xref: /netbsd-src/sys/kern/uipc_mbuf.c (revision 4472dbe5e3bd91ef2540bada7a7ca7384627ff9b)
1 /*	$NetBSD: uipc_mbuf.c,v 1.45 2000/03/01 12:49:28 itojun Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the NetBSD
22  *	Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
41  * Copyright (c) 1982, 1986, 1988, 1991, 1993
42  *	The Regents of the University of California.  All rights reserved.
43  *
44  * Redistribution and use in source and binary forms, with or without
45  * modification, are permitted provided that the following conditions
46  * are met:
47  * 1. Redistributions of source code must retain the above copyright
48  *    notice, this list of conditions and the following disclaimer.
49  * 2. Redistributions in binary form must reproduce the above copyright
50  *    notice, this list of conditions and the following disclaimer in the
51  *    documentation and/or other materials provided with the distribution.
52  * 3. All advertising materials mentioning features or use of this software
53  *    must display the following acknowledgement:
54  *	This product includes software developed by the University of
55  *	California, Berkeley and its contributors.
56  * 4. Neither the name of the University nor the names of its contributors
57  *    may be used to endorse or promote products derived from this software
58  *    without specific prior written permission.
59  *
60  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
61  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
62  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
63  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
64  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
65  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
66  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
67  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
68  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
69  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
70  * SUCH DAMAGE.
71  *
72  *	@(#)uipc_mbuf.c	8.4 (Berkeley) 2/14/95
73  */
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/proc.h>
78 #include <sys/malloc.h>
79 #include <sys/map.h>
80 #define MBTYPES
81 #include <sys/mbuf.h>
82 #include <sys/kernel.h>
83 #include <sys/syslog.h>
84 #include <sys/domain.h>
85 #include <sys/protosw.h>
86 #include <sys/pool.h>
87 #include <sys/socket.h>
88 #include <net/if.h>
89 
90 #include <vm/vm.h>
91 #include <vm/vm_kern.h>
92 
93 #include <uvm/uvm_extern.h>
94 
95 #include <sys/sysctl.h>
96 
97 struct	pool mbpool;		/* mbuf pool */
98 struct	pool mclpool;		/* mbuf cluster pool */
99 
100 struct mbstat mbstat;
101 int	max_linkhdr;
102 int	max_protohdr;
103 int	max_hdr;
104 int	max_datalen;
105 
106 void	*mclpool_alloc __P((unsigned long, int, int));
107 void	mclpool_release __P((void *, unsigned long, int));
108 static struct mbuf *m_copym0 __P((struct mbuf *, int, int, int, int));
109 
110 const char *mclpool_warnmsg =
111     "WARNING: mclpool limit reached; increase NMBCLUSTERS";
112 
113 /*
114  * Initialize the mbuf allcator.
115  */
116 void
117 mbinit()
118 {
119 
120 	pool_init(&mbpool, msize, 0, 0, 0, "mbpl", 0, NULL, NULL, 0);
121 	pool_init(&mclpool, mclbytes, 0, 0, 0, "mclpl", 0, mclpool_alloc,
122 	    mclpool_release, 0);
123 
124 	/*
125 	 * Set the hard limit on the mclpool to the number of
126 	 * mbuf clusters the kernel is to support.  Log the limit
127 	 * reached message max once a minute.
128 	 */
129 	pool_sethardlimit(&mclpool, nmbclusters, mclpool_warnmsg, 60);
130 
131 	/*
132 	 * Set a low water mark for both mbufs and clusters.  This should
133 	 * help ensure that they can be allocated in a memory starvation
134 	 * situation.  This is important for e.g. diskless systems which
135 	 * must allocate mbufs in order for the pagedaemon to clean pages.
