	top = 0;
617 	while (len > 0) {
618 		if (m == NULL) {
619 			KASSERT(len == M_COPYALL,
620 			    ("m_copym, length > size of mbuf chain"));
621 			break;
622 		}
623 		if (copyhdr)
624 			n = m_gethdr(wait, m->m_type);
625 		else
626 			n = m_get(wait, m->m_type);
627 		*np = n;
628 		if (n == NULL)
629 			goto nospace;
630 		if (copyhdr) {
631 			if (!m_dup_pkthdr(n, m, wait))
632 				goto nospace;
633 			if (len == M_COPYALL)
634 				n->m_pkthdr.len -= off0;
635 			else
636 				n->m_pkthdr.len = len;
637 			copyhdr = 0;
638 		}
639 		n->m_len = min(len, m->m_len - off);
640 		if (m->m_flags & M_EXT) {
641 			n->m_data = m->m_data + off;
642 			mb_dupcl(n, m);
643 		} else
644 			bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
645 			    (u_int)n->m_len);
646 		if (len != M_COPYALL)
647 			len -= n->m_len;
648 		off = 0;
649 		m = m->m_next;
650 		np = &n->m_next;
651 	}
652 
653 	return (top);
654 nospace:
655 	m_freem(top);
656 	return (NULL);
657 }
658 
659 /*
660  * Returns mbuf chain with new head for the prepending case.
661  * Copies from mbuf (chain) n from off for len to mbuf (chain) m
662  * either prepending or appending the data.
663  * The resulting mbuf (chain) m is fully writeable.
664  * m is destination (is made writeable)
665  * n is source, off is offset in source, len is len from offset
666  * dir, 0 append, 1 prepend
667  * how, wait or nowait
668  */
669 
670 static int
671 m_bcopyxxx(void *s, void *t, u_int len)
672 {
673 	bcopy(s, t, (size_t)len);
674 	return 0;
675 }
676 
677 struct mbuf *
678 m_copymdata(struct mbuf *m, struct mbuf *n, int off, int len,
679     int prep, int how)
680 {
681 	struct mbuf *mm, *x, *z, *prev = NULL;
682 	caddr_t p;
683 	int i, nlen = 0;
684 	caddr_t buf[MLEN];
685 
686 	KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source"));
687 	KASSERT(off >= 0, ("m_copymdata, negative off %d", off));
688 	KASSERT(len >= 0, ("m_copymdata, negative len %d", len));
689 	KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep));
690 
691 	mm = m;
692 	if (!prep) {
693 		while(mm->m_next) {
694 			prev = mm;
695 			mm = mm->m_next;
696 		}
697 	}
698 	for (z = n; z != NULL; z = z->m_next)
699 		nlen += z->m_len;
700 	if (len == M_COPYALL)
701 		len = nlen - off;
702 	if (off + len > nlen || len < 1)
703 		return NULL;
704 
705 	if (!M_WRITABLE(mm)) {
706 		/* XXX: Use proper m_xxx function instead. */
707 		x = m_getcl(how, MT_DATA, mm->m_flags);
708 		if (x == NULL)
709 			return NULL;
710 		bcopy(mm->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size);
711 		p = x->m_ext.ext_buf + (mm->m_data - mm->m_ext.ext_buf);
712 		x->m_data = p;
713 		mm->m_next = NULL;
714 		if (mm != m)
715 			prev->m_next = x;
716 		m_free(mm);
717 		mm = x;
718 	}
719 
720 	/*
721 	 * Append/prepend the data.  Allocating mbufs as necessary.
722 	 */
723 	/* Shortcut if enough free space in first/last mbuf. */
724 	if (!prep && M_TRAILINGSPACE(mm) >= len) {
725 		m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) +
726 			 mm->m_len);
727 		mm->m_len += len;
728 		mm->m_pkthdr.len += len;
729 		return m;
730 	}
731 	if (prep && M_LEADINGSPACE(mm) >= len) {
732 		mm->m_data = mtod(mm, caddr_t) - len;
733 		m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t));
734 		mm->m_len += len;
735 		mm->m_pkthdr.len += len;
736 		return mm;
737 	}
738 
739 	/* Expand first/last mbuf to cluster if possible. */
740 	if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) {
741 		bcopy(mm->m_data, &buf, mm->m_len);
742 		m_clget(mm, how);
743 		if (!(mm->m_flags & M_EXT))
744 			return NULL;
745 		bcopy(&buf, mm->m_ext.ext_buf, mm->m_len);
746 		mm->m_data = mm->m_ext.ext_buf;
747 		mm->m_pkthdr.header = NULL;
748 	}
749 	if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) {
750 		bcopy(mm->m_data, &buf, mm->m_len);
751 		m_clget(mm, how);
752 		if (!(mm->m_flags & M_EXT))
753 			return NULL;
754 		bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf +
755 		       mm->m_ext.ext_size - mm->m_len, mm->m_len);
756 		mm->m_data = (caddr_t)mm->m_ext.ext_buf +
757 			      mm->m_ext.ext_size - mm->m_len;
758 		mm->m_pkthdr.header = NULL;
759 	}
760 
761 	/* Append/prepend as many mbuf (clusters) as necessary to fit len. */
762 	if (!prep && len > M_TRAILINGSPACE(mm)) {
763 		if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA))
764 			return NULL;
765 	}
766 	if (prep && len > M_LEADINGSPACE(mm)) {
767 		if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA)))
768 			return NULL;
769 		i = 0;
770 		for (x = z; x != NULL; x = x->m_next) {
771 			i += x->m_flags & M_EXT ? x->m_ext.ext_size :
772 			      (x->m_flags & M_PKTHDR ? MHLEN : MLEN);
773 			if (!x->m_next)
774 				break;
775 		}
776 		z->m_data += i - len;
777 		m_move_pkthdr(mm, z);
778 		x->m_next = mm;
779 		mm = z;
780 	}
781 
782 	/* Seek to start position in source mbuf. Optimization for long chains. */
783 	while (off > 0) {
784 		if (off < n->m_len)
785 			break;
786 		off -= n->m_len;
787 		n = n->m_next;
788 	}
789 
790 	/* Copy data into target mbuf. */
791 	z = mm;
792 	while (len > 0) {
793 		KASSERT(z != NULL, ("m_copymdata, falling off target edge"));
794 		i = M_TRAILINGSPACE(z);
795 		m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len);
796 		z->m_len += i;
797 		/* fixup pkthdr.len if necessary */
798 		if ((prep ? mm : m)->m_flags & M_PKTHDR)
799 			(prep ? mm : m)->m_pkthdr.len += i;
800 		off += i;
801 		len -= i;
802 		z = z->m_next;
803 	}
804 	return (prep ? mm : m);
805 }
806 
807 /*
808  * Copy an entire packet, including header (which must be present).
