xref: /netbsd-src/sys/kern/uipc_mbuf.c (revision 481fca6e59249d8ffcf24fef7cfbe7b131bfb080)
1 /*	$NetBSD: uipc_mbuf.c,v 1.47 2000/06/27 17:41:44 mrg 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 <uvm/uvm_extern.h>
91 
92 #include <sys/sysctl.h>
93 
94 struct	pool mbpool;		/* mbuf pool */
95 struct	pool mclpool;		/* mbuf cluster pool */
96 
97 struct mbstat mbstat;
98 int	max_linkhdr;
99 int	max_protohdr;
100 int	max_hdr;
101 int	max_datalen;
102 
103 void	*mclpool_alloc __P((unsigned long, int, int));
104 void	mclpool_release __P((void *, unsigned long, int));
105 static struct mbuf *m_copym0 __P((struct mbuf *, int, int, int, int));
106 
107 const char *mclpool_warnmsg =
108     "WARNING: mclpool limit reached; increase NMBCLUSTERS";
109 
110 /*
111  * Initialize the mbuf allcator.
112  */
113 void
114 mbinit()
115 {
116 
117 	pool_init(&mbpool, msize, 0, 0, 0, "mbpl", 0, NULL, NULL, 0);
118 	pool_init(&mclpool, mclbytes, 0, 0, 0, "mclpl", 0, mclpool_alloc,
119 	    mclpool_release, 0);
120 
121 	/*
122 	 * Set the hard limit on the mclpool to the number of
123 	 * mbuf clusters the kernel is to support.  Log the limit
124 	 * reached message max once a minute.
125 	 */
126 	pool_sethardlimit(&mclpool, nmbclusters, mclpool_warnmsg, 60);
127 
128 	/*
129 	 * Set a low water mark for both mbufs and clusters.  This should
130 	 * help ensure that they can be allocated in a memory starvation
131 	 * situation.  This is important for e.g. diskless systems which
132 	 * must allocate mbufs in order for the pagedaemon to clean pages.
133 	 */
134 	pool_setlowat(&mbpool, mblowat);
135 	pool_setlowat(&mclpool, mcllowat);
136 }
137 
138 int
139 sysctl_dombuf(name, namelen, oldp, oldlenp, newp, newlen)
140 	int *name;
141 	u_int namelen;
142 	void *oldp;
143 	size_t *oldlenp;
144 	void *newp;
145 	size_t newlen;
146 {
147 	int error, newval;
148 
149 	/* All sysctl names at this level are terminal. */
150 	if (namelen != 1)
151 		return (ENOTDIR);		/* overloaded */
152 
153 	switch (name[0]) {
154 	case MBUF_MSIZE:
155 		return (sysctl_rdint(oldp, oldlenp, newp, msize));
156 	case MBUF_MCLBYTES:
157 		return (sysctl_rdint(oldp, oldlenp, newp, mclbytes));
158 	case MBUF_NMBCLUSTERS:
159 		/*
160 		 * If we have direct-mapped pool pages, we can adjust this
161 		 * number on the fly.  If not, we're limited by the size
162 		 * of mb_map, and cannot change this value.
163 		 *
164 		 * Note: we only allow the value to be increased, never
165 		 * decreased.
166 		 */
167 		if (mb_map == NULL) {
168 			newval = nmbclusters;
169 			error = sysctl_int(oldp, oldlenp, newp, newlen,
170 			    &newval);
171 			if (error != 0)
172 				return (error);
173 			if (newp != NULL) {
174 				if (newval >= nmbclusters) {
175 					nmbclusters = newval;
176 					pool_sethardlimit(&mclpool,
177 					    nmbclusters, mclpool_warnmsg, 60);
178 				} else
179 					error = EINVAL;
180 			}
181 			return (error);
182 		} else
183 			return (sysctl_rdint(oldp, oldlenp, newp, nmbclusters));
184 	case MBUF_MBLOWAT:
185 	case MBUF_MCLLOWAT:
186 		/* New value must be >= 0. */
187 		newval = (name[0] == MBUF_MBLOWAT) ? mblowat : mcllowat;
188 		error = sysctl_int(oldp, oldlenp, newp, newlen, &newval);
189 		if (error != 0)
190 			return (error);
191 		if (newp != NULL) {
192 			if (newval >= 0) {
193 				if (name[0] == MBUF_MBLOWAT) {
194 					mblowat = newval;
195 					pool_setlowat(&mbpool, newval);
196 				} else {
197 					mcllowat = newval;
198 					pool_setlowat(&mclpool, newval);
199 				}
200 			} else
201 				error = EINVAL;
202 		}
203 		return (error);
204 	default:
205 		return (EOPNOTSUPP);
206 	}
207 	/* NOTREACHED */
208 }
209 
210 void *
211 mclpool_alloc(sz, flags, mtype)
212 	unsigned long sz;
213 	int flags;
214 	int mtype;
215 {
216 	boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
217 
218 	return ((void *)uvm_km_alloc_poolpage1(mb_map, uvmexp.mb_object,
219 	    waitok));
220 }
221 
222 void
223 mclpool_release(v, sz, mtype)
224 	void *v;
225 	unsigned long sz;
226 	int mtype;
227 {
228 
229 	uvm_km_free_poolpage1(mb_map, (vaddr_t)v);
230 }
231 
232 /*
233  * When MGET failes, ask protocols to free space when short of memory,
234  * then re-attempt to allocate an mbuf.
