xref: /netbsd/sys/kern/uipc_mbuf.c (revision c4a72b64)
1 /*	$NetBSD: uipc_mbuf.c,v 1.61 2002/09/25 22:21:45 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 1.61 2002/09/25 22:21:45 thorpej Exp $");
77 
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/proc.h>
81 #include <sys/malloc.h>
82 #define MBTYPES
83 #include <sys/mbuf.h>
84 #include <sys/kernel.h>
85 #include <sys/syslog.h>
86 #include <sys/domain.h>
87 #include <sys/protosw.h>
88 #include <sys/pool.h>
89 #include <sys/socket.h>
90 #include <sys/sysctl.h>
91 
92 #include <net/if.h>
93 
94 #include <uvm/uvm_extern.h>
95 
96 
97 struct	pool mbpool;		/* mbuf pool */
98 struct	pool mclpool;		/* mbuf cluster pool */
99 
100 struct pool_cache mbpool_cache;
101 struct pool_cache mclpool_cache;
102 
103 struct mbstat mbstat;
104 int	max_linkhdr;
105 int	max_protohdr;
106 int	max_hdr;
107 int	max_datalen;
108 
109 void	*mclpool_alloc __P((struct pool *, int));
110 void	mclpool_release __P((struct pool *, void *));
111 
112 struct pool_allocator mclpool_allocator = {
113 	mclpool_alloc, mclpool_release, 0,
114 };
115 
116 static struct mbuf *m_copym0 __P((struct mbuf *, int, int, int, int));
117 
118 const char mclpool_warnmsg[] =
119     "WARNING: mclpool limit reached; increase NMBCLUSTERS";
120 
121 /*
122  * Initialize the mbuf allcator.
123  */
124 void
125 mbinit()
126 {
127 
128 	pool_init(&mbpool, msize, 0, 0, 0, "mbpl", NULL);
129 	pool_init(&mclpool, mclbytes, 0, 0, 0, "mclpl", &mclpool_allocator);
130 
131 	pool_set_drain_hook(&mbpool, m_reclaim, NULL);
132 	pool_set_drain_hook(&mclpool, m_reclaim, NULL);
133 
134 	pool_cache_init(&mbpool_cache, &mbpool, NULL, NULL, NULL);
135 	pool_cache_init(&mclpool_cache, &mclpool, NULL, NULL, NULL);
136 
137 	/*
138 	 * Set the hard limit on the mclpool to the number of
139 	 * mbuf clusters the kernel is to support.  Log the limit
140 	 * reached message max once a minute.
141 	 */
142 	pool_sethardlimit(&mclpool, nmbclusters, mclpool_warnmsg, 60);
143 
144 	/*
145 	 * Set a low water mark for both mbufs and clusters.  This should
146 	 * help ensure that they can be allocated in a memory starvation
147 	 * situation.  This is important for e.g. diskless systems which
148 	 * must allocate mbufs in order for the pagedaemon to clean pages.
149 	 */
150 	pool_setlowat(&mbpool, mblowat);
151 	pool_setlowat(&mclpool, mcllowat);
152 }
153 
154 int
155 sysctl_dombuf(name, namelen, oldp, oldlenp, newp, newlen)
156 	int *name;
157 	u_int namelen;
158 	void *oldp;
159 	size_t *oldlenp;
160 	void *newp;
161 	size_t newlen;
162 {
163 	int error, newval;
164 
165 	/* All sysctl names at this level are terminal. */
166 	if (namelen != 1)
167 		return (ENOTDIR);		/* overloaded */
168 
169 	switch (name[0]) {
170 	case MBUF_MSIZE:
171 		return (sysctl_rdint(oldp, oldlenp, newp, msize));
172 	case MBUF_MCLBYTES:
173 		return (sysctl_rdint(oldp, oldlenp, newp, mclbytes));
174 	case MBUF_NMBCLUSTERS:
175 		/*
176 		 * If we have direct-mapped pool pages, we can adjust this
177 		 * number on the fly.  If not, we're limited by the size
178 		 * of mb_map, and cannot change this value.