809  * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
810  * Note that the copy is read-only, because clusters are not copied,
811  * only their reference counts are incremented.
812  * Preserve alignment of the first mbuf so if the creator has left
813  * some room at the beginning (e.g. for inserting protocol headers)
814  * the copies still have the room available.
815  */
816 struct mbuf *
817 m_copypacket(struct mbuf *m, int how)
818 {
819 	struct mbuf *top, *n, *o;
820 
821 	MBUF_CHECKSLEEP(how);
822 	n = m_get(how, m->m_type);
823 	top = n;
824 	if (n == NULL)
825 		goto nospace;
826 
827 	if (!m_dup_pkthdr(n, m, how))
828 		goto nospace;
829 	n->m_len = m->m_len;
830 	if (m->m_flags & M_EXT) {
831 		n->m_data = m->m_data;
832 		mb_dupcl(n, m);
833 	} else {
834 		n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
835 		bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
836 	}
837 
838 	m = m->m_next;
839 	while (m) {
840 		o = m_get(how, m->m_type);
841 		if (o == NULL)
842 			goto nospace;
843 
844 		n->m_next = o;
845 		n = n->m_next;
846 
847 		n->m_len = m->m_len;
848 		if (m->m_flags & M_EXT) {
849 			n->m_data = m->m_data;
850 			mb_dupcl(n, m);
851 		} else {
852 			bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
853 		}
854 
855 		m = m->m_next;
856 	}
857 	return top;
858 nospace:
859 	m_freem(top);
860 	return (NULL);
861 }
862 
863 /*
864  * Copy data from an mbuf chain starting "off" bytes from the beginning,
865  * continuing for "len" bytes, into the indicated buffer.
866  */
867 void
868 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
869 {
870 	u_int count;
871 
872 	KASSERT(off >= 0, ("m_copydata, negative off %d", off));
873 	KASSERT(len >= 0, ("m_copydata, negative len %d", len));
874 	while (off > 0) {
875 		KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
876 		if (off < m->m_len)
877 			break;
878 		off -= m->m_len;
879 		m = m->m_next;
880 	}
881 	while (len > 0) {
882 		KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
883 		count = min(m->m_len - off, len);
884 		bcopy(mtod(m, caddr_t) + off, cp, count);
885 		len -= count;
886 		cp += count;
887 		off = 0;
888 		m = m->m_next;
889 	}
890 }
891 
892 /*
893  * Copy a packet header mbuf chain into a completely new chain, including
894  * copying any mbuf clusters.  Use this instead of m_copypacket() when
895  * you need a writable copy of an mbuf chain.
896  */
897 struct mbuf *
898 m_dup(struct mbuf *m, int how)
899 {
900 	struct mbuf **p, *top = NULL;
901 	int remain, moff, nsize;
902 
903 	MBUF_CHECKSLEEP(how);
904 	/* Sanity check */
905 	if (m == NULL)
906 		return (NULL);
907 	M_ASSERTPKTHDR(m);
908 
909 	/* While there's more data, get a new mbuf, tack it on, and fill it */
910 	remain = m->m_pkthdr.len;
911 	moff = 0;
912 	p = &top;
913 	while (remain > 0 || top == NULL) {	/* allow m->m_pkthdr.len == 0 */
914 		struct mbuf *n;
915 
916 		/* Get the next new mbuf */
917 		if (remain >= MINCLSIZE) {
918 			n = m_getcl(how, m->m_type, 0);
919 			nsize = MCLBYTES;
920 		} else {
921 			n = m_get(how, m->m_type);
922 			nsize = MLEN;
923 		}
924 		if (n == NULL)
925 			goto nospace;
926 
927 		if (top == NULL) {		/* First one, must be PKTHDR */
928 			if (!m_dup_pkthdr(n, m, how)) {
929 				m_free(n);
930 				goto nospace;
931 			}
932 			if ((n->m_flags & M_EXT) == 0)
933 				nsize = MHLEN;
934 		}
935 		n->m_len = 0;
936 
937 		/* Link it into the new chain */
938 		*p = n;
939 		p = &n->m_next;
940 
941 		/* Copy data from original mbuf(s) into new mbuf */
942 		while (n->m_len < nsize && m != NULL) {
943 			int chunk = min(nsize - n->m_len, m->m_len - moff);
944 
945 			bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
946 			moff += chunk;
947 			n->m_len += chunk;
948 			remain -= chunk;
949 			if (moff == m->m_len) {
950 				m = m->m_next;
951 				moff = 0;
952 			}
953 		}
954 
955 		/* Check correct total mbuf length */
956 		KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
957 		    	("%s: bogus m_pkthdr.len", __func__));
958 	}
959 	return (top);
960 
961 nospace:
962 	m_freem(top);
963 	return (NULL);
964 }
965 
966 /*
967  * Concatenate mbuf chain n to m.
968  * Both chains must be of the same type (e.g. MT_DATA).
969  * Any m_pkthdr is not updated.
970  */
971 void
972 m_cat(struct mbuf *m, struct mbuf *n)
973 {
974 	while (m->m_next)
975 		m = m->m_next;
976 	while (n) {
977 		if (!M_WRITABLE(m) ||
978 		    M_TRAILINGSPACE(m) < n->m_len) {
979 			/* just join the two chains */
980 			m->m_next = n;
981 			return;
982 		}
983 		/* splat the data from one into the other */
984 		bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
985 		    (u_int)n->m_len);
986 		m->m_len += n->m_len;
987 		n = m_free(n);
988 	}
989 }
990 
991 void
992 m_adj(struct mbuf *mp, int req_len)
993 {
994 	int len = req_len;
995 	struct mbuf *m;
996 	int count;
997 
998 	if ((m = mp) == NULL)
999 		return;
1000 	if (len >= 0) {
1001 		/*
1002 		 * Trim from head.
1003 		 */
1004 		while (m != NULL && len > 0) {
1005 			if (m->m_len <= len) {
1006 				len -= m->m_len;
1007 				m->m_len = 0;
1008 				m = m->m_next;
1009 			} else {
1010 				m->m_len -= len;
1011 				m->m_data += len;
1012 				len = 0;
1013 			}
1014 		}
1015 		if (mp->m_flags & M_PKTHDR)
1016 			mp->m_pkthdr.len -= (req_len - len);
1017 	} else {
1018 		/*
1019 		 * Trim from tail.  Scan the mbuf chain,
1020 		 * calculating its length and finding the last mbuf.
1021 		 * If the adjustment only affects this mbuf, then just
1022 		 * adjust and return.  Otherwise, rescan and truncate
1023 		 * after the remaining size.