136 	 */
137 	pool_setlowat(&mbpool, mblowat);
138 	pool_setlowat(&mclpool, mcllowat);
139 }
140 
141 int
142 sysctl_dombuf(name, namelen, oldp, oldlenp, newp, newlen)
143 	int *name;
144 	u_int namelen;
145 	void *oldp;
146 	size_t *oldlenp;
147 	void *newp;
148 	size_t newlen;
149 {
150 	int error, newval;
151 
152 	/* All sysctl names at this level are terminal. */
153 	if (namelen != 1)
154 		return (ENOTDIR);		/* overloaded */
155 
156 	switch (name[0]) {
157 	case MBUF_MSIZE:
158 		return (sysctl_rdint(oldp, oldlenp, newp, msize));
159 	case MBUF_MCLBYTES:
160 		return (sysctl_rdint(oldp, oldlenp, newp, mclbytes));
161 	case MBUF_NMBCLUSTERS:
162 		/*
163 		 * If we have direct-mapped pool pages, we can adjust this
164 		 * number on the fly.  If not, we're limited by the size
165 		 * of mb_map, and cannot change this value.
166 		 *
167 		 * Note: we only allow the value to be increased, never
168 		 * decreased.
169 		 */
170 		if (mb_map == NULL) {
171 			newval = nmbclusters;
172 			error = sysctl_int(oldp, oldlenp, newp, newlen,
173 			    &newval);
174 			if (error != 0)
175 				return (error);
176 			if (newp != NULL) {
177 				if (newval >= nmbclusters) {
178 					nmbclusters = newval;
179 					pool_sethardlimit(&mclpool,
180 					    nmbclusters, mclpool_warnmsg, 60);
181 				} else
182 					error = EINVAL;
183 			}
184 			return (error);
185 		} else
186 			return (sysctl_rdint(oldp, oldlenp, newp, nmbclusters));
187 	case MBUF_MBLOWAT:
188 	case MBUF_MCLLOWAT:
189 		/* New value must be >= 0. */
190 		newval = (name[0] == MBUF_MBLOWAT) ? mblowat : mcllowat;
191 		error = sysctl_int(oldp, oldlenp, newp, newlen, &newval);
192 		if (error != 0)
193 			return (error);
194 		if (newp != NULL) {
195 			if (newval >= 0) {
196 				if (name[0] == MBUF_MBLOWAT) {
197 					mblowat = newval;
198 					pool_setlowat(&mbpool, newval);
199 				} else {
200 					mcllowat = newval;
201 					pool_setlowat(&mclpool, newval);
202 				}
203 			} else
204 				error = EINVAL;
205 		}
206 		return (error);
207 	default:
208 		return (EOPNOTSUPP);
209 	}
210 	/* NOTREACHED */
211 }
212 
213 void *
214 mclpool_alloc(sz, flags, mtype)
215 	unsigned long sz;
216 	int flags;
217 	int mtype;
218 {
219 	boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
220 
221 	return ((void *)uvm_km_alloc_poolpage1(mb_map, uvmexp.mb_object,
222 	    waitok));
223 }
224 
225 void
226 mclpool_release(v, sz, mtype)
227 	void *v;
228 	unsigned long sz;
229 	int mtype;
230 {
231 
232 	uvm_km_free_poolpage1(mb_map, (vaddr_t)v);
233 }
234 
235 /*
236  * When MGET failes, ask protocols to free space when short of memory,
237  * then re-attempt to allocate an mbuf.
238  */
239 struct mbuf *
240 m_retry(i, t)
241 	int i, t;
242 {
243 	struct mbuf *m;
244 
245 	m_reclaim(i);
246 #define m_retry(i, t)	(struct mbuf *)0
247 	MGET(m, i, t);
248 #undef m_retry
249 	if (m != NULL)
250 		mbstat.m_wait++;
251 	else
252 		mbstat.m_drops++;
253 	return (m);
254 }
255 
256 /*
257  * As above; retry an MGETHDR.
