xref: /netbsd/sys/kern/uipc_mbuf.c (revision bf9ec67e)
1 /*	$NetBSD: uipc_mbuf.c,v 1.59 2002/03/09 01:46:33 thorpej Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2001 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/cdefs.h>
76 __KERNEL_RCSID(0, "$NetBSD: uipc_mbuf.c,v 1.59 2002/03/09 01:46:33 thorpej Exp $");
77 
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/proc.h>
81 #include <sys/malloc.h>
82 #include <sys/map.h>
83 #define MBTYPES
84 #include <sys/mbuf.h>
85 #include <sys/kernel.h>
86 #include <sys/syslog.h>
87 #include <sys/domain.h>
88 #include <sys/protosw.h>
89 #include <sys/pool.h>
90 #include <sys/socket.h>
91 #include <sys/sysctl.h>
92 
93 #include <net/if.h>
94 
95 #include <uvm/uvm_extern.h>
96 
97 
98 struct	pool mbpool;		/* mbuf pool */
99 struct	pool mclpool;		/* mbuf cluster pool */
100 
101 struct pool_cache mbpool_cache;
102 struct pool_cache mclpool_cache;
103 
104 struct mbstat mbstat;
105 int	max_linkhdr;
106 int	max_protohdr;
107 int	max_hdr;
108 int	max_datalen;
109 
110 void	*mclpool_alloc __P((struct pool *, int));
111 void	mclpool_release __P((struct pool *, void *));
112 
113 struct pool_allocator mclpool_allocator = {
114 	mclpool_alloc, mclpool_release, 0,
115 };
116 
117 static struct mbuf *m_copym0 __P((struct mbuf *, int, int, int, int));
118 
119 const char mclpool_warnmsg[] =
120     "WARNING: mclpool limit reached; increase NMBCLUSTERS";
121 
122 /*
123  * Initialize the mbuf allcator.
124  */
125 void
126 mbinit()
127 {
128 
129 	pool_init(&mbpool, msize, 0, 0, 0, "mbpl", NULL);
130 	pool_init(&mclpool, mclbytes, 0, 0, 0, "mclpl", &mclpool_allocator);
131 
132 	pool_set_drain_hook(&mbpool, m_reclaim, NULL);
133 	pool_set_drain_hook(&mclpool, m_reclaim, NULL);
134 
135 	pool_cache_init(&mbpool_cache, &mbpool, NULL, NULL, NULL);
136 	pool_cache_init(&mclpool_cache, &mclpool, NULL, NULL, NULL);
137 
138 	/*
139 	 * Set the hard limit on the mclpool to the number of
140 	 * mbuf clusters the kernel is to support.  Log the limit
141 	 * reached message max once a minute.
142 	 */
143 	pool_sethardlimit(&mclpool, nmbclusters, mclpool_warnmsg, 60);
144 
145 	/*
146 	 * Set a low water mark for both mbufs and clusters.  This should
147 	 * help ensure that they can be allocated in a memory starvation
148 	 * situation.  This is important for e.g. diskless systems which
149 	 * must allocate mbufs in order for the pagedaemon to clean pages.
150 	 */
151 	pool_setlowat(&mbpool, mblowat);
152 	pool_setlowat(&mclpool, mcllowat);
153 }
154 
155 int
156 sysctl_dombuf(name, namelen, oldp, oldlenp, newp, newlen)
157 	int *name;
158 	u_int namelen;
159 	void *oldp;
160 	size_t *oldlenp;
161 	void *newp;
162 	size_t newlen;
163 {
164 	int error, newval;
165 
166 	/* All sysctl names at this level are terminal. */
167 	if (namelen != 1)
168 		return (ENOTDIR);		/* overloaded */
169 
170 	switch (name[0]) {
171 	case MBUF_MSIZE:
172 		return (sysctl_rdint(oldp, oldlenp, newp, msize));
173 	case MBUF_MCLBYTES:
174 		return (sysctl_rdint(oldp, oldlenp, newp, mclbytes));
175 	case MBUF_NMBCLUSTERS:
176 		/*
177 		 * If we have direct-mapped pool pages, we can adjust this
178 		 * number on the fly.  If not, we're limited by the size
179 		 * of mb_map, and cannot change this value.
180 		 *
181 		 * Note: we only allow the value to be increased, never
182 		 * decreased.