1024 		 */
1025 		len = -len;
1026 		count = 0;
1027 		for (;;) {
1028 			count += m->m_len;
1029 			if (m->m_next == (struct mbuf *)0)
1030 				break;
1031 			m = m->m_next;
1032 		}
1033 		if (m->m_len >= len) {
1034 			m->m_len -= len;
1035 			if (mp->m_flags & M_PKTHDR)
1036 				mp->m_pkthdr.len -= len;
1037 			return;
1038 		}
1039 		count -= len;
1040 		if (count < 0)
1041 			count = 0;
1042 		/*
1043 		 * Correct length for chain is "count".
1044 		 * Find the mbuf with last data, adjust its length,
1045 		 * and toss data from remaining mbufs on chain.
1046 		 */
1047 		m = mp;
1048 		if (m->m_flags & M_PKTHDR)
1049 			m->m_pkthdr.len = count;
1050 		for (; m; m = m->m_next) {
1051 			if (m->m_len >= count) {
1052 				m->m_len = count;
1053 				if (m->m_next != NULL) {
1054 					m_freem(m->m_next);
1055 					m->m_next = NULL;
1056 				}
1057 				break;
1058 			}
1059 			count -= m->m_len;
1060 		}
1061 	}
1062 }
1063 
1064 /*
1065  * Rearange an mbuf chain so that len bytes are contiguous
1066  * and in the data area of an mbuf (so that mtod will work
1067  * for a structure of size len).  Returns the resulting
1068  * mbuf chain on success, frees it and returns null on failure.
1069  * If there is room, it will add up to max_protohdr-len extra bytes to the
1070  * contiguous region in an attempt to avoid being called next time.
1071  */
1072 struct mbuf *
1073 m_pullup(struct mbuf *n, int len)
1074 {
1075 	struct mbuf *m;
1076 	int count;
1077 	int space;
1078 
1079 	/*
1080 	 * If first mbuf has no cluster, and has room for len bytes
1081 	 * without shifting current data, pullup into it,
1082 	 * otherwise allocate a new mbuf to prepend to the chain.
1083 	 */
1084 	if ((n->m_flags & M_EXT) == 0 &&
1085 	    n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
1086 		if (n->m_len >= len)
1087 			return (n);
1088 		m = n;
1089 		n = n->m_next;
1090 		len -= m->m_len;
1091 	} else {
1092 		if (len > MHLEN)
1093 			goto bad;
1094 		m = m_get(M_NOWAIT, n->m_type);
1095 		if (m == NULL)
1096 			goto bad;
1097 		if (n->m_flags & M_PKTHDR)
1098 			m_move_pkthdr(m, n);
1099 	}
1100 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1101 	do {
1102 		count = min(min(max(len, max_protohdr), space), n->m_len);
1103 		bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1104 		  (u_int)count);
1105 		len -= count;
1106 		m->m_len += count;
1107 		n->m_len -= count;
1108 		space -= count;
1109 		if (n->m_len)
1110 			n->m_data += count;
1111 		else
1112 			n = m_free(n);
1113 	} while (len > 0 && n);
1114 	if (len > 0) {
1115 		(void) m_free(m);
1116 		goto bad;
1117 	}
1118 	m->m_next = n;
1119 	return (m);
1120 bad:
1121 	m_freem(n);
1122 	return (NULL);
1123 }
1124 
1125 /*
1126  * Like m_pullup(), except a new mbuf is always allocated, and we allow
1127  * the amount of empty space before the data in the new mbuf to be specified
1128  * (in the event that the caller expects to prepend later).
1129  */
1130 int MSFail;
1131 
1132 struct mbuf *
1133 m_copyup(struct mbuf *n, int len, int dstoff)
1134 {
1135 	struct mbuf *m;
1136 	int count, space;
1137 
1138 	if (len > (MHLEN - dstoff))
1139 		goto bad;
1140 	m = m_get(M_NOWAIT, n->m_type);
1141 	if (m == NULL)
1142 		goto bad;
1143 	if (n->m_flags & M_PKTHDR)
1144 		m_move_pkthdr(m, n);
1145 	m->m_data += dstoff;
1146 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1147 	do {
1148 		count = min(min(max(len, max_protohdr), space), n->m_len);
1149 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
1150 		    (unsigned)count);
1151 		len -= count;
1152 		m->m_len += count;
1153 		n->m_len -= count;
1154 		space -= count;
1155 		if (n->m_len)
1156 			n->m_data += count;
1157 		else
1158 			n = m_free(n);
1159 	} while (len > 0 && n);
1160 	if (len > 0) {
1161 		(void) m_free(m);
1162 		goto bad;
1163 	}
1164 	m->m_next = n;
1165 	return (m);
1166  bad:
1167 	m_freem(n);
1168 	MSFail++;
1169 	return (NULL);
1170 }
1171 
1172 /*
1173  * Partition an mbuf chain in two pieces, returning the tail --
1174  * all but the first len0 bytes.  In case of failure, it returns NULL and
1175  * attempts to restore the chain to its original state.
1176  *
1177  * Note that the resulting mbufs might be read-only, because the new
1178  * mbuf can end up sharing an mbuf cluster with the original mbuf if
1179  * the "breaking point" happens to lie within a cluster mbuf. Use the
1180  * M_WRITABLE() macro to check for this case.
1181  */
1182 struct mbuf *
1183 m_split(struct mbuf *m0, int len0, int wait)
1184 {
1185 	struct mbuf *m, *n;
1186 	u_int len = len0, remain;
1187 
1188 	MBUF_CHECKSLEEP(wait);
1189 	for (m = m0; m && len > m->m_len; m = m->m_next)
1190 		len -= m->m_len;
1191 	if (m == NULL)
1192 		return (NULL);
1193 	remain = m->m_len - len;
1194 	if (m0->m_flags & M_PKTHDR && remain == 0) {
1195 		n = m_gethdr(wait, m0->m_type);
1196 			return (NULL);
1197 		n->m_next = m->m_next;
1198 		m->m_next = NULL;
1199 		n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1200 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1201 		m0->m_pkthdr.len = len0;
1202 		return (n);
1203 	} else if (m0->m_flags & M_PKTHDR) {
1204 		n = m_gethdr(wait, m0->m_type);
1205 		if (n == NULL)
1206 			return (NULL);
1207 		n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1208 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1209 		m0->m_pkthdr.len = len0;
1210 		if (m->m_flags & M_EXT)
1211 			goto extpacket;
1212 		if (remain > MHLEN) {
1213 			/* m can't be the lead packet */
1214 			MH_ALIGN(n, 0);
1215 			n->m_next = m_split(m, len, wait);
1216 			if (n->m_next == NULL) {
1217 				(void) m_free(n);
1218 				return (NULL);
1219 			} else {
1220 				n->m_len = 0;
1221 				return (n);
1222 			}
1223 		} else
1224 			MH_ALIGN(n, remain);
1225 	} else if (remain == 0) {
1226 		n = m->m_next;
1227 		m->m_next = NULL;
1228 		return (n);
1229 	} else {
1230 		n = m_get(wait, m->m_type);
1231 		if (n == NULL)
1232 			return (NULL);
1233 		M_ALIGN(n, remain);
1234 	}
1235 extpacket:
1236 	if (m->m_flags & M_EXT) {
1237 		n->m_data = m->m_data + len;
1238 		mb_dupcl(n, m);
1239 	} else {
1240 		bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1241 	}
1242 	n->m_len = remain;
1243 	m->m_len = len;
1244 	n->m_next = m->m_next;
1245 	m->m_next = NULL;
1246 	return (n);
1247 }
1248 /*
1249  * Routine to copy from device local memory into mbufs.