179 		 *
180 		 * Note: we only allow the value to be increased, never
181 		 * decreased.
182 		 */
183 		if (mb_map == NULL) {
184 			newval = nmbclusters;
185 			error = sysctl_int(oldp, oldlenp, newp, newlen,
186 			    &newval);
187 			if (error != 0)
188 				return (error);
189 			if (newp != NULL) {
190 				if (newval >= nmbclusters) {
191 					nmbclusters = newval;
192 					pool_sethardlimit(&mclpool,
193 					    nmbclusters, mclpool_warnmsg, 60);
194 				} else
195 					error = EINVAL;
196 			}
197 			return (error);
198 		} else
199 			return (sysctl_rdint(oldp, oldlenp, newp, nmbclusters));
200 	case MBUF_MBLOWAT:
201 	case MBUF_MCLLOWAT:
202 		/* New value must be >= 0. */
203 		newval = (name[0] == MBUF_MBLOWAT) ? mblowat : mcllowat;
204 		error = sysctl_int(oldp, oldlenp, newp, newlen, &newval);
205 		if (error != 0)
206 			return (error);
207 		if (newp != NULL) {
208 			if (newval >= 0) {
209 				if (name[0] == MBUF_MBLOWAT) {
210 					mblowat = newval;
211 					pool_setlowat(&mbpool, newval);
212 				} else {
213 					mcllowat = newval;
214 					pool_setlowat(&mclpool, newval);
215 				}
216 			} else
217 				error = EINVAL;
218 		}
219 		return (error);
220 	default:
221 		return (EOPNOTSUPP);
222 	}
223 	/* NOTREACHED */
224 }
225 
226 void *
227 mclpool_alloc(pp, flags)
228 	struct pool *pp;
229 	int flags;
230 {
231 	boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
232 
233 	return ((void *)uvm_km_alloc_poolpage1(mb_map, NULL, waitok));
234 }
235 
236 void
237 mclpool_release(pp, v)
238 	struct pool *pp;
239 	void *v;
240 {
241 
242 	uvm_km_free_poolpage1(mb_map, (vaddr_t)v);
243 }
244 
245 void
246 m_reclaim(void *arg, int flags)
247 {
248 	struct domain *dp;
249 	struct protosw *pr;
250 	struct ifnet *ifp;
251 	int s = splvm();
252 
253 	for (dp = domains; dp; dp = dp->dom_next)
254 		for (pr = dp->dom_protosw;
255 		     pr < dp->dom_protoswNPROTOSW; pr++)
256 			if (pr->pr_drain)
257 				(*pr->pr_drain)();
258 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list))
259 		if (ifp->if_drain)
260 			(*ifp->if_drain)(ifp);
261 	splx(s);
262 	mbstat.m_drain++;
263 }
264 
265 /*
266  * Space allocation routines.
267  * These are also available as macros
268  * for critical paths.
269  */
270 struct mbuf *
271 m_get(nowait, type)
272 	int nowait, type;
273 {
274 	struct mbuf *m;
275 
276 	MGET(m, nowait, type);
277 	return (m);
278 }
279 
280 struct mbuf *
281 m_gethdr(nowait, type)
282 	int nowait, type;
283 {
284 	struct mbuf *m;
285 
286 	MGETHDR(m, nowait, type);
287 	return (m);
288 }
289 
290 struct mbuf *
291 m_getclr(nowait, type)
292 	int nowait, type;
293 {
294 	struct mbuf *m;
295 
296 	MGET(m, nowait, type);
297 	if (m == 0)
298 		return (0);
299 	memset(mtod(m, caddr_t), 0, MLEN);
300 	return (m);
301 }
302 
303 struct mbuf *
304 m_free(m)
305 	struct mbuf *m;
306 {
307 	struct mbuf *n;
308 
309 	MFREE(m, n);
310 	return (n);
311 }
312 
313 void
314 m_freem(m)
315 	struct mbuf *m;
316 {
317 	struct mbuf *n;
318 
319 	if (m == NULL)
320 		return;
321 	do {
322 		MFREE(m, n);
323 		m = n;
324 	} while (m);
325 }
326 
327 /*
328  * Mbuffer utility routines.