183 		 */
184 		if (mb_map == NULL) {
185 			newval = nmbclusters;
186 			error = sysctl_int(oldp, oldlenp, newp, newlen,
187 			    &newval);
188 			if (error != 0)
189 				return (error);
190 			if (newp != NULL) {
191 				if (newval >= nmbclusters) {
192 					nmbclusters = newval;
193 					pool_sethardlimit(&mclpool,
194 					    nmbclusters, mclpool_warnmsg, 60);
195 				} else
196 					error = EINVAL;
197 			}
198 			return (error);
199 		} else
200 			return (sysctl_rdint(oldp, oldlenp, newp, nmbclusters));
201 	case MBUF_MBLOWAT:
202 	case MBUF_MCLLOWAT:
203 		/* New value must be >= 0. */
204 		newval = (name[0] == MBUF_MBLOWAT) ? mblowat : mcllowat;
205 		error = sysctl_int(oldp, oldlenp, newp, newlen, &newval);
206 		if (error != 0)
207 			return (error);
208 		if (newp != NULL) {
209 			if (newval >= 0) {
210 				if (name[0] == MBUF_MBLOWAT) {
211 					mblowat = newval;
212 					pool_setlowat(&mbpool, newval);
213 				} else {
214 					mcllowat = newval;
215 					pool_setlowat(&mclpool, newval);
216 				}
217 			} else
218 				error = EINVAL;
219 		}
220 		return (error);
221 	default:
222 		return (EOPNOTSUPP);
223 	}
224 	/* NOTREACHED */
225 }
226 
227 void *
228 mclpool_alloc(pp, flags)
229 	struct pool *pp;
230 	int flags;
231 {
232 	boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
233 
234 	return ((void *)uvm_km_alloc_poolpage1(mb_map, NULL, waitok));
235 }
236 
237 void
238 mclpool_release(pp, v)
239 	struct pool *pp;
240 	void *v;
241 {
242 
243 	uvm_km_free_poolpage1(mb_map, (vaddr_t)v);
244 }
245 
246 void
247 m_reclaim(void *arg, int flags)
248 {
249 	struct domain *dp;
250 	struct protosw *pr;
251 	struct ifnet *ifp;
252 	int s = splvm();
253 
254 	for (dp = domains; dp; dp = dp->dom_next)
255 		for (pr = dp->dom_protosw;
256 		     pr < dp->dom_protoswNPROTOSW; pr++)
257 			if (pr->pr_drain)
258 				(*pr->pr_drain)();
259 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list))
260 		if (ifp->if_drain)
261 			(*ifp->if_drain)(ifp);
262 	splx(s);
263 	mbstat.m_drain++;
264 }
265 
266 /*
267  * Space allocation routines.
268  * These are also available as macros
269  * for critical paths.
270  */
271 struct mbuf *
272 m_get(nowait, type)
273 	int nowait, type;
274 {
275 	struct mbuf *m;
276 
277 	MGET(m, nowait, type);
278 	return (m);
279 }
280 
281 struct mbuf *
282 m_gethdr(nowait, type)
283 	int nowait, type;
284 {
285 	struct mbuf *m;
286 
287 	MGETHDR(m, nowait, type);
288 	return (m);
289 }
290 
291 struct mbuf *
292 m_getclr(nowait, type)
293 	int nowait, type;
294 {
295 	struct mbuf *m;
296 
297 	MGET(m, nowait, type);
298 	if (m == 0)
299 		return (0);
300 	memset(mtod(m, caddr_t), 0, MLEN);
301 	return (m);
302 }
303 
304 struct mbuf *
305 m_free(m)
306 	struct mbuf *m;
307 {
308 	struct mbuf *n;
309 
310 	MFREE(m, n);
311 	return (n);
312 }
313 
314 void
315 m_freem(m)
316 	struct mbuf *m;
317 {
318 	struct mbuf *n;
319 
320 	if (m == NULL)
321 		return;
322 	do {
323 		MFREE(m, n);
324 		m = n;
325 	} while (m);
326 }
327 
328 /*
329  * Mbuffer utility routines.
330  */
331 
332 /*
333  * Lesser-used path for M_PREPEND:
334  * allocate new mbuf to prepend to chain,
335  * copy junk along.