235  */
236 struct mbuf *
237 m_retry(i, t)
238 	int i, t;
239 {
240 	struct mbuf *m;
241 
242 	m_reclaim(i);
243 #define m_retry(i, t)	(struct mbuf *)0
244 	MGET(m, i, t);
245 #undef m_retry
246 	if (m != NULL)
247 		mbstat.m_wait++;
248 	else
249 		mbstat.m_drops++;
250 	return (m);
251 }
252 
253 /*
254  * As above; retry an MGETHDR.
255  */
256 struct mbuf *
257 m_retryhdr(i, t)
258 	int i, t;
259 {
260 	struct mbuf *m;
261 
262 	m_reclaim(i);
263 #define m_retryhdr(i, t) (struct mbuf *)0
264 	MGETHDR(m, i, t);
265 #undef m_retryhdr
266 	if (m != NULL)
267 		mbstat.m_wait++;
268 	else
269 		mbstat.m_drops++;
270 	return (m);
271 }
272 
273 void
274 m_reclaim(how)
275 	int how;
276 {
277 	struct domain *dp;
278 	struct protosw *pr;
279 	struct ifnet *ifp;
280 	int s = splimp();
281 
282 	for (dp = domains; dp; dp = dp->dom_next)
283 		for (pr = dp->dom_protosw;
284 		     pr < dp->dom_protoswNPROTOSW; pr++)
285 			if (pr->pr_drain)
286 				(*pr->pr_drain)();
287 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list))
288 		if (ifp->if_drain)
289 			(*ifp->if_drain)(ifp);
290 	splx(s);
291 	mbstat.m_drain++;
292 }
293 
294 /*
295  * Space allocation routines.
296  * These are also available as macros
297  * for critical paths.
298  */
299 struct mbuf *
300 m_get(nowait, type)
301 	int nowait, type;
302 {
303 	struct mbuf *m;
304 
305 	MGET(m, nowait, type);
306 	return (m);
307 }
308 
309 struct mbuf *
310 m_gethdr(nowait, type)
311 	int nowait, type;
312 {
313 	struct mbuf *m;
314 
315 	MGETHDR(m, nowait, type);
316 	return (m);
317 }
318 
319 struct mbuf *
320 m_getclr(nowait, type)
321 	int nowait, type;
322 {
323 	struct mbuf *m;
324 
325 	MGET(m, nowait, type);
326 	if (m == 0)
327 		return (0);
328 	memset(mtod(m, caddr_t), 0, MLEN);
329 	return (m);
330 }
331 
332 struct mbuf *
333 m_free(m)
334 	struct mbuf *m;
335 {
336 	struct mbuf *n;
337 
338 	MFREE(m, n);
339 	return (n);
340 }
341 
342 void
343 m_freem(m)
344 	struct mbuf *m;
345 {
346 	struct mbuf *n;
347 
348 	if (m == NULL)
349 		return;
350 	if ((m->m_flags & M_PKTHDR) != 0 && m->m_pkthdr.aux) {
351 		m_freem(m->m_pkthdr.aux);
352 		m->m_pkthdr.aux = NULL;
353 	}
354 	do {
355 		MFREE(m, n);
356 		m = n;
357 	} while (m);
358 }
359 
360 /*
361  * Mbuffer utility routines.
362  */
363 
364 /*
365  * Lesser-used path for M_PREPEND:
366  * allocate new mbuf to prepend to chain,
367  * copy junk along.