329  */
330 
331 /*
332  * Lesser-used path for M_PREPEND:
333  * allocate new mbuf to prepend to chain,
334  * copy junk along.
335  */
336 struct mbuf *
337 m_prepend(m, len, how)
338 	struct mbuf *m;
339 	int len, how;
340 {
341 	struct mbuf *mn;
342 
343 	MGET(mn, how, m->m_type);
344 	if (mn == (struct mbuf *)NULL) {
345 		m_freem(m);
346 		return ((struct mbuf *)NULL);
347 	}
348 	if (m->m_flags & M_PKTHDR) {
349 		M_COPY_PKTHDR(mn, m);
350 		m->m_flags &= ~M_PKTHDR;
351 	}
352 	mn->m_next = m;
353 	m = mn;
354 	if (len < MHLEN)
355 		MH_ALIGN(m, len);
356 	m->m_len = len;
357 	return (m);
358 }
359 
360 /*
361  * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
362  * continuing for "len" bytes.  If len is M_COPYALL, copy to end of mbuf.
363  * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
364  */
365 int MCFail;
366 
367 struct mbuf *
368 m_copym(m, off0, len, wait)
369 	struct mbuf *m;
370 	int off0, wait;
371 	int len;
372 {
373 	return m_copym0(m, off0, len, wait, 0);	/* shallow copy on M_EXT */
374 }
375 
376 struct mbuf *
377 m_dup(m, off0, len, wait)
378 	struct mbuf *m;
379 	int off0, wait;
380 	int len;
381 {
382 	return m_copym0(m, off0, len, wait, 1);	/* deep copy */
383 }
384 
385 static struct mbuf *
386 m_copym0(m, off0, len, wait, deep)
387 	struct mbuf *m;
388 	int off0, wait;
389 	int len;
390 	int deep;	/* deep copy */
391 {
392 	struct mbuf *n, **np;
393 	int off = off0;
394 	struct mbuf *top;
395 	int copyhdr = 0;
396 
397 	if (off < 0 || len < 0)
398 		panic("m_copym: off %d, len %d", off, len);
399 	if (off == 0 && m->m_flags & M_PKTHDR)
400 		copyhdr = 1;
401 	while (off > 0) {
402 		if (m == 0)
403 			panic("m_copym: m == 0");
404 		if (off < m->m_len)
405 			break;
406 		off -= m->m_len;
407 		m = m->m_next;
408 	}
409 	np = &top;
410 	top = 0;
411 	while (len > 0) {
412 		if (m == 0) {
413 			if (len != M_COPYALL)
414 				panic("m_copym: m == 0 and not COPYALL");
415 			break;
416 		}
417 		MGET(n, wait, m->m_type);
418 		*np = n;
419 		if (n == 0)
420 			goto nospace;
421 		if (copyhdr) {
422 			M_COPY_PKTHDR(n, m);
423 			if (len == M_COPYALL)
424 				n->m_pkthdr.len -= off0;
425 			else
426 				n->m_pkthdr.len = len;
427 			copyhdr = 0;
428 		}
429 		n->m_len = min(len, m->m_len - off);
430 		if (m->m_flags & M_EXT) {
431 			if (!deep) {
432 				n->m_data = m->m_data + off;
433 				n->m_ext = m->m_ext;
434 				MCLADDREFERENCE(m, n);
435 			} else {
436 				/*
437 				 * we are unsure about the way m was allocated.
438 				 * copy into multiple MCLBYTES cluster mbufs.