336  */
337 struct mbuf *
338 m_prepend(m, len, how)
339 	struct mbuf *m;
340 	int len, how;
341 {
342 	struct mbuf *mn;
343 
344 	MGET(mn, how, m->m_type);
345 	if (mn == (struct mbuf *)NULL) {
346 		m_freem(m);
347 		return ((struct mbuf *)NULL);
348 	}
349 	if (m->m_flags & M_PKTHDR) {
350 		M_COPY_PKTHDR(mn, m);
351 		m->m_flags &= ~M_PKTHDR;
352 	}
353 	mn->m_next = m;
354 	m = mn;
355 	if (len < MHLEN)
356 		MH_ALIGN(m, len);
357 	m->m_len = len;
358 	return (m);
359 }
360 
361 /*
362  * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
363  * continuing for "len" bytes.  If len is M_COPYALL, copy to end of mbuf.
364  * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
365  */
366 int MCFail;
367 
368 struct mbuf *
369 m_copym(m, off0, len, wait)
370 	struct mbuf *m;
371 	int off0, wait;
372 	int len;
373 {
374 	return m_copym0(m, off0, len, wait, 0);	/* shallow copy on M_EXT */
375 }
376 
377 struct mbuf *
378 m_dup(m, off0, len, wait)
379 	struct mbuf *m;
380 	int off0, wait;
381 	int len;
382 {
383 	return m_copym0(m, off0, len, wait, 1);	/* deep copy */
384 }
385 
386 static struct mbuf *
387 m_copym0(m, off0, len, wait, deep)
388 	struct mbuf *m;
389 	int off0, wait;
390 	int len;
391 	int deep;	/* deep copy */
392 {
393 	struct mbuf *n, **np;
394 	int off = off0;
395 	struct mbuf *top;
396 	int copyhdr = 0;
397 
398 	if (off < 0 || len < 0)
399 		panic("m_copym: off %d, len %d", off, len);
400 	if (off == 0 && m->m_flags & M_PKTHDR)
401 		copyhdr = 1;
402 	while (off > 0) {
403 		if (m == 0)
404 			panic("m_copym: m == 0");
405 		if (off < m->m_len)
406 			break;
407 		off -= m->m_len;
408 		m = m->m_next;
409 	}
410 	np = &top;
411 	top = 0;
412 	while (len > 0) {
413 		if (m == 0) {
414 			if (len != M_COPYALL)
415 				panic("m_copym: m == 0 and not COPYALL");
416 			break;
417 		}
418 		MGET(n, wait, m->m_type);
419 		*np = n;
420 		if (n == 0)
421 			goto nospace;
422 		if (copyhdr) {
423 			M_COPY_PKTHDR(n, m);
424 			if (len == M_COPYALL)
425 				n->m_pkthdr.len -= off0;
426 			else
427 				n->m_pkthdr.len = len;
428 			copyhdr = 0;
429 		}
430 		n->m_len = min(len, m->m_len - off);
431 		if (m->m_flags & M_EXT) {
432 			if (!deep) {
433 				n->m_data = m->m_data + off;
434 				n->m_ext = m->m_ext;
435 				MCLADDREFERENCE(m, n);
436 			} else {
437 				/*
438 				 * we are unsure about the way m was allocated.
439 				 * copy into multiple MCLBYTES cluster mbufs.
440 				 */
441 				MCLGET(n, wait);
442 				n->m_len = 0;
443 				n->m_len = M_TRAILINGSPACE(n);
444 				n->m_len = min(n->m_len, len);
445 				n->m_len = min(n->m_len, m->m_len - off);
446 				memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off,
447 				    (unsigned)n->m_len);
448 			}
449 		} else
450 			memcpy(mtod(n, caddr_t), mtod(m, caddr_t)+off,
451 			    (unsigned)n->m_len);
452 		if (len != M_COPYALL)
453 			len -= n->m_len;
454 		off += n->m_len;
455 #ifdef DIAGNOSTIC
456 		if (off > m->m_len)
457 			panic("m_copym0 overrun");
458 #endif
459 		if (off == m->m_len) {
460 			m = m->m_next;
461 			off = 0;
462 		}
463 		np = &n->m_next;
464 	}
465 	if (top == 0)
466 		MCFail++;
467 	return (top);
468 nospace:
469 	m_freem(top);
470 	MCFail++;
471 	return (0);
472 }
473 
474 /*
475  * Copy an entire packet, including header (which must be present).