258  */
259 struct mbuf *
260 m_retryhdr(i, t)
261 	int i, t;
262 {
263 	struct mbuf *m;
264 
265 	m_reclaim(i);
266 #define m_retryhdr(i, t) (struct mbuf *)0
267 	MGETHDR(m, i, t);
268 #undef m_retryhdr
269 	if (m != NULL)
270 		mbstat.m_wait++;
271 	else
272 		mbstat.m_drops++;
273 	return (m);
274 }
275 
276 void
277 m_reclaim(how)
278 	int how;
279 {
280 	struct domain *dp;
281 	struct protosw *pr;
282 	struct ifnet *ifp;
283 	int s = splimp();
284 
285 	for (dp = domains; dp; dp = dp->dom_next)
286 		for (pr = dp->dom_protosw;
287 		     pr < dp->dom_protoswNPROTOSW; pr++)
288 			if (pr->pr_drain)
289 				(*pr->pr_drain)();
290 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list))
291 		if (ifp->if_drain)
292 			(*ifp->if_drain)(ifp);
293 	splx(s);
294 	mbstat.m_drain++;
295 }
296 
297 /*
298  * Space allocation routines.
299  * These are also available as macros
300  * for critical paths.
301  */
302 struct mbuf *
303 m_get(nowait, type)
304 	int nowait, type;
305 {
306 	struct mbuf *m;
307 
308 	MGET(m, nowait, type);
309 	return (m);
310 }
311 
312 struct mbuf *
313 m_gethdr(nowait, type)
314 	int nowait, type;
315 {
316 	struct mbuf *m;
317 
318 	MGETHDR(m, nowait, type);
319 	return (m);
320 }
321 
322 struct mbuf *
323 m_getclr(nowait, type)
324 	int nowait, type;
325 {
326 	struct mbuf *m;
327 
328 	MGET(m, nowait, type);
329 	if (m == 0)
330 		return (0);
331 	memset(mtod(m, caddr_t), 0, MLEN);
332 	return (m);
333 }
334 
335 struct mbuf *
336 m_free(m)
337 	struct mbuf *m;
338 {
339 	struct mbuf *n;
340 
341 	MFREE(m, n);
342 	return (n);
343 }
344 
345 void
346 m_freem(m)
347 	struct mbuf *m;
348 {
349 	struct mbuf *n;
350 
351 	if (m == NULL)
352 		return;
353 	if ((m->m_flags & M_PKTHDR) != 0 && m->m_pkthdr.aux) {
354 		m_freem(m->m_pkthdr.aux);
355 		m->m_pkthdr.aux = NULL;
356 	}
357 	do {
358 		MFREE(m, n);
359 		m = n;
360 	} while (m);
361 }
362 
363 /*
364  * Mbuffer utility routines.
365  */
366 
367 /*
368  * Lesser-used path for M_PREPEND:
369  * allocate new mbuf to prepend to chain,
370  * copy junk along.
371  */
372 struct mbuf *
373 m_prepend(m, len, how)
374 	struct mbuf *m;
375 	int len, how;
376 {
377 	struct mbuf *mn;
378 
379 	MGET(mn, how, m->m_type);
380 	if (mn == (struct mbuf *)NULL) {
381 		m_freem(m);
382 		return ((struct mbuf *)NULL);
383 	}
384 	if (m->m_flags & M_PKTHDR) {
385 		M_COPY_PKTHDR(mn, m);
386 		m->m_flags &= ~M_PKTHDR;
387 	}
388 	mn->m_next = m;
389 	m = mn;
390 	if (len < MHLEN)
391 		MH_ALIGN(m, len);
392 	m->m_len = len;
393 	return (m);
394 }
395 
396 /*
397  * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
398  * continuing for "len" bytes.  If len is M_COPYALL, copy to end of mbuf.