439 				 */
440 				MCLGET(n, wait);
441 				n->m_len = 0;
442 				n->m_len = M_TRAILINGSPACE(n);
443 				n->m_len = min(n->m_len, len);
444 				n->m_len = min(n->m_len, m->m_len - off);
445 				memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off,
446 				    (unsigned)n->m_len);
447 			}
448 		} else
449 			memcpy(mtod(n, caddr_t), mtod(m, caddr_t)+off,
450 			    (unsigned)n->m_len);
451 		if (len != M_COPYALL)
452 			len -= n->m_len;
453 		off += n->m_len;
454 #ifdef DIAGNOSTIC
455 		if (off > m->m_len)
456 			panic("m_copym0 overrun");
457 #endif
458 		if (off == m->m_len) {
459 			m = m->m_next;
460 			off = 0;
461 		}
462 		np = &n->m_next;
463 	}
464 	if (top == 0)
465 		MCFail++;
466 	return (top);
467 nospace:
468 	m_freem(top);
469 	MCFail++;
470 	return (0);
471 }
472 
473 /*
474  * Copy an entire packet, including header (which must be present).
475  * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
476  */
477 struct mbuf *
478 m_copypacket(m, how)
479 	struct mbuf *m;
480 	int how;
481 {
482 	struct mbuf *top, *n, *o;
483 
484 	MGET(n, how, m->m_type);
485 	top = n;
486 	if (!n)
487 		goto nospace;
488 
489 	M_COPY_PKTHDR(n, m);
490 	n->m_len = m->m_len;
491 	if (m->m_flags & M_EXT) {
492 		n->m_data = m->m_data;
493 		n->m_ext = m->m_ext;
494 		MCLADDREFERENCE(m, n);
495 	} else {
496 		memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
497 	}
498 
499 	m = m->m_next;
500 	while (m) {
501 		MGET(o, how, m->m_type);
502 		if (!o)
503 			goto nospace;
504 
505 		n->m_next = o;
506 		n = n->m_next;
507 
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 	}
519 	return top;
520 nospace:
521 	m_freem(top);
522 	MCFail++;
523 	return 0;
524 }
525 
526 /*
527  * Copy data from an mbuf chain starting "off" bytes from the beginning,
528  * continuing for "len" bytes, into the indicated buffer.
529  */
530 void
531 m_copydata(m, off, len, cp)
532 	struct mbuf *m;
533 	int off;
534 	int len;
535 	caddr_t cp;
536 {
537 	unsigned count;
538 
539 	if (off < 0 || len < 0)
540 		panic("m_copydata");
541 	while (off > 0) {
542 		if (m == 0)
543 			panic("m_copydata");
544 		if (off < m->m_len)
545 			break;
546 		off -= m->m_len;
547 		m = m->m_next;
548 	}
549 	while (len > 0) {
550 		if (m == 0)
551 			panic("m_copydata");
552 		count = min(m->m_len - off, len);
553 		memcpy(cp, mtod(m, caddr_t) + off, count);
554 		len -= count;
555 		cp += count;
556 		off = 0;
557 		m = m->m_next;
558 	}
559 }
560 
561 /*
562  * Concatenate mbuf chain n to m.
563  * Both chains must be of the same type (e.g. MT_DATA).
564  * Any m_pkthdr is not updated.
565  */
566 void
567 m_cat(m, n)
568 	struct mbuf *m, *n;
569 {
570 	while (m->m_next)
571 		m = m->m_next;
572 	while (n) {
573 		if (m->m_flags & M_EXT ||
574 		    m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
575 			/* just join the two chains */
576 			m->m_next = n;
577 			return;
578 		}
579 		/* splat the data from one into the other */
580 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
581 		    (u_int)n->m_len);
582 		m->m_len += n->m_len;
583 		n = m_free(n);
584 	}
585 }
586 
587 void
588 m_adj(mp, req_len)
589 	struct mbuf *mp;
590 	int req_len;
591 {
592 	int len = req_len;
593 	struct mbuf *m;
594 	int count;
595 
596 	if ((m = mp) == NULL)
597 		return;
598 	if (len >= 0) {
599 		/*
600 		 * Trim from head.