476  * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
477  */
478 struct mbuf *
479 m_copypacket(m, how)
480 	struct mbuf *m;
481 	int how;
482 {
483 	struct mbuf *top, *n, *o;
484 
485 	MGET(n, how, m->m_type);
486 	top = n;
487 	if (!n)
488 		goto nospace;
489 
490 	M_COPY_PKTHDR(n, m);
491 	n->m_len = m->m_len;
492 	if (m->m_flags & M_EXT) {
493 		n->m_data = m->m_data;
494 		n->m_ext = m->m_ext;
495 		MCLADDREFERENCE(m, n);
496 	} else {
497 		memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
498 	}
499 
500 	m = m->m_next;
501 	while (m) {
502 		MGET(o, how, m->m_type);
503 		if (!o)
504 			goto nospace;
505 
506 		n->m_next = o;
507 		n = n->m_next;
508 
509 		n->m_len = m->m_len;
510 		if (m->m_flags & M_EXT) {
511 			n->m_data = m->m_data;
512 			n->m_ext = m->m_ext;
513 			MCLADDREFERENCE(m, n);
514 		} else {
515 			memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
516 		}
517 
518 		m = m->m_next;
519 	}
520 	return top;
521 nospace:
522 	m_freem(top);
523 	MCFail++;
524 	return 0;
525 }
526 
527 /*
528  * Copy data from an mbuf chain starting "off" bytes from the beginning,
529  * continuing for "len" bytes, into the indicated buffer.
530  */
531 void
532 m_copydata(m, off, len, cp)
533 	struct mbuf *m;
534 	int off;
535 	int len;
536 	caddr_t cp;
537 {
538 	unsigned count;
539 
540 	if (off < 0 || len < 0)
541 		panic("m_copydata");
542 	while (off > 0) {
543 		if (m == 0)
544 			panic("m_copydata");
545 		if (off < m->m_len)
546 			break;
547 		off -= m->m_len;
548 		m = m->m_next;
549 	}
550 	while (len > 0) {
551 		if (m == 0)
552 			panic("m_copydata");
553 		count = min(m->m_len - off, len);
554 		memcpy(cp, mtod(m, caddr_t) + off, count);
555 		len -= count;
556 		cp += count;
557 		off = 0;
558 		m = m->m_next;
559 	}
560 }
561 
562 /*
563  * Concatenate mbuf chain n to m.
564  * Both chains must be of the same type (e.g. MT_DATA).
565  * Any m_pkthdr is not updated.
566  */
567 void
568 m_cat(m, n)
569 	struct mbuf *m, *n;
570 {
571 	while (m->m_next)
572 		m = m->m_next;
573 	while (n) {
574 		if (m->m_flags & M_EXT ||
575 		    m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
576 			/* just join the two chains */
577 			m->m_next = n;
578 			return;
579 		}
580 		/* splat the data from one into the other */
581 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
582 		    (u_int)n->m_len);
583 		m->m_len += n->m_len;
584 		n = m_free(n);
585 	}
586 }
587 
588 void
589 m_adj(mp, req_len)
590 	struct mbuf *mp;
591 	int req_len;
592 {
593 	int len = req_len;
594 	struct mbuf *m;
595 	int count;
596 
597 	if ((m = mp) == NULL)
598 		return;
599 	if (len >= 0) {
600 		/*
601 		 * Trim from head.
602 		 */
603 		while (m != NULL && len > 0) {
604 			if (m->m_len <= len) {
605 				len -= m->m_len;
606 				m->m_len = 0;
607 				m = m->m_next;
608 			} else {
609 				m->m_len -= len;
610 				m->m_data += len;
611 				len = 0;
612 			}
613 		}
614 		m = mp;
615 		if (mp->m_flags & M_PKTHDR)
616 			m->m_pkthdr.len -= (req_len - len);
617 	} else {
618 		/*
619 		 * Trim from tail.  Scan the mbuf chain,
620 		 * calculating its length and finding the last mbuf.
621 		 * If the adjustment only affects this mbuf, then just
622 		 * adjust and return.  Otherwise, rescan and truncate
623 		 * after the remaining size.
624 		 */
625 		len = -len;
626 		count = 0;
627 		for (;;) {
628 			count += m->m_len;
629 			if (m->m_next == (struct mbuf *)0)
630 				break;
631 			m = m->m_next;
632 		}
633 		if (m->m_len >= len) {
634 			m->m_len -= len;
635 			if (mp->m_flags & M_PKTHDR)
636 				mp->m_pkthdr.len -= len;
637 			return;
638 		}
639 		count -= len;
640 		if (count < 0)
641 			count = 0;
642 		/*
643 		 * Correct length for chain is "count".
644 		 * Find the mbuf with last data, adjust its length,
645 		 * and toss data from remaining mbufs on chain.