399  * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
400  */
401 int MCFail;
402 
403 struct mbuf *
404 m_copym(m, off0, len, wait)
405 	struct mbuf *m;
406 	int off0, wait;
407 	int len;
408 {
409 	return m_copym0(m, off0, len, wait, 0);	/* shallow copy on M_EXT */
410 }
411 
412 struct mbuf *
413 m_dup(m, off0, len, wait)
414 	struct mbuf *m;
415 	int off0, wait;
416 	int len;
417 {
418 	return m_copym0(m, off0, len, wait, 1);	/* deep copy */
419 }
420 
421 static struct mbuf *
422 m_copym0(m, off0, len, wait, deep)
423 	struct mbuf *m;
424 	int off0, wait;
425 	int len;
426 	int deep;	/* deep copy */
427 {
428 	struct mbuf *n, **np;
429 	int off = off0;
430 	struct mbuf *top;
431 	int copyhdr = 0;
432 
433 	if (off < 0 || len < 0)
434 		panic("m_copym: off %d, len %d", off, len);
435 	if (off == 0 && m->m_flags & M_PKTHDR)
436 		copyhdr = 1;
437 	while (off > 0) {
438 		if (m == 0)
439 			panic("m_copym: m == 0");
440 		if (off < m->m_len)
441 			break;
442 		off -= m->m_len;
443 		m = m->m_next;
444 	}
445 	np = &top;
446 	top = 0;
447 	while (len > 0) {
448 		if (m == 0) {
449 			if (len != M_COPYALL)
450 				panic("m_copym: m == 0 and not COPYALL");
451 			break;
452 		}
453 		MGET(n, wait, m->m_type);
454 		*np = n;
455 		if (n == 0)
456 			goto nospace;
457 		if (copyhdr) {
458 			M_COPY_PKTHDR(n, m);
459 			if (len == M_COPYALL)
460 				n->m_pkthdr.len -= off0;
461 			else
462 				n->m_pkthdr.len = len;
463 			copyhdr = 0;
464 		}
465 		n->m_len = min(len, m->m_len - off);
466 		if (m->m_flags & M_EXT) {
467 			if (!deep) {
468 				n->m_data = m->m_data + off;
469 				n->m_ext = m->m_ext;
470 				MCLADDREFERENCE(m, n);
471 			} else {
472 				MCLGET(n, wait);
473 				memcpy(mtod(n, caddr_t), mtod(m, caddr_t)+off,
474 				    (unsigned)n->m_len);
475 			}
476 		} else
477 			memcpy(mtod(n, caddr_t), mtod(m, caddr_t)+off,
478 			    (unsigned)n->m_len);
479 		if (len != M_COPYALL)
480 			len -= n->m_len;
481 		off = 0;
482 		m = m->m_next;
483 		np = &n->m_next;
484 	}
485 	if (top == 0)
486 		MCFail++;
487 	return (top);
488 nospace:
489 	m_freem(top);
490 	MCFail++;
491 	return (0);
492 }
493 
494 /*
495  * Copy an entire packet, including header (which must be present).
496  * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
497  */
498 struct mbuf *
499 m_copypacket(m, how)
500 	struct mbuf *m;
501 	int how;
502 {
503 	struct mbuf *top, *n, *o;
504 
505 	MGET(n, how, m->m_type);
506 	top = n;
507 	if (!n)
508 		goto nospace;
509 
510 	M_COPY_PKTHDR(n, m);
511 	n->m_len = m->m_len;
512 	if (m->m_flags & M_EXT) {
513 		n->m_data = m->m_data;
514 		n->m_ext = m->m_ext;
515 		MCLADDREFERENCE(m, n);
516 	} else {
517 		memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
518 	}
519 
520 	m = m->m_next;
521 	while (m) {
522 		MGET(o, how, m->m_type);
523 		if (!o)
524 			goto nospace;
525 
526 		n->m_next = o;
527 		n = n->m_next;
528 
529 		n->m_len = m->m_len;
530 		if (m->m_flags & M_EXT) {
531 			n->m_data = m->m_data;
532 			n->m_ext = m->m_ext;
533 			MCLADDREFERENCE(m, n);
534 		} else {
535 			memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
536 		}
537 
538 		m = m->m_next;
539 	}
540 	return top;
541 nospace:
542 	m_freem(top);
543 	MCFail++;
544 	return 0;
545 }
546 
547 /*
548  * Copy data from an mbuf chain starting "off" bytes from the beginning,
549  * continuing for "len" bytes, into the indicated buffer.