601 		 */
602 		while (m != NULL && len > 0) {
603 			if (m->m_len <= len) {
604 				len -= m->m_len;
605 				m->m_len = 0;
606 				m = m->m_next;
607 			} else {
608 				m->m_len -= len;
609 				m->m_data += len;
610 				len = 0;
611 			}
612 		}
613 		m = mp;
614 		if (mp->m_flags & M_PKTHDR)
615 			m->m_pkthdr.len -= (req_len - len);
616 	} else {
617 		/*
618 		 * Trim from tail.  Scan the mbuf chain,
619 		 * calculating its length and finding the last mbuf.
620 		 * If the adjustment only affects this mbuf, then just
621 		 * adjust and return.  Otherwise, rescan and truncate
622 		 * after the remaining size.
623 		 */
624 		len = -len;
625 		count = 0;
626 		for (;;) {
627 			count += m->m_len;
628 			if (m->m_next == (struct mbuf *)0)
629 				break;
630 			m = m->m_next;
631 		}
632 		if (m->m_len >= len) {
633 			m->m_len -= len;
634 			if (mp->m_flags & M_PKTHDR)
635 				mp->m_pkthdr.len -= len;
636 			return;
637 		}
638 		count -= len;
639 		if (count < 0)
640 			count = 0;
641 		/*
642 		 * Correct length for chain is "count".
643 		 * Find the mbuf with last data, adjust its length,
644 		 * and toss data from remaining mbufs on chain.
645 		 */
646 		m = mp;
647 		if (m->m_flags & M_PKTHDR)
648 			m->m_pkthdr.len = count;
649 		for (; m; m = m->m_next) {
650 			if (m->m_len >= count) {
651 				m->m_len = count;
652 				break;
653 			}
654 			count -= m->m_len;
655 		}
656 		while (m->m_next)
657 			(m = m->m_next) ->m_len = 0;
658 	}
659 }
660 
661 /*
662  * Rearange an mbuf chain so that len bytes are contiguous
663  * and in the data area of an mbuf (so that mtod and dtom
664  * will work for a structure of size len).  Returns the resulting
665  * mbuf chain on success, frees it and returns null on failure.
666  * If there is room, it will add up to max_protohdr-len extra bytes to the
667  * contiguous region in an attempt to avoid being called next time.
668  */
669 int MPFail;
670 
671 struct mbuf *
672 m_pullup(n, len)
673 	struct mbuf *n;
674 	int len;
675 {
676 	struct mbuf *m;
677 	int count;
678 	int space;
679 
680 	/*
681 	 * If first mbuf has no cluster, and has room for len bytes
682 	 * without shifting current data, pullup into it,
683 	 * otherwise allocate a new mbuf to prepend to the chain.
684 	 */
685 	if ((n->m_flags & M_EXT) == 0 &&
686 	    n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
687 		if (n->m_len >= len)
688 			return (n);
689 		m = n;
690 		n = n->m_next;
691 		len -= m->m_len;
692 	} else {
693 		if (len > MHLEN)
694 			goto bad;
695 		MGET(m, M_DONTWAIT, n->m_type);
696 		if (m == 0)
697 			goto bad;
698 		m->m_len = 0;
699 		if (n->m_flags & M_PKTHDR) {
700 			M_COPY_PKTHDR(m, n);
701 			n->m_flags &= ~M_PKTHDR;
702 		}
703 	}
704 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
705 	do {
706 		count = min(min(max(len, max_protohdr), space), n->m_len);
707 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
708 		  (unsigned)count);
709 		len -= count;
710 		m->m_len += count;
711 		n->m_len -= count;
712 		space -= count;
713 		if (n->m_len)
714 			n->m_data += count;
715 		else
716 			n = m_free(n);
717 	} while (len > 0 && n);
718 	if (len > 0) {
719 		(void) m_free(m);
720 		goto bad;
721 	}
722 	m->m_next = n;
723 	return (m);
724 bad:
725 	m_freem(n);
726 	MPFail++;
727 	return (0);
728 }
729 
730 /*
731  * Like m_pullup(), except a new mbuf is always allocated, and we allow
732  * the amount of empty space before the data in the new mbuf to be specified
733  * (in the event that the caller expects to prepend later).