1250  * Note that `off' argument is offset into first mbuf of target chain from
1251  * which to begin copying the data to.
1252  */
1253 struct mbuf *
1254 m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
1255     void (*copy)(char *from, caddr_t to, u_int len))
1256 {
1257 	struct mbuf *m;
1258 	struct mbuf *top = NULL, **mp = &top;
1259 	int len;
1260 
1261 	if (off < 0 || off > MHLEN)
1262 		return (NULL);
1263 
1264 	while (totlen > 0) {
1265 		if (top == NULL) {	/* First one, must be PKTHDR */
1266 			if (totlen + off >= MINCLSIZE) {
1267 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1268 				len = MCLBYTES;
1269 			} else {
1270 				m = m_gethdr(M_NOWAIT, MT_DATA);
1271 				len = MHLEN;
1272 
1273 				/* Place initial small packet/header at end of mbuf */
1274 				if (m && totlen + off + max_linkhdr <= MLEN) {
1275 					m->m_data += max_linkhdr;
1276 					len -= max_linkhdr;
1277 				}
1278 			}
1279 			if (m == NULL)
1280 				return NULL;
1281 			m->m_pkthdr.rcvif = ifp;
1282 			m->m_pkthdr.len = totlen;
1283 		} else {
1284 			if (totlen + off >= MINCLSIZE) {
1285 				m = m_getcl(M_NOWAIT, MT_DATA, 0);
1286 				len = MCLBYTES;
1287 			} else {
1288 				m = m_get(M_NOWAIT, MT_DATA);
1289 				len = MLEN;
1290 			}
1291 			if (m == NULL) {
1292 				m_freem(top);
1293 				return NULL;
1294 			}
1295 		}
1296 		if (off) {
1297 			m->m_data += off;
1298 			len -= off;
1299 			off = 0;
1300 		}
1301 		m->m_len = len = min(totlen, len);
1302 		if (copy)
1303 			copy(buf, mtod(m, caddr_t), (u_int)len);
1304 		else
1305 			bcopy(buf, mtod(m, caddr_t), (u_int)len);
1306 		buf += len;
1307 		*mp = m;
1308 		mp = &m->m_next;
1309 		totlen -= len;
1310 	}
1311 	return (top);
1312 }
1313 
1314 /*
1315  * Copy data from a buffer back into the indicated mbuf chain,
1316  * starting "off" bytes from the beginning, extending the mbuf
1317  * chain if necessary.
1318  */
1319 void
1320 m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
1321 {
1322 	int mlen;
1323 	struct mbuf *m = m0, *n;
1324 	int totlen = 0;
1325 
1326 	if (m0 == NULL)
1327 		return;
1328 	while (off > (mlen = m->m_len)) {
1329 		off -= mlen;
1330 		totlen += mlen;
1331 		if (m->m_next == NULL) {
1332 			n = m_get(M_NOWAIT, m->m_type);
1333 			if (n == NULL)
1334 				goto out;
1335 			bzero(mtod(n, caddr_t), MLEN);
1336 			n->m_len = min(MLEN, len + off);
1337 			m->m_next = n;
1338 		}
1339 		m = m->m_next;
1340 	}
1341 	while (len > 0) {
1342 		if (m->m_next == NULL && (len > m->m_len - off)) {
1343 			m->m_len += min(len - (m->m_len - off),
1344 			    M_TRAILINGSPACE(m));
1345 		}
1346 		mlen = min (m->m_len - off, len);
1347 		bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
1348 		cp += mlen;
1349 		len -= mlen;
1350 		mlen += off;
1351 		off = 0;
1352 		totlen += mlen;
1353 		if (len == 0)
1354 			break;
1355 		if (m->m_next == NULL) {
1356 			n = m_get(M_NOWAIT, m->m_type);
1357 			if (n == NULL)
1358 				break;
1359 			n->m_len = min(MLEN, len);
1360 			m->m_next = n;
1361 		}
1362 		m = m->m_next;
1363 	}
1364 out:	if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1365 		m->m_pkthdr.len = totlen;
1366 }
1367 
1368 /*
1369  * Append the specified data to the indicated mbuf chain,
1370  * Extend the mbuf chain if the new data does not fit in
1371  * existing space.
1372  *
1373  * Return 1 if able to complete the job; otherwise 0.
1374  */
1375 int
1376 m_append(struct mbuf *m0, int len, c_caddr_t cp)
1377 {
1378 	struct mbuf *m, *n;
1379 	int remainder, space;
1380 
1381 	for (m = m0; m->m_next != NULL; m = m->m_next)
1382 		;
1383 	remainder = len;
1384 	space = M_TRAILINGSPACE(m);
1385 	if (space > 0) {
1386 		/*
1387 		 * Copy into available space.
1388 		 */
1389 		if (space > remainder)
1390 			space = remainder;
1391 		bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1392 		m->m_len += space;
1393 		cp += space, remainder -= space;
1394 	}
1395 	while (remainder > 0) {
1396 		/*
1397 		 * Allocate a new mbuf; could check space
1398 		 * and allocate a cluster instead.
1399 		 */
1400 		n = m_get(M_NOWAIT, m->m_type);
1401 		if (n == NULL)
1402 			break;
1403 		n->m_len = min(MLEN, remainder);
1404 		bcopy(cp, mtod(n, caddr_t), n->m_len);
1405 		cp += n->m_len, remainder -= n->m_len;
1406 		m->m_next = n;
1407 		m = n;
1408 	}
1409 	if (m0->m_flags & M_PKTHDR)
1410 		m0->m_pkthdr.len += len - remainder;
1411 	return (remainder == 0);
1412 }
1413 
1414 /*
1415  * Apply function f to the data in an mbuf chain starting "off" bytes from
1416  * the beginning, continuing for "len" bytes.