550  */
551 void
552 m_copydata(m, off, len, cp)
553 	struct mbuf *m;
554 	int off;
555 	int len;
556 	caddr_t cp;
557 {
558 	unsigned count;
559 
560 	if (off < 0 || len < 0)
561 		panic("m_copydata");
562 	while (off > 0) {
563 		if (m == 0)
564 			panic("m_copydata");
565 		if (off < m->m_len)
566 			break;
567 		off -= m->m_len;
568 		m = m->m_next;
569 	}
570 	while (len > 0) {
571 		if (m == 0)
572 			panic("m_copydata");
573 		count = min(m->m_len - off, len);
574 		memcpy(cp, mtod(m, caddr_t) + off, count);
575 		len -= count;
576 		cp += count;
577 		off = 0;
578 		m = m->m_next;
579 	}
580 }
581 
582 /*
583  * Concatenate mbuf chain n to m.
584  * Both chains must be of the same type (e.g. MT_DATA).
585  * Any m_pkthdr is not updated.
586  */
587 void
588 m_cat(m, n)
589 	struct mbuf *m, *n;
590 {
591 	while (m->m_next)
592 		m = m->m_next;
593 	while (n) {
594 		if (m->m_flags & M_EXT ||
595 		    m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
596 			/* just join the two chains */
597 			m->m_next = n;
598 			return;
599 		}
600 		/* splat the data from one into the other */
601 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
602 		    (u_int)n->m_len);
603 		m->m_len += n->m_len;
604 		n = m_free(n);
605 	}
606 }
607 
608 void
609 m_adj(mp, req_len)
610 	struct mbuf *mp;
611 	int req_len;
612 {
613 	int len = req_len;
614 	struct mbuf *m;
615 	int count;
616 
617 	if ((m = mp) == NULL)
618 		return;
619 	if (len >= 0) {
620 		/*
621 		 * Trim from head.
622 		 */
623 		while (m != NULL && len > 0) {
624 			if (m->m_len <= len) {
625 				len -= m->m_len;
626 				m->m_len = 0;
627 				m = m->m_next;
628 			} else {
629 				m->m_len -= len;
630 				m->m_data += len;
631 				len = 0;
632 			}
633 		}
634 		m = mp;
635 		if (mp->m_flags & M_PKTHDR)
636 			m->m_pkthdr.len -= (req_len - len);
637 	} else {
638 		/*
639 		 * Trim from tail.  Scan the mbuf chain,
640 		 * calculating its length and finding the last mbuf.
641 		 * If the adjustment only affects this mbuf, then just
642 		 * adjust and return.  Otherwise, rescan and truncate
643 		 * after the remaining size.
644 		 */
645 		len = -len;
646 		count = 0;
647 		for (;;) {
648 			count += m->m_len;
649 			if (m->m_next == (struct mbuf *)0)
650 				break;
651 			m = m->m_next;
652 		}
653 		if (m->m_len >= len) {
654 			m->m_len -= len;
655 			if (mp->m_flags & M_PKTHDR)
656 				mp->m_pkthdr.len -= len;
657 			return;
658 		}
659 		count -= len;
660 		if (count < 0)
661 			count = 0;
662 		/*
663 		 * Correct length for chain is "count".
664 		 * Find the mbuf with last data, adjust its length,
665 		 * and toss data from remaining mbufs on chain.
666 		 */
667 		m = mp;
668 		if (m->m_flags & M_PKTHDR)
669 			m->m_pkthdr.len = count;
670 		for (; m; m = m->m_next) {
671 			if (m->m_len >= count) {
672 				m->m_len = count;
673 				break;
674 			}
675 			count -= m->m_len;
676 		}
677 		while (m->m_next)
678 			(m = m->m_next) ->m_len = 0;
679 	}
680 }
681 
682 /*
683  * Rearange an mbuf chain so that len bytes are contiguous
684  * and in the data area of an mbuf (so that mtod and dtom
685  * will work for a structure of size len).  Returns the resulting
686  * mbuf chain on success, frees it and returns null on failure.
687  * If there is room, it will add up to max_protohdr-len extra bytes to the
688  * contiguous region in an attempt to avoid being called next time.