734  */
735 int MSFail;
736 
737 struct mbuf *
738 m_copyup(struct mbuf *n, int len, int dstoff)
739 {
740 	struct mbuf *m;
741 	int count, space;
742 
743 	if (len > (MHLEN - dstoff))
744 		goto bad;
745 	MGET(m, M_DONTWAIT, n->m_type);
746 	if (m == NULL)
747 		goto bad;
748 	m->m_len = 0;
749 	if (n->m_flags & M_PKTHDR) {
750 		M_COPY_PKTHDR(m, n);
751 		n->m_flags &= ~M_PKTHDR;
752 	}
753 	m->m_data += dstoff;
754 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
755 	do {
756 		count = min(min(max(len, max_protohdr), space), n->m_len);
757 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
758 		    (unsigned)count);
759 		len -= count;
760 		m->m_len += count;
761 		n->m_len -= count;
762 		space -= count;
763 		if (n->m_len)
764 			n->m_data += count;
765 		else
766 			n = m_free(n);
767 	} while (len > 0 && n);
768 	if (len > 0) {
769 		(void) m_free(m);
770 		goto bad;
771 	}
772 	m->m_next = n;
773 	return (m);
774  bad:
775 	m_freem(n);
776 	MSFail++;
777 	return (NULL);
778 }
779 
780 /*
781  * Partition an mbuf chain in two pieces, returning the tail --
782  * all but the first len0 bytes.  In case of failure, it returns NULL and
783  * attempts to restore the chain to its original state.
784  */
785 struct mbuf *
786 m_split(m0, len0, wait)
787 	struct mbuf *m0;
788 	int len0, wait;
789 {
790 	struct mbuf *m, *n;
791 	unsigned len = len0, remain, len_save;
792 
793 	for (m = m0; m && len > m->m_len; m = m->m_next)
794 		len -= m->m_len;
795 	if (m == 0)
796 		return (0);
797 	remain = m->m_len - len;
798 	if (m0->m_flags & M_PKTHDR) {
799 		MGETHDR(n, wait, m0->m_type);
800 		if (n == 0)
801 			return (0);
802 		n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
803 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
804 		len_save = m0->m_pkthdr.len;
805 		m0->m_pkthdr.len = len0;
806 		if (m->m_flags & M_EXT)
807 			goto extpacket;
808 		if (remain > MHLEN) {
809 			/* m can't be the lead packet */
810 			MH_ALIGN(n, 0);
811 			n->m_next = m_split(m, len, wait);
812 			if (n->m_next == 0) {
813 				(void) m_free(n);
814 				m0->m_pkthdr.len = len_save;
815 				return (0);
816 			} else
817 				return (n);
818 		} else
819 			MH_ALIGN(n, remain);
820 	} else if (remain == 0) {
821 		n = m->m_next;
822 		m->m_next = 0;
823 		return (n);
824 	} else {
825 		MGET(n, wait, m->m_type);
826 		if (n == 0)
827 			return (0);
828 		M_ALIGN(n, remain);
829 	}
830 extpacket:
831 	if (m->m_flags & M_EXT) {
832 		n->m_ext = m->m_ext;
833 		MCLADDREFERENCE(m, n);
834 		n->m_data = m->m_data + len;
835 	} else {
836 		memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + len, remain);
837 	}
838 	n->m_len = remain;
839 	m->m_len = len;
840 	n->m_next = m->m_next;
841 	m->m_next = 0;
842 	return (n);
843 }
844 /*
845  * Routine to copy from device local memory into mbufs.