1417  */
1418 int
1419 m_apply(struct mbuf *m, int off, int len,
1420     int (*f)(void *, void *, u_int), void *arg)
1421 {
1422 	u_int count;
1423 	int rval;
1424 
1425 	KASSERT(off >= 0, ("m_apply, negative off %d", off));
1426 	KASSERT(len >= 0, ("m_apply, negative len %d", len));
1427 	while (off > 0) {
1428 		KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1429 		if (off < m->m_len)
1430 			break;
1431 		off -= m->m_len;
1432 		m = m->m_next;
1433 	}
1434 	while (len > 0) {
1435 		KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1436 		count = min(m->m_len - off, len);
1437 		rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1438 		if (rval)
1439 			return (rval);
1440 		len -= count;
1441 		off = 0;
1442 		m = m->m_next;
1443 	}
1444 	return (0);
1445 }
1446 
1447 /*
1448  * Return a pointer to mbuf/offset of location in mbuf chain.
1449  */
1450 struct mbuf *
1451 m_getptr(struct mbuf *m, int loc, int *off)
1452 {
1453 
1454 	while (loc >= 0) {
1455 		/* Normal end of search. */
1456 		if (m->m_len > loc) {
1457 			*off = loc;
1458 			return (m);
1459 		} else {
1460 			loc -= m->m_len;
1461 			if (m->m_next == NULL) {
1462 				if (loc == 0) {
1463 					/* Point at the end of valid data. */
1464 					*off = m->m_len;
1465 					return (m);
1466 				}
1467 				return (NULL);
1468 			}
1469 			m = m->m_next;
1470 		}
1471 	}
1472 	return (NULL);
1473 }
1474 
1475 void
1476 m_print(const struct mbuf *m, int maxlen)
1477 {
1478 	int len;
1479 	int pdata;
1480 	const struct mbuf *m2;
1481 
1482 	if (m == NULL) {
1483 		printf("mbuf: %p\n", m);
1484 		return;
1485 	}
1486 
1487 	if (m->m_flags & M_PKTHDR)
1488 		len = m->m_pkthdr.len;
1489 	else
1490 		len = -1;
1491 	m2 = m;
1492 	while (m2 != NULL && (len == -1 || len)) {
1493 		pdata = m2->m_len;
1494 		if (maxlen != -1 && pdata > maxlen)
1495 			pdata = maxlen;
1496 		printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
1497 		    m2->m_next, m2->m_flags, "\20\20freelist\17skipfw"
1498 		    "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
1499 		    "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
1500 		if (pdata)
1501 			printf(", %*D\n", pdata, (u_char *)m2->m_data, "-");
1502 		if (len != -1)
1503 			len -= m2->m_len;
1504 		m2 = m2->m_next;
1505 	}
1506 	if (len > 0)
1507 		printf("%d bytes unaccounted for.\n", len);
1508 	return;
1509 }
1510 
1511 u_int
1512 m_fixhdr(struct mbuf *m0)
1513 {
1514 	u_int len;
1515 
1516 	len = m_length(m0, NULL);
1517 	m0->m_pkthdr.len = len;
1518 	return (len);
1519 }
1520 
1521 u_int
1522 m_length(struct mbuf *m0, struct mbuf **last)
1523 {
1524 	struct mbuf *m;
1525 	u_int len;
1526 
1527 	len = 0;
1528 	for (m = m0; m != NULL; m = m->m_next) {
1529 		len += m->m_len;
1530 		if (m->m_next == NULL)
1531 			break;
1532 	}
1533 	if (last != NULL)
1534 		*last = m;
1535 	return (len);
1536 }
1537 
1538 /*
1539  * Defragment a mbuf chain, returning the shortest possible
1540  * chain of mbufs and clusters.  If allocation fails and
1541  * this cannot be completed, NULL will be returned, but
1542  * the passed in chain will be unchanged.  Upon success,
1543  * the original chain will be freed, and the new chain
1544  * will be returned.
1545  *
1546  * If a non-packet header is passed in, the original
1547  * mbuf (chain?) will be returned unharmed.
1548  */
1549 struct mbuf *
1550 m_defrag(struct mbuf *m0, int how)
1551 {
1552 	struct mbuf *m_new = NULL, *m_final = NULL;
1553 	int progress = 0, length;
1554 
1555 	MBUF_CHECKSLEEP(how);
1556 	if (!(m0->m_flags & M_PKTHDR))
1557 		return (m0);
1558 
1559 	m_fixhdr(m0); /* Needed sanity check */
1560 
1561 #ifdef MBUF_STRESS_TEST
1562 	if (m_defragrandomfailures) {
1563 		int temp = arc4random() & 0xff;
1564 		if (temp == 0xba)
1565 			goto nospace;
1566 	}
1567 #endif
1568 
1569 	if (m0->m_pkthdr.len > MHLEN)
1570 		m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1571 	else
1572 		m_final = m_gethdr(how, MT_DATA);
1573 
1574 	if (m_final == NULL)
1575 		goto nospace;
1576 
1577 	if (m_dup_pkthdr(m_final, m0, how) == 0)
1578 		goto nospace;
1579 
1580 	m_new = m_final;
1581 
1582 	while (progress < m0->m_pkthdr.len) {
1583 		length = m0->m_pkthdr.len - progress;
1584 		if (length > MCLBYTES)
1585 			length = MCLBYTES;
1586 
1587 		if (m_new == NULL) {
1588 			if (length > MLEN)
1589 				m_new = m_getcl(how, MT_DATA, 0);
1590 			else
1591 				m_new = m_get(how, MT_DATA);
1592 			if (m_new == NULL)
1593 				goto nospace;
1594 		}
1595 
1596 		m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1597 		progress += length;
1598 		m_new->m_len = length;
1599 		if (m_new != m_final)
1600 			m_cat(m_final, m_new);
1601 		m_new = NULL;
1602 	}
1603 #ifdef MBUF_STRESS_TEST
1604 	if (m0->m_next == NULL)
1605 		m_defraguseless++;
1606 #endif
1607 	m_freem(m0);
1608 	m0 = m_final;
1609 #ifdef MBUF_STRESS_TEST
1610 	m_defragpackets++;
1611 	m_defragbytes += m0->m_pkthdr.len;
1612 #endif
1613 	return (m0);
1614 nospace:
1615 #ifdef MBUF_STRESS_TEST
1616 	m_defragfailure++;
1617 #endif
1618 	if (m_final)
1619 		m_freem(m_final);
1620 	return (NULL);
1621 }
1622 
1623 /*
1624  * Defragment an mbuf chain, returning at most maxfrags separate
1625  * mbufs+clusters.  If this is not possible NULL is returned and
1626  * the original mbuf chain is left in it's present (potentially
1627  * modified) state.  We use two techniques: collapsing consecutive
1628  * mbufs and replacing consecutive mbufs by a cluster.