646 		 */
647 		m = mp;
648 		if (m->m_flags & M_PKTHDR)
649 			m->m_pkthdr.len = count;
650 		for (; m; m = m->m_next) {
651 			if (m->m_len >= count) {
652 				m->m_len = count;
653 				break;
654 			}
655 			count -= m->m_len;
656 		}
657 		while (m->m_next)
658 			(m = m->m_next) ->m_len = 0;
659 	}
660 }
661 
662 /*
663  * Rearange an mbuf chain so that len bytes are contiguous
664  * and in the data area of an mbuf (so that mtod and dtom
665  * will work for a structure of size len).  Returns the resulting
666  * mbuf chain on success, frees it and returns null on failure.
667  * If there is room, it will add up to max_protohdr-len extra bytes to the
668  * contiguous region in an attempt to avoid being called next time.
669  */
670 int MPFail;
671 
672 struct mbuf *
673 m_pullup(n, len)
674 	struct mbuf *n;
675 	int len;
676 {
677 	struct mbuf *m;
678 	int count;
679 	int space;
680 
681 	/*
682 	 * If first mbuf has no cluster, and has room for len bytes
683 	 * without shifting current data, pullup into it,
684 	 * otherwise allocate a new mbuf to prepend to the chain.
685 	 */
686 	if ((n->m_flags & M_EXT) == 0 &&
687 	    n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
688 		if (n->m_len >= len)
689 			return (n);
690 		m = n;
691 		n = n->m_next;
692 		len -= m->m_len;
693 	} else {
694 		if (len > MHLEN)
695 			goto bad;
696 		MGET(m, M_DONTWAIT, n->m_type);
697 		if (m == 0)
698 			goto bad;
699 		m->m_len = 0;
700 		if (n->m_flags & M_PKTHDR) {
701 			M_COPY_PKTHDR(m, n);
702 			n->m_flags &= ~M_PKTHDR;
703 		}
704 	}
705 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
706 	do {
707 		count = min(min(max(len, max_protohdr), space), n->m_len);
708 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
709 		  (unsigned)count);
710 		len -= count;
711 		m->m_len += count;
712 		n->m_len -= count;
713 		space -= count;
714 		if (n->m_len)
715 			n->m_data += count;
716 		else
717 			n = m_free(n);
718 	} while (len > 0 && n);
719 	if (len > 0) {
720 		(void) m_free(m);
721 		goto bad;
722 	}
723 	m->m_next = n;
724 	return (m);
725 bad:
726 	m_freem(n);
727 	MPFail++;
728 	return (0);
729 }
730 
731 /*
732  * Partition an mbuf chain in two pieces, returning the tail --
733  * all but the first len0 bytes.  In case of failure, it returns NULL and
734  * attempts to restore the chain to its original state.
735  */
736 struct mbuf *
737 m_split(m0, len0, wait)
738 	struct mbuf *m0;
739 	int len0, wait;
740 {
741 	struct mbuf *m, *n;
742 	unsigned len = len0, remain, len_save;
743 
744 	for (m = m0; m && len > m->m_len; m = m->m_next)
745 		len -= m->m_len;
746 	if (m == 0)
747 		return (0);
748 	remain = m->m_len - len;
749 	if (m0->m_flags & M_PKTHDR) {
750 		MGETHDR(n, wait, m0->m_type);
751 		if (n == 0)
752 			return (0);
753 		n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
754 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
755 		len_save = m0->m_pkthdr.len;
756 		m0->m_pkthdr.len = len0;
757 		if (m->m_flags & M_EXT)
758 			goto extpacket;
759 		if (remain > MHLEN) {
760 			/* m can't be the lead packet */
761 			MH_ALIGN(n, 0);
762 			n->m_next = m_split(m, len, wait);
763 			if (n->m_next == 0) {
764 				(void) m_free(n);
765 				m0->m_pkthdr.len = len_save;
766 				return (0);
767 			} else
768 				return (n);
769 		} else
770 			MH_ALIGN(n, remain);
771 	} else if (remain == 0) {
772 		n = m->m_next;
773 		m->m_next = 0;
774 		return (n);
775 	} else {
776 		MGET(n, wait, m->m_type);
777 		if (n == 0)
778 			return (0);
779 		M_ALIGN(n, remain);
780 	}
781 extpacket:
782 	if (m->m_flags & M_EXT) {
783 		n->m_ext = m->m_ext;
784 		MCLADDREFERENCE(m, n);
785 		n->m_data = m->m_data + len;
786 	} else {
787 		memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + len, remain);
788 	}
789 	n->m_len = remain;
790 	m->m_len = len;
791 	n->m_next = m->m_next;
792 	m->m_next = 0;
793 	return (n);
794 }
795 /*
796  * Routine to copy from device local memory into mbufs.