689  */
690 int MPFail;
691 
692 struct mbuf *
693 m_pullup(n, len)
694 	struct mbuf *n;
695 	int len;
696 {
697 	struct mbuf *m;
698 	int count;
699 	int space;
700 
701 	/*
702 	 * If first mbuf has no cluster, and has room for len bytes
703 	 * without shifting current data, pullup into it,
704 	 * otherwise allocate a new mbuf to prepend to the chain.
705 	 */
706 	if ((n->m_flags & M_EXT) == 0 &&
707 	    n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
708 		if (n->m_len >= len)
709 			return (n);
710 		m = n;
711 		n = n->m_next;
712 		len -= m->m_len;
713 	} else {
714 		if (len > MHLEN)
715 			goto bad;
716 		MGET(m, M_DONTWAIT, n->m_type);
717 		if (m == 0)
718 			goto bad;
719 		m->m_len = 0;
720 		if (n->m_flags & M_PKTHDR) {
721 			M_COPY_PKTHDR(m, n);
722 			n->m_flags &= ~M_PKTHDR;
723 		}
724 	}
725 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
726 	do {
727 		count = min(min(max(len, max_protohdr), space), n->m_len);
728 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
729 		  (unsigned)count);
730 		len -= count;
731 		m->m_len += count;
732 		n->m_len -= count;
733 		space -= count;
734 		if (n->m_len)
735 			n->m_data += count;
736 		else
737 			n = m_free(n);
738 	} while (len > 0 && n);
739 	if (len > 0) {
740 		(void) m_free(m);
741 		goto bad;
742 	}
743 	m->m_next = n;
744 	return (m);
745 bad:
746 	m_freem(n);
747 	MPFail++;
748 	return (0);
749 }
750 
751 /*
752  * Partition an mbuf chain in two pieces, returning the tail --
753  * all but the first len0 bytes.  In case of failure, it returns NULL and
754  * attempts to restore the chain to its original state.
755  */
756 struct mbuf *
757 m_split(m0, len0, wait)
758 	struct mbuf *m0;
759 	int len0, wait;
760 {
761 	struct mbuf *m, *n;
762 	unsigned len = len0, remain, len_save;
763 
764 	for (m = m0; m && len > m->m_len; m = m->m_next)
765 		len -= m->m_len;
766 	if (m == 0)
767 		return (0);
768 	remain = m->m_len - len;
769 	if (m0->m_flags & M_PKTHDR) {
770 		MGETHDR(n, wait, m0->m_type);
771 		if (n == 0)
772 			return (0);
773 		n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
774 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
775 		len_save = m0->m_pkthdr.len;
776 		m0->m_pkthdr.len = len0;
777 		if (m->m_flags & M_EXT)
778 			goto extpacket;
779 		if (remain > MHLEN) {
780 			/* m can't be the lead packet */
781 			MH_ALIGN(n, 0);
782 			n->m_next = m_split(m, len, wait);
783 			if (n->m_next == 0) {
784 				(void) m_free(n);
785 				m0->m_pkthdr.len = len_save;
786 				return (0);
787 			} else
788 				return (n);
789 		} else
790 			MH_ALIGN(n, remain);
791 	} else if (remain == 0) {
792 		n = m->m_next;
793 		m->m_next = 0;
794 		return (n);
795 	} else {
796 		MGET(n, wait, m->m_type);
797 		if (n == 0)
798 			return (0);
799 		M_ALIGN(n, remain);
800 	}
801 extpacket:
802 	if (m->m_flags & M_EXT) {
803 		n->m_ext = m->m_ext;
804 		MCLADDREFERENCE(m, n);
805 		n->m_data = m->m_data + len;
806 	} else {
807 		memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + len, remain);
808 	}
809 	n->m_len = remain;
810 	m->m_len = len;
811 	n->m_next = m->m_next;
812 	m->m_next = 0;
813 	return (n);
814 }
815 /*
816  * Routine to copy from device local memory into mbufs.