846  */
847 struct mbuf *
848 m_devget(buf, totlen, off0, ifp, copy)
849 	char *buf;
850 	int totlen, off0;
851 	struct ifnet *ifp;
852 	void (*copy) __P((const void *from, void *to, size_t len));
853 {
854 	struct mbuf *m;
855 	struct mbuf *top = 0, **mp = &top;
856 	int off = off0, len;
857 	char *cp;
858 	char *epkt;
859 
860 	cp = buf;
861 	epkt = cp + totlen;
862 	if (off) {
863 		/*
864 		 * If 'off' is non-zero, packet is trailer-encapsulated,
865 		 * so we have to skip the type and length fields.
866 		 */
867 		cp += off + 2 * sizeof(u_int16_t);
868 		totlen -= 2 * sizeof(u_int16_t);
869 	}
870 	MGETHDR(m, M_DONTWAIT, MT_DATA);
871 	if (m == 0)
872 		return (0);
873 	m->m_pkthdr.rcvif = ifp;
874 	m->m_pkthdr.len = totlen;
875 	m->m_len = MHLEN;
876 
877 	while (totlen > 0) {
878 		if (top) {
879 			MGET(m, M_DONTWAIT, MT_DATA);
880 			if (m == 0) {
881 				m_freem(top);
882 				return (0);
883 			}
884 			m->m_len = MLEN;
885 		}
886 		len = min(totlen, epkt - cp);
887 		if (len >= MINCLSIZE) {
888 			MCLGET(m, M_DONTWAIT);
889 			if ((m->m_flags & M_EXT) == 0) {
890 				m_free(m);
891 				m_freem(top);
892 				return (0);
893 			}
894 			m->m_len = len = min(len, MCLBYTES);
895 		} else {
896 			/*
897 			 * Place initial small packet/header at end of mbuf.
898 			 */
899 			if (len < m->m_len) {
900 				if (top == 0 && len + max_linkhdr <= m->m_len)
901 					m->m_data += max_linkhdr;
902 				m->m_len = len;
903 			} else
904 				len = m->m_len;
905 		}
906 		if (copy)
907 			copy(cp, mtod(m, caddr_t), (size_t)len);
908 		else
909 			memcpy(mtod(m, caddr_t), cp, (size_t)len);
910 		cp += len;
911 		*mp = m;
912 		mp = &m->m_next;
913 		totlen -= len;
914 		if (cp == epkt)
915 			cp = buf;
916 	}
917 	return (top);
918 }
919 
920 /*
921  * Copy data from a buffer back into the indicated mbuf chain,
922  * starting "off" bytes from the beginning, extending the mbuf
923  * chain if necessary.
924  */
925 void
926 m_copyback(m0, off, len, cp)
927 	struct	mbuf *m0;
928 	int off;
929 	int len;
930 	caddr_t cp;
931 {
932 	int mlen;
933 	struct mbuf *m = m0, *n;
934 	int totlen = 0;
935 
936 	if (m0 == 0)
937 		return;
938 	while (off > (mlen = m->m_len)) {
939 		off -= mlen;
940 		totlen += mlen;
941 		if (m->m_next == 0) {
942 			n = m_getclr(M_DONTWAIT, m->m_type);
943 			if (n == 0)
944 				goto out;
945 			n->m_len = min(MLEN, len + off);
946 			m->m_next = n;
947 		}
948 		m = m->m_next;
949 	}
950 	while (len > 0) {
951 		mlen = min (m->m_len - off, len);
952 		memcpy(mtod(m, caddr_t) + off, cp, (unsigned)mlen);
953 		cp += mlen;
954 		len -= mlen;
955 		mlen += off;
956 		off = 0;
957 		totlen += mlen;
958 		if (len == 0)
959 			break;
960 		if (m->m_next == 0) {
961 			n = m_get(M_DONTWAIT, m->m_type);
962 			if (n == 0)
963 				break;
964 			n->m_len = min(MLEN, len);
965 			m->m_next = n;
966 		}
967 		m = m->m_next;
968 	}
969 out:	if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
970 		m->m_pkthdr.len = totlen;
971 }
972