368  */
369 struct mbuf *
370 m_prepend(m, len, how)
371 	struct mbuf *m;
372 	int len, how;
373 {
374 	struct mbuf *mn;
375 
376 	MGET(mn, how, m->m_type);
377 	if (mn == (struct mbuf *)NULL) {
378 		m_freem(m);
379 		return ((struct mbuf *)NULL);
380 	}
381 	if (m->m_flags & M_PKTHDR) {
382 		M_COPY_PKTHDR(mn, m);
383 		m->m_flags &= ~M_PKTHDR;
384 	}
385 	mn->m_next = m;
386 	m = mn;
387 	if (len < MHLEN)
388 		MH_ALIGN(m, len);
389 	m->m_len = len;
390 	return (m);
391 }
392 
393 /*
394  * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
395  * continuing for "len" bytes.  If len is M_COPYALL, copy to end of mbuf.
396  * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
397  */
398 int MCFail;
399 
400 struct mbuf *
401 m_copym(m, off0, len, wait)
402 	struct mbuf *m;
403 	int off0, wait;
404 	int len;
405 {
406 	return m_copym0(m, off0, len, wait, 0);	/* shallow copy on M_EXT */
407 }
408 
409 struct mbuf *
410 m_dup(m, off0, len, wait)
411 	struct mbuf *m;
412 	int off0, wait;
413 	int len;
414 {
415 	return m_copym0(m, off0, len, wait, 1);	/* deep copy */
416 }
417 
418 static struct mbuf *
419 m_copym0(m, off0, len, wait, deep)
420 	struct mbuf *m;
421 	int off0, wait;
422 	int len;
423 	int deep;	/* deep copy */
424 {
425 	struct mbuf *n, **np;
426 	int off = off0;
427 	struct mbuf *top;
428 	int copyhdr = 0;
429 
430 	if (off < 0 || len < 0)
431 		panic("m_copym: off %d, len %d", off, len);
432 	if (off == 0 && m->m_flags & M_PKTHDR)
433 		copyhdr = 1;
434 	while (off > 0) {
435 		if (m == 0)
436 			panic("m_copym: m == 0");
437 		if (off < m->m_len)
438 			break;
439 		off -= m->m_len;
440 		m = m->m_next;
441 	}
442 	np = &top;
443 	top = 0;
444 	while (len > 0) {
445 		if (m == 0) {
446 			if (len != M_COPYALL)
447 				panic("m_copym: m == 0 and not COPYALL");
448 			break;
449 		}
450 		MGET(n, wait, m->m_type);
451 		*np = n;
452 		if (n == 0)
453 			goto nospace;
454 		if (copyhdr) {
455 			M_COPY_PKTHDR(n, m);
456 			if (len == M_COPYALL)
457 				n->m_pkthdr.len -= off0;
458 			else
459 				n->m_pkthdr.len = len;
460 			copyhdr = 0;
461 		}
462 		n->m_len = min(len, m->m_len - off);
463 		if (m->m_flags & M_EXT) {
464 			if (!deep) {
465 				n->m_data = m->m_data + off;
466 				n->m_ext = m->m_ext;
467 				MCLADDREFERENCE(m, n);
468 			} else {
469 				MCLGET(n, wait);
470 				memcpy(mtod(n, caddr_t), mtod(m, caddr_t)+off,
471 				    (unsigned)n->m_len);
472 			}
473 		} else
474 			memcpy(mtod(n, caddr_t), mtod(m, caddr_t)+off,
475 			    (unsigned)n->m_len);
476 		if (len != M_COPYALL)
477 			len -= n->m_len;
478 		off = 0;
479 		m = m->m_next;
480 		np = &n->m_next;
481 	}
482 	if (top == 0)
483 		MCFail++;
484 	return (top);
485 nospace:
486 	m_freem(top);
487 	MCFail++;
488 	return (0);
489 }
490 
491 /*
492  * Copy an entire packet, including header (which must be present).