817  */
818 struct mbuf *
819 m_devget(buf, totlen, off0, ifp, copy)
820 	char *buf;
821 	int totlen, off0;
822 	struct ifnet *ifp;
823 	void (*copy) __P((const void *from, void *to, size_t len));
824 {
825 	struct mbuf *m;
826 	struct mbuf *top = 0, **mp = &top;
827 	int off = off0, len;
828 	char *cp;
829 	char *epkt;
830 
831 	cp = buf;
832 	epkt = cp + totlen;
833 	if (off) {
834 		/*
835 		 * If 'off' is non-zero, packet is trailer-encapsulated,
836 		 * so we have to skip the type and length fields.
837 		 */
838 		cp += off + 2 * sizeof(u_int16_t);
839 		totlen -= 2 * sizeof(u_int16_t);
840 	}
841 	MGETHDR(m, M_DONTWAIT, MT_DATA);
842 	if (m == 0)
843 		return (0);
844 	m->m_pkthdr.rcvif = ifp;
845 	m->m_pkthdr.len = totlen;
846 	m->m_len = MHLEN;
847 
848 	while (totlen > 0) {
849 		if (top) {
850 			MGET(m, M_DONTWAIT, MT_DATA);
851 			if (m == 0) {
852 				m_freem(top);
853 				return (0);
854 			}
855 			m->m_len = MLEN;
856 		}
857 		len = min(totlen, epkt - cp);
858 		if (len >= MINCLSIZE) {
859 			MCLGET(m, M_DONTWAIT);
860 			if ((m->m_flags & M_EXT) == 0) {
861 				m_free(m);
862 				m_freem(top);
863 				return (0);
864 			}
865 			m->m_len = len = min(len, MCLBYTES);
866 		} else {
867 			/*
868 			 * Place initial small packet/header at end of mbuf.
869 			 */
870 			if (len < m->m_len) {
871 				if (top == 0 && len + max_linkhdr <= m->m_len)
872 					m->m_data += max_linkhdr;
873 				m->m_len = len;
874 			} else
875 				len = m->m_len;
876 		}
877 		if (copy)
878 			copy(cp, mtod(m, caddr_t), (size_t)len);
879 		else
880 			memcpy(mtod(m, caddr_t), cp, (size_t)len);
881 		cp += len;
882 		*mp = m;
883 		mp = &m->m_next;
884 		totlen -= len;
885 		if (cp == epkt)
886 			cp = buf;
887 	}
888 	return (top);
889 }
890 
891 /*
892  * Copy data from a buffer back into the indicated mbuf chain,
893  * starting "off" bytes from the beginning, extending the mbuf
894  * chain if necessary.
895  */
896 void
897 m_copyback(m0, off, len, cp)
898 	struct	mbuf *m0;
899 	int off;
900 	int len;
901 	caddr_t cp;
902 {
903 	int mlen;
904 	struct mbuf *m = m0, *n;
905 	int totlen = 0;
906 
907 	if (m0 == 0)
908 		return;
909 	while (off > (mlen = m->m_len)) {
910 		off -= mlen;
911 		totlen += mlen;
912 		if (m->m_next == 0) {
913 			n = m_getclr(M_DONTWAIT, m->m_type);
914 			if (n == 0)
915 				goto out;
916 			n->m_len = min(MLEN, len + off);
917 			m->m_next = n;
918 		}
919 		m = m->m_next;
920 	}
921 	while (len > 0) {
922 		mlen = min (m->m_len - off, len);
923 		memcpy(mtod(m, caddr_t) + off, cp, (unsigned)mlen);
924 		cp += mlen;
925 		len -= mlen;
926 		mlen += off;
927 		off = 0;
928 		totlen += mlen;
929 		if (len == 0)
930 			break;
931 		if (m->m_next == 0) {
932 			n = m_get(M_DONTWAIT, m->m_type);
933 			if (n == 0)
934 				break;
935 			n->m_len = min(MLEN, len);
936 			m->m_next = n;
937 		}
938 		m = m->m_next;
939 	}
940 out:	if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
941 		m->m_pkthdr.len = totlen;
942 }
943