1629  *
1630  * NB: this should really be named m_defrag but that name is taken
1631  */
1632 struct mbuf *
1633 m_collapse(struct mbuf *m0, int how, int maxfrags)
1634 {
1635 	struct mbuf *m, *n, *n2, **prev;
1636 	u_int curfrags;
1637 
1638 	/*
1639 	 * Calculate the current number of frags.
1640 	 */
1641 	curfrags = 0;
1642 	for (m = m0; m != NULL; m = m->m_next)
1643 		curfrags++;
1644 	/*
1645 	 * First, try to collapse mbufs.  Note that we always collapse
1646 	 * towards the front so we don't need to deal with moving the
1647 	 * pkthdr.  This may be suboptimal if the first mbuf has much
1648 	 * less data than the following.
1649 	 */
1650 	m = m0;
1651 again:
1652 	for (;;) {
1653 		n = m->m_next;
1654 		if (n == NULL)
1655 			break;
1656 		if (M_WRITABLE(m) &&
1657 		    n->m_len < M_TRAILINGSPACE(m)) {
1658 			bcopy(mtod(n, void *), mtod(m, char *) + m->m_len,
1659 				n->m_len);
1660 			m->m_len += n->m_len;
1661 			m->m_next = n->m_next;
1662 			m_free(n);
1663 			if (--curfrags <= maxfrags)
1664 				return m0;
1665 		} else
1666 			m = n;
1667 	}
1668 	KASSERT(maxfrags > 1,
1669 		("maxfrags %u, but normal collapse failed", maxfrags));
1670 	/*
1671 	 * Collapse consecutive mbufs to a cluster.
1672 	 */
1673 	prev = &m0->m_next;		/* NB: not the first mbuf */
1674 	while ((n = *prev) != NULL) {
1675 		if ((n2 = n->m_next) != NULL &&
1676 		    n->m_len + n2->m_len < MCLBYTES) {
1677 			m = m_getcl(how, MT_DATA, 0);
1678 			if (m == NULL)
1679 				goto bad;
1680 			bcopy(mtod(n, void *), mtod(m, void *), n->m_len);
1681 			bcopy(mtod(n2, void *), mtod(m, char *) + n->m_len,
1682 				n2->m_len);
1683 			m->m_len = n->m_len + n2->m_len;
1684 			m->m_next = n2->m_next;
1685 			*prev = m;
1686 			m_free(n);
1687 			m_free(n2);
1688 			if (--curfrags <= maxfrags)	/* +1 cl -2 mbufs */
1689 				return m0;
1690 			/*
1691 			 * Still not there, try the normal collapse
1692 			 * again before we allocate another cluster.
1693 			 */
1694 			goto again;
1695 		}
1696 		prev = &n->m_next;
1697 	}
1698 	/*
1699 	 * No place where we can collapse to a cluster; punt.
1700 	 * This can occur if, for example, you request 2 frags
1701 	 * but the packet requires that both be clusters (we
1702 	 * never reallocate the first mbuf to avoid moving the
1703 	 * packet header).
1704 	 */
1705 bad:
1706 	return NULL;
1707 }
1708 
1709 #ifdef MBUF_STRESS_TEST
1710 
1711 /*
1712  * Fragment an mbuf chain.  There's no reason you'd ever want to do
1713  * this in normal usage, but it's great for stress testing various
1714  * mbuf consumers.
1715  *
1716  * If fragmentation is not possible, the original chain will be
1717  * returned.
1718  *
1719  * Possible length values:
1720  * 0	 no fragmentation will occur
1721  * > 0	each fragment will be of the specified length
1722  * -1	each fragment will be the same random value in length
1723  * -2	each fragment's length will be entirely random
1724  * (Random values range from 1 to 256)
1725  */
1726 struct mbuf *
1727 m_fragment(struct mbuf *m0, int how, int length)
1728 {
1729 	struct mbuf *m_new = NULL, *m_final = NULL;
1730 	int progress = 0;
1731 
1732 	if (!(m0->m_flags & M_PKTHDR))
1733 		return (m0);
1734 
1735 	if ((length == 0) || (length < -2))
1736 		return (m0);
1737 
1738 	m_fixhdr(m0); /* Needed sanity check */
1739 
1740 	m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1741 
1742 	if (m_final == NULL)
1743 		goto nospace;
1744 
1745 	if (m_dup_pkthdr(m_final, m0, how) == 0)
1746 		goto nospace;
1747 
1748 	m_new = m_final;
1749 
1750 	if (length == -1)
1751 		length = 1 + (arc4random() & 255);
1752 
1753 	while (progress < m0->m_pkthdr.len) {
1754 		int fraglen;
1755 
1756 		if (length > 0)
1757 			fraglen = length;
1758 		else
1759 			fraglen = 1 + (arc4random() & 255);
1760 		if (fraglen > m0->m_pkthdr.len - progress)
1761 			fraglen = m0->m_pkthdr.len - progress;
1762 
1763 		if (fraglen > MCLBYTES)
1764 			fraglen = MCLBYTES;
1765 
1766 		if (m_new == NULL) {
1767 			m_new = m_getcl(how, MT_DATA, 0);
1768 			if (m_new == NULL)
1769 				goto nospace;
1770 		}
1771 
1772 		m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t));
1773 		progress += fraglen;
1774 		m_new->m_len = fraglen;
1775 		if (m_new != m_final)
1776 			m_cat(m_final, m_new);
1777 		m_new = NULL;
1778 	}
1779 	m_freem(m0);
1780 	m0 = m_final;
1781 	return (m0);
1782 nospace:
1783 	if (m_final)
1784 		m_freem(m_final);
1785 	/* Return the original chain on failure */
1786 	return (m0);
1787 }
1788 
1789 #endif
1790 
1791 /*
1792  * Copy the contents of uio into a properly sized mbuf chain.
1793  */
1794 struct mbuf *
1795 m_uiotombuf(struct uio *uio, int how, int len, int align, int flags)
1796 {
1797 	struct mbuf *m, *mb;
1798 	int error, length;
1799 	ssize_t total;
1800 	int progress = 0;
1801 
1802 	/*
1803 	 * len can be zero or an arbitrary large value bound by
1804 	 * the total data supplied by the uio.
1805 	 */
1806 	if (len > 0)
1807 		total = min(uio->uio_resid, len);
1808 	else
1809 		total = uio->uio_resid;
1810 
1811 	/*
1812 	 * The smallest unit returned by m_getm2() is a single mbuf
1813 	 * with pkthdr.  We can't align past it.
1814 	 */
1815 	if (align >= MHLEN)
1816 		return (NULL);
1817 
1818 	/*
1819 	 * Give us the full allocation or nothing.
1820 	 * If len is zero return the smallest empty mbuf.