797  */
798 struct mbuf *
799 m_devget(buf, totlen, off0, ifp, copy)
800 	char *buf;
801 	int totlen, off0;
802 	struct ifnet *ifp;
803 	void (*copy) __P((const void *from, void *to, size_t len));
804 {
805 	struct mbuf *m;
806 	struct mbuf *top = 0, **mp = &top;
807 	int off = off0, len;
808 	char *cp;
809 	char *epkt;
810 
811 	cp = buf;
812 	epkt = cp + totlen;
813 	if (off) {
814 		/*
815 		 * If 'off' is non-zero, packet is trailer-encapsulated,
816 		 * so we have to skip the type and length fields.
817 		 */
818 		cp += off + 2 * sizeof(u_int16_t);
819 		totlen -= 2 * sizeof(u_int16_t);
820 	}
821 	MGETHDR(m, M_DONTWAIT, MT_DATA);
822 	if (m == 0)
823 		return (0);
824 	m->m_pkthdr.rcvif = ifp;
825 	m->m_pkthdr.len = totlen;
826 	m->m_len = MHLEN;
827 
828 	while (totlen > 0) {
829 		if (top) {
830 			MGET(m, M_DONTWAIT, MT_DATA);
831 			if (m == 0) {
832 				m_freem(top);
833 				return (0);
834 			}
835 			m->m_len = MLEN;
836 		}
837 		len = min(totlen, epkt - cp);
838 		if (len >= MINCLSIZE) {
839 			MCLGET(m, M_DONTWAIT);
840 			if ((m->m_flags & M_EXT) == 0) {
841 				m_free(m);
842 				m_freem(top);
843 				return (0);
844 			}
845 			m->m_len = len = min(len, MCLBYTES);
846 		} else {
847 			/*
848 			 * Place initial small packet/header at end of mbuf.
849 			 */
850 			if (len < m->m_len) {
851 				if (top == 0 && len + max_linkhdr <= m->m_len)
852 					m->m_data += max_linkhdr;
853 				m->m_len = len;
854 			} else
855 				len = m->m_len;
856 		}
857 		if (copy)
858 			copy(cp, mtod(m, caddr_t), (size_t)len);
859 		else
860 			memcpy(mtod(m, caddr_t), cp, (size_t)len);
861 		cp += len;
862 		*mp = m;
863 		mp = &m->m_next;
864 		totlen -= len;
865 		if (cp == epkt)
866 			cp = buf;
867 	}
868 	return (top);
869 }
870 
871 /*
872  * Copy data from a buffer back into the indicated mbuf chain,
873  * starting "off" bytes from the beginning, extending the mbuf
874  * chain if necessary.
875  */
876 void
877 m_copyback(m0, off, len, cp)
878 	struct	mbuf *m0;
879 	int off;
880 	int len;
881 	caddr_t cp;
882 {
883 	int mlen;
884 	struct mbuf *m = m0, *n;
885 	int totlen = 0;
886 
887 	if (m0 == 0)
888 		return;
889 	while (off > (mlen = m->m_len)) {
890 		off -= mlen;
891 		totlen += mlen;
892 		if (m->m_next == 0) {
893 			n = m_getclr(M_DONTWAIT, m->m_type);
894 			if (n == 0)
895 				goto out;
896 			n->m_len = min(MLEN, len + off);
897 			m->m_next = n;
898 		}
899 		m = m->m_next;
900 	}
901 	while (len > 0) {
902 		mlen = min (m->m_len - off, len);
903 		memcpy(mtod(m, caddr_t) + off, cp, (unsigned)mlen);
904 		cp += mlen;
905 		len -= mlen;
906 		mlen += off;
907 		off = 0;
908 		totlen += mlen;
909 		if (len == 0)
910 			break;
911 		if (m->m_next == 0) {
912 			n = m_get(M_DONTWAIT, m->m_type);
913 			if (n == 0)
914 				break;
915 			n->m_len = min(MLEN, len);
916 			m->m_next = n;
917 		}
918 		m = m->m_next;
919 	}
920 out:	if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
921 		m->m_pkthdr.len = totlen;
922 }
923