493  * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
494  */
495 struct mbuf *
496 m_copypacket(m, how)
497 	struct mbuf *m;
498 	int how;
499 {
500 	struct mbuf *top, *n, *o;
501 
502 	MGET(n, how, m->m_type);
503 	top = n;
504 	if (!n)
505 		goto nospace;
506 
507 	M_COPY_PKTHDR(n, m);
508 	n->m_len = m->m_len;
509 	if (m->m_flags & M_EXT) {
510 		n->m_data = m->m_data;
511 		n->m_ext = m->m_ext;
512 		MCLADDREFERENCE(m, n);
513 	} else {
514 		memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
515 	}
516 
517 	m = m->m_next;
518 	while (m) {
519 		MGET(o, how, m->m_type);
520 		if (!o)
521 			goto nospace;
522 
523 		n->m_next = o;
524 		n = n->m_next;
525 
526 		n->m_len = m->m_len;
527 		if (m->m_flags & M_EXT) {
528 			n->m_data = m->m_data;
529 			n->m_ext = m->m_ext;
530 			MCLADDREFERENCE(m, n);
531 		} else {
532 			memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
533 		}
534 
535 		m = m->m_next;
536 	}
537 	return top;
538 nospace:
539 	m_freem(top);
540 	MCFail++;
541 	return 0;
542 }
543 
544 /*
545  * Copy data from an mbuf chain starting "off" bytes from the beginning,
546  * continuing for "len" bytes, into the indicated buffer.
547  */
548 void
549 m_copydata(m, off, len, cp)
550 	struct mbuf *m;
551 	int off;
552 	int len;
553 	caddr_t cp;
554 {
555 	unsigned count;
556 
557 	if (off < 0 || len < 0)
558 		panic("m_copydata");
559 	while (off > 0) {
560 		if (m == 0)
561 			panic("m_copydata");
562 		if (off < m->m_len)
563 			break;
564 		off -= m->m_len;
565 		m = m->m_next;
566 	}
567 	while (len > 0) {
568 		if (m == 0)
569 			panic("m_copydata");
570 		count = min(m->m_len - off, len);
571 		memcpy(cp, mtod(m, caddr_t) + off, count);
572 		len -= count;
573 		cp += count;
574 		off = 0;
575 		m = m->m_next;
576 	}
577 }
578 
579 /*
580  * Concatenate mbuf chain n to m.
581  * Both chains must be of the same type (e.g. MT_DATA).
582  * Any m_pkthdr is not updated.
583  */
584 void
585 m_cat(m, n)
586 	struct mbuf *m, *n;
587 {
588 	while (m->m_next)
589 		m = m->m_next;
590 	while (n) {
591 		if (m->m_flags & M_EXT ||
592 		    m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
593 			/* just join the two chains */
594 			m->m_next = n;
595 			return;
596 		}
597 		/* splat the data from one into the other */
598 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
599 		    (u_int)n->m_len);
600 		m->m_len += n->m_len;
601 		n = m_free(n);
602 	}
603 }
604 
605 void
606 m_adj(mp, req_len)
607 	struct mbuf *mp;
608 	int req_len;
609 {
610 	int len = req_len;
611 	struct mbuf *m;
612 	int count;
613 
614 	if ((m = mp) == NULL)
615 		return;
616 	if (len >= 0) {
617 		/*
618 		 * Trim from head.
619 		 */
620 		while (m != NULL && len > 0) {
621 			if (m->m_len <= len) {
622 				len -= m->m_len;
623 				m->m_len = 0;
624 				m = m->m_next;
625 			} else {
626 				m->m_len -= len;
627 				m->m_data += len;
628 				len = 0;
629 			}
630 		}
631 		m = mp;
632 		if (mp->m_flags & M_PKTHDR)
633 			m->m_pkthdr.len -= (req_len - len);
634 	} else {
635 		/*
636 		 * Trim from tail.  Scan the mbuf chain,
637 		 * calculating its length and finding the last mbuf.
638 		 * If the adjustment only affects this mbuf, then just
639 		 * adjust and return.  Otherwise, rescan and truncate
640 		 * after the remaining size.
641 		 */
642 		len = -len;
643 		count = 0;
644 		for (;;) {
645 			count += m->m_len;
646 			if (m->m_next == (struct mbuf *)0)
647 				break;
648 			m = m->m_next;
649 		}
650 		if (m->m_len >= len) {
651 			m->m_len -= len;
652 			if (mp->m_flags & M_PKTHDR)
653 				mp->m_pkthdr.len -= len;
654 			return;
655 		}
656 		count -= len;
657 		if (count < 0)
658 			count = 0;
659 		/*
660 		 * Correct length for chain is "count".
661 		 * Find the mbuf with last data, adjust its length,
662 		 * and toss data from remaining mbufs on chain.