1821 	 */
1822 	m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags);
1823 	if (m == NULL)
1824 		return (NULL);
1825 	m->m_data += align;
1826 
1827 	/* Fill all mbufs with uio data and update header information. */
1828 	for (mb = m; mb != NULL; mb = mb->m_next) {
1829 		length = min(M_TRAILINGSPACE(mb), total - progress);
1830 
1831 		error = uiomove(mtod(mb, void *), length, uio);
1832 		if (error) {
1833 			m_freem(m);
1834 			return (NULL);
1835 		}
1836 
1837 		mb->m_len = length;
1838 		progress += length;
1839 		if (flags & M_PKTHDR)
1840 			m->m_pkthdr.len += length;
1841 	}
1842 	KASSERT(progress == total, ("%s: progress != total", __func__));
1843 
1844 	return (m);
1845 }
1846 
1847 /*
1848  * Copy an mbuf chain into a uio limited by len if set.
1849  */
1850 int
1851 m_mbuftouio(struct uio *uio, struct mbuf *m, int len)
1852 {
1853 	int error, length, total;
1854 	int progress = 0;
1855 
1856 	if (len > 0)
1857 		total = min(uio->uio_resid, len);
1858 	else
1859 		total = uio->uio_resid;
1860 
1861 	/* Fill the uio with data from the mbufs. */
1862 	for (; m != NULL; m = m->m_next) {
1863 		length = min(m->m_len, total - progress);
1864 
1865 		error = uiomove(mtod(m, void *), length, uio);
1866 		if (error)
1867 			return (error);
1868 
1869 		progress += length;
1870 	}
1871 
1872 	return (0);
1873 }
1874 
1875 /*
1876  * Set the m_data pointer of a newly-allocated mbuf
1877  * to place an object of the specified size at the
1878  * end of the mbuf, longword aligned.
1879  */
1880 void
1881 m_align(struct mbuf *m, int len)
1882 {
1883 #ifdef INVARIANTS
1884 	const char *msg = "%s: not a virgin mbuf";
1885 #endif
1886 	int adjust;
1887 
1888 	if (m->m_flags & M_EXT) {
1889 		KASSERT(m->m_data == m->m_ext.ext_buf, (msg, __func__));
1890 		adjust = m->m_ext.ext_size - len;
1891 	} else if (m->m_flags & M_PKTHDR) {
1892 		KASSERT(m->m_data == m->m_pktdat, (msg, __func__));
1893 		adjust = MHLEN - len;
1894 	} else {
1895 		KASSERT(m->m_data == m->m_dat, (msg, __func__));
1896 		adjust = MLEN - len;
1897 	}
1898 
1899 	m->m_data += adjust &~ (sizeof(long)-1);
1900 }
1901 
1902 /*
1903  * Create a writable copy of the mbuf chain.  While doing this
1904  * we compact the chain with a goal of producing a chain with
1905  * at most two mbufs.  The second mbuf in this chain is likely
1906  * to be a cluster.  The primary purpose of this work is to create
1907  * a writable packet for encryption, compression, etc.  The
1908  * secondary goal is to linearize the data so the data can be
1909  * passed to crypto hardware in the most efficient manner possible.
1910  */
1911 struct mbuf *
1912 m_unshare(struct mbuf *m0, int how)
1913 {
1914 	struct mbuf *m, *mprev;
1915 	struct mbuf *n, *mfirst, *mlast;
1916 	int len, off;
1917 
1918 	mprev = NULL;
1919 	for (m = m0; m != NULL; m = mprev->m_next) {
1920 		/*
1921 		 * Regular mbufs are ignored unless there's a cluster
1922 		 * in front of it that we can use to coalesce.  We do
1923 		 * the latter mainly so later clusters can be coalesced
1924 		 * also w/o having to handle them specially (i.e. convert
1925 		 * mbuf+cluster -> cluster).  This optimization is heavily
1926 		 * influenced by the assumption that we're running over
1927 		 * Ethernet where MCLBYTES is large enough that the max
1928 		 * packet size will permit lots of coalescing into a
1929 		 * single cluster.  This in turn permits efficient
1930 		 * crypto operations, especially when using hardware.
1931 		 */
1932 		if ((m->m_flags & M_EXT) == 0) {
1933 			if (mprev && (mprev->m_flags & M_EXT) &&
1934 			    m->m_len <= M_TRAILINGSPACE(mprev)) {
1935 				/* XXX: this ignores mbuf types */
1936 				memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1937 				       mtod(m, caddr_t), m->m_len);
1938 				mprev->m_len += m->m_len;
1939 				mprev->m_next = m->m_next;	/* unlink from chain */
1940 				m_free(m);			/* reclaim mbuf */
1941 #if 0
1942 				newipsecstat.ips_mbcoalesced++;
1943 #endif
1944 			} else {
1945 				mprev = m;
1946 			}
1947 			continue;
1948 		}
1949 		/*
1950 		 * Writable mbufs are left alone (for now).
1951 		 */
1952 		if (M_WRITABLE(m)) {
1953 			mprev = m;
1954 			continue;
1955 		}
1956 
1957 		/*
1958 		 * Not writable, replace with a copy or coalesce with
1959 		 * the previous mbuf if possible (since we have to copy
1960 		 * it anyway, we try to reduce the number of mbufs and
1961 		 * clusters so that future work is easier).
1962 		 */
1963 		KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags));
1964 		/* NB: we only coalesce into a cluster or larger */
1965 		if (mprev != NULL && (mprev->m_flags & M_EXT) &&
1966 		    m->m_len <= M_TRAILINGSPACE(mprev)) {
1967 			/* XXX: this ignores mbuf types */
1968 			memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1969 			       mtod(m, caddr_t), m->m_len);
1970 			mprev->m_len += m->m_len;
1971 			mprev->m_next = m->m_next;	/* unlink from chain */
1972 			m_free(m);			/* reclaim mbuf */
1973 #if 0
1974 			newipsecstat.ips_clcoalesced++;
1975 #endif
1976 			continue;
1977 		}
1978 
1979 		/*
1980 		 * Allocate new space to hold the copy and copy the data.
1981 		 * We deal with jumbo mbufs (i.e. m_len > MCLBYTES) by
1982 		 * splitting them into clusters.  We could just malloc a
1983 		 * buffer and make it external but too many device drivers
1984 		 * don't know how to break up the non-contiguous memory when
1985 		 * doing DMA.