663 		 */
664 		m = mp;
665 		if (m->m_flags & M_PKTHDR)
666 			m->m_pkthdr.len = count;
667 		for (; m; m = m->m_next) {
668 			if (m->m_len >= count) {
669 				m->m_len = count;
670 				break;
671 			}
672 			count -= m->m_len;
673 		}
674 		while (m->m_next)
675 			(m = m->m_next) ->m_len = 0;
676 	}
677 }
678 
679 /*
680  * Rearange an mbuf chain so that len bytes are contiguous
681  * and in the data area of an mbuf (so that mtod and dtom
682  * will work for a structure of size len).  Returns the resulting
683  * mbuf chain on success, frees it and returns null on failure.
684  * If there is room, it will add up to max_protohdr-len extra bytes to the
685  * contiguous region in an attempt to avoid being called next time.
686  */
687 int MPFail;
688 
689 struct mbuf *
690 m_pullup(n, len)
691 	struct mbuf *n;
692 	int len;
693 {
694 	struct mbuf *m;
695 	int count;
696 	int space;
697 
698 	/*
699 	 * If first mbuf has no cluster, and has room for len bytes
700 	 * without shifting current data, pullup into it,
701 	 * otherwise allocate a new mbuf to prepend to the chain.
702 	 */
703 	if ((n->m_flags & M_EXT) == 0 &&
704 	    n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
705 		if (n->m_len >= len)
706 			return (n);
707 		m = n;
708 		n = n->m_next;
709 		len -= m->m_len;
710 	} else {
711 		if (len > MHLEN)
712 			goto bad;
713 		MGET(m, M_DONTWAIT, n->m_type);
714 		if (m == 0)
715 			goto bad;
716 		m->m_len = 0;
717 		if (n->m_flags & M_PKTHDR) {
718 			M_COPY_PKTHDR(m, n);
719 			n->m_flags &= ~M_PKTHDR;
720 		}
721 	}
722 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
723 	do {
724 		count = min(min(max(len, max_protohdr), space), n->m_len);
725 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
726 		  (unsigned)count);
727 		len -= count;
728 		m->m_len += count;
729 		n->m_len -= count;
730 		space -= count;
731 		if (n->m_len)
732 			n->m_data += count;
733 		else
734 			n = m_free(n);
735 	} while (len > 0 && n);
736 	if (len > 0) {
737 		(void) m_free(m);
738 		goto bad;
739 	}
740 	m->m_next = n;
741 	return (m);
742 bad:
743 	m_freem(n);
744 	MPFail++;
745 	return (0);
746 }
747 
748 /*
749  * Partition an mbuf chain in two pieces, returning the tail --
750  * all but the first len0 bytes.  In case of failure, it returns NULL and
751  * attempts to restore the chain to its original state.
752  */
753 struct mbuf *
754 m_split(m0, len0, wait)
755 	struct mbuf *m0;
756 	int len0, wait;
757 {
758 	struct mbuf *m, *n;
759 	unsigned len = len0, remain, len_save;
760 
761 	for (m = m0; m && len > m->m_len; m = m->m_next)
762 		len -= m->m_len;
763 	if (m == 0)
764 		return (0);
765 	remain = m->m_len - len;
766 	if (m0->m_flags & M_PKTHDR) {
767 		MGETHDR(n, wait, m0->m_type);
768 		if (n == 0)
769 			return (0);
770 		n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
771 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
772 		len_save = m0->m_pkthdr.len;
773 		m0->m_pkthdr.len = len0;
774 		if (m->m_flags & M_EXT)
775 			goto extpacket;
776 		if (remain > MHLEN) {
777 			/* m can't be the lead packet */
778 			MH_ALIGN(n, 0);
779 			n->m_next = m_split(m, len, wait);
780 			if (n->m_next == 0) {
781 				(void) m_free(n);
782 				m0->m_pkthdr.len = len_save;
783 				return (0);
784 			} else
785 				return (n);
786 		} else
787 			MH_ALIGN(n, remain);
788 	} else if (remain == 0) {
789 		n = m->m_next;
790 		m->m_next = 0;
791 		return (n);
792 	} else {
793 		MGET(n, wait, m->m_type);
794 		if (n == 0)
795 			return (0);
796 		M_ALIGN(n, remain);
797 	}
798 extpacket:
799 	if (m->m_flags & M_EXT) {
800 		n->m_ext = m->m_ext;
801 		MCLADDREFERENCE(m, n);
802 		n->m_data = m->m_data + len;
803 	} else {
804 		memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + len, remain);
805 	}
806 	n->m_len = remain;
807 	m->m_len = len;
808 	n->m_next = m->m_next;
809 	m->m_next = 0;
810 	return (n);
811 }
812 /*
813  * Routine to copy from device local memory into mbufs.
814  */
815 struct mbuf *
816 m_devget(buf, totlen, off0, ifp, copy)
817 	char *buf;
818 	int totlen, off0;
819 	struct ifnet *ifp;
820 	void (*copy) __P((const void *from, void *to, size_t len));
821 {
822 	struct mbuf *m;
823 	struct mbuf *top = 0, **mp = &top;
824 	int off = off0, len;
825 	char *cp;
826 	char *epkt;
827 
828 	cp = buf;
829 	epkt = cp + totlen;
830 	if (off) {
831 		/*
832 		 * If 'off' is non-zero, packet is trailer-encapsulated,
833 		 * so we have to skip the type and length fields.
834 		 */
835 		cp += off + 2 * sizeof(u_int16_t);
836 		totlen -= 2 * sizeof(u_int16_t);
837 	}
838 	MGETHDR(m, M_DONTWAIT, MT_DATA);
839 	if (m == 0)
840 		return (0);
841 	m->m_pkthdr.rcvif = ifp;
842 	m->m_pkthdr.len = totlen;
843 	m->m_len = MHLEN;
844 
845 	while (totlen > 0) {
846 		if (top) {
847 			MGET(m, M_DONTWAIT, MT_DATA);
848 			if (m == 0) {
849 				m_freem(top);
850 				return (0);
851 			}
852 			m->m_len = MLEN;
853 		}
854 		len = min(totlen, epkt - cp);
855 		if (len >= MINCLSIZE) {
856 			MCLGET(m, M_DONTWAIT);
857 			if ((m->m_flags & M_EXT) == 0) {
858 				m_free(m);
859 				m_freem(top);
860 				return (0);
861 			}
862 			m->m_len = len = min(len, MCLBYTES);
863 		} else {
864 			/*
865 			 * Place initial small packet/header at end of mbuf.
866 			 */
867 			if (len < m->m_len) {
868 				if (top == 0 && len + max_linkhdr <= m->m_len)
869 					m->m_data += max_linkhdr;
870 				m->m_len = len;
871 			} else
872 				len = m->m_len;
873 		}
874 		if (copy)
875 			copy(cp, mtod(m, caddr_t), (size_t)len);
876 		else
877 			memcpy(mtod(m, caddr_t), cp, (size_t)len);
878 		cp += len;
879 		*mp = m;
880 		mp = &m->m_next;
881 		totlen -= len;
882 		if (cp == epkt)
883 			cp = buf;
884 	}
885 	return (top);
886 }
887 
888 /*
889  * Copy data from a buffer back into the indicated mbuf chain,
890  * starting "off" bytes from the beginning, extending the mbuf
891  * chain if necessary.
892  */
893 void
894 m_copyback(m0, off, len, cp)
895 	struct	mbuf *m0;
896 	int off;
897 	int len;
898 	caddr_t cp;
899 {
900 	int mlen;
901 	struct mbuf *m = m0, *n;
902 	int totlen = 0;
903 
904 	if (m0 == 0)
905 		return;
906 	while (off > (mlen = m->m_len)) {
907 		off -= mlen;
908 		totlen += mlen;
909 		if (m->m_next == 0) {
910 			n = m_getclr(M_DONTWAIT, m->m_type);
911 			if (n == 0)
912 				goto out;
913 			n->m_len = min(MLEN, len + off);
914 			m->m_next = n;
915 		}
916 		m = m->m_next;
917 	}
918 	while (len > 0) {
919 		mlen = min (m->m_len - off, len);
920 		memcpy(mtod(m, caddr_t) + off, cp, (unsigned)mlen);
921 		cp += mlen;
922 		len -= mlen;
923 		mlen += off;
924 		off = 0;
925 		totlen += mlen;
926 		if (len == 0)
927 			break;
928 		if (m->m_next == 0) {
929 			n = m_get(M_DONTWAIT, m->m_type);
930 			if (n == 0)
931 				break;
932 			n->m_len = min(MLEN, len);
933 			m->m_next = n;
934 		}
935 		m = m->m_next;
936 	}
937 out:	if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
938 		m->m_pkthdr.len = totlen;
939 }
940