1986 		 */
1987 		n = m_getcl(how, m->m_type, m->m_flags);
1988 		if (n == NULL) {
1989 			m_freem(m0);
1990 			return (NULL);
1991 		}
1992 		len = m->m_len;
1993 		off = 0;
1994 		mfirst = n;
1995 		mlast = NULL;
1996 		for (;;) {
1997 			int cc = min(len, MCLBYTES);
1998 			memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc);
1999 			n->m_len = cc;
2000 			if (mlast != NULL)
2001 				mlast->m_next = n;
2002 			mlast = n;
2003 #if 0
2004 			newipsecstat.ips_clcopied++;
2005 #endif
2006 
2007 			len -= cc;
2008 			if (len <= 0)
2009 				break;
2010 			off += cc;
2011 
2012 			n = m_getcl(how, m->m_type, m->m_flags);
2013 			if (n == NULL) {
2014 				m_freem(mfirst);
2015 				m_freem(m0);
2016 				return (NULL);
2017 			}
2018 		}
2019 		n->m_next = m->m_next;
2020 		if (mprev == NULL)
2021 			m0 = mfirst;		/* new head of chain */
2022 		else
2023 			mprev->m_next = mfirst;	/* replace old mbuf */
2024 		m_free(m);			/* release old mbuf */
2025 		mprev = mfirst;
2026 	}
2027 	return (m0);
2028 }
2029 
2030 #ifdef MBUF_PROFILING
2031 
2032 #define MP_BUCKETS 32 /* don't just change this as things may overflow.*/
2033 struct mbufprofile {
2034 	uintmax_t wasted[MP_BUCKETS];
2035 	uintmax_t used[MP_BUCKETS];
2036 	uintmax_t segments[MP_BUCKETS];
2037 } mbprof;
2038 
2039 #define MP_MAXDIGITS 21	/* strlen("16,000,000,000,000,000,000") == 21 */
2040 #define MP_NUMLINES 6
2041 #define MP_NUMSPERLINE 16
2042 #define MP_EXTRABYTES 64	/* > strlen("used:\nwasted:\nsegments:\n") */
2043 /* work out max space needed and add a bit of spare space too */
2044 #define MP_MAXLINE ((MP_MAXDIGITS+1) * MP_NUMSPERLINE)
2045 #define MP_BUFSIZE ((MP_MAXLINE * MP_NUMLINES) + 1 + MP_EXTRABYTES)
2046 
2047 char mbprofbuf[MP_BUFSIZE];
2048 
2049 void
2050 m_profile(struct mbuf *m)
2051 {
2052 	int segments = 0;
2053 	int used = 0;
2054 	int wasted = 0;
2055 
2056 	while (m) {
2057 		segments++;
2058 		used += m->m_len;
2059 		if (m->m_flags & M_EXT) {
2060 			wasted += MHLEN - sizeof(m->m_ext) +
2061 			    m->m_ext.ext_size - m->m_len;
2062 		} else {
2063 			if (m->m_flags & M_PKTHDR)
2064 				wasted += MHLEN - m->m_len;
2065 			else
2066 				wasted += MLEN - m->m_len;
2067 		}
2068 		m = m->m_next;
2069 	}
2070 	/* be paranoid.. it helps */
2071 	if (segments > MP_BUCKETS - 1)
2072 		segments = MP_BUCKETS - 1;
2073 	if (used > 100000)
2074 		used = 100000;
2075 	if (wasted > 100000)
2076 		wasted = 100000;
2077 	/* store in the appropriate bucket */
2078 	/* don't bother locking. if it's slightly off, so what? */
2079 	mbprof.segments[segments]++;
2080 	mbprof.used[fls(used)]++;
2081 	mbprof.wasted[fls(wasted)]++;
2082 }
2083 
2084 static void
2085 mbprof_textify(void)
2086 {
2087 	int offset;
2088 	char *c;
2089 	uint64_t *p;
2090 
2091 
2092 	p = &mbprof.wasted[0];
2093 	c = mbprofbuf;
2094 	offset = snprintf(c, MP_MAXLINE + 10,
2095 	    "wasted:\n"
2096 	    "%ju %ju %ju %ju %ju %ju %ju %ju "
2097 	    "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2098 	    p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2099 	    p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2100 #ifdef BIG_ARRAY
2101 	p = &mbprof.wasted[16];
2102 	c += offset;
2103 	offset = snprintf(c, MP_MAXLINE,
2104 	    "%ju %ju %ju %ju %ju %ju %ju %ju "
2105 	    "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2106 	    p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2107 	    p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2108 #endif
2109 	p = &mbprof.used[0];
2110 	c += offset;
2111 	offset = snprintf(c, MP_MAXLINE + 10,
2112 	    "used:\n"
2113 	    "%ju %ju %ju %ju %ju %ju %ju %ju "
2114 	    "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2115 	    p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2116 	    p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2117 #ifdef BIG_ARRAY
2118 	p = &mbprof.used[16];
2119 	c += offset;
2120 	offset = snprintf(c, MP_MAXLINE,
2121 	    "%ju %ju %ju %ju %ju %ju %ju %ju "
2122 	    "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2123 	    p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2124 	    p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2125 #endif
2126 	p = &mbprof.segments[0];
2127 	c += offset;
2128 	offset = snprintf(c, MP_MAXLINE + 10,
2129 	    "segments:\n"
2130 	    "%ju %ju %ju %ju %ju %ju %ju %ju "
2131 	    "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2132 	    p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2133 	    p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2134 #ifdef BIG_ARRAY
2135 	p = &mbprof.segments[16];
2136 	c += offset;
2137 	offset = snprintf(c, MP_MAXLINE,
2138 	    "%ju %ju %ju %ju %ju %ju %ju %ju "
2139 	    "%ju %ju %ju %ju %ju %ju %ju %jju",
2140 	    p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2141 	    p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2142 #endif
2143 }
2144 
2145 static int
2146 mbprof_handler(SYSCTL_HANDLER_ARGS)
2147 {
2148 	int error;
2149 
2150 	mbprof_textify();
2151 	error = SYSCTL_OUT(req, mbprofbuf, strlen(mbprofbuf) + 1);
2152 	return (error);
2153 }
2154 
2155 static int
2156 mbprof_clr_handler(SYSCTL_HANDLER_ARGS)
2157 {
2158 	int clear, error;
2159 
2160 	clear = 0;
2161 	error = sysctl_handle_int(oidp, &clear, 0, req);
2162 	if (error || !req->newptr)
2163 		return (error);
2164 
2165 	if (clear) {
2166 		bzero(&mbprof, sizeof(mbprof));
2167 	}
2168 
2169 	return (error);
2170 }
2171 
2172 
2173 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofile, CTLTYPE_STRING|CTLFLAG_RD,
2174 	    NULL, 0, mbprof_handler, "A", "mbuf profiling statistics");
2175 
2176 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofileclr, CTLTYPE_INT|CTLFLAG_RW,
2177 	    NULL, 0, mbprof_clr_handler, "I", "clear mbuf profiling statistics");
2178 #endif
2179 
2180