xref: /freebsd/sys/kern/uipc_mbuf.c (revision 9c8b8baa)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)uipc_mbuf.c	8.2 (Berkeley) 1/4/94
34  *	$Id: uipc_mbuf.c,v 1.39 1999/04/12 10:07:15 des Exp $
35  */
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/kernel.h>
42 #include <sys/sysctl.h>
43 #include <sys/domain.h>
44 #include <sys/protosw.h>
45 
46 #include <vm/vm.h>
47 #include <vm/vm_kern.h>
48 #include <vm/vm_extern.h>
49 
50 static void mbinit __P((void *));
51 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbinit, NULL)
52 
53 struct mbuf *mbutl;
54 char	*mclrefcnt;
55 struct mbstat mbstat;
56 struct mbuf *mmbfree;
57 union mcluster *mclfree;
58 int	max_linkhdr;
59 int	max_protohdr;
60 int	max_hdr;
61 int	max_datalen;
62 
63 SYSCTL_DECL(_kern_ipc);
64 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW,
65 	   &max_linkhdr, 0, "");
66 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW,
67 	   &max_protohdr, 0, "");
68 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, "");
69 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW,
70 	   &max_datalen, 0, "");
71 SYSCTL_STRUCT(_kern_ipc, KIPC_MBSTAT, mbstat, CTLFLAG_RW, &mbstat, mbstat, "");
72 
73 static void	m_reclaim __P((void));
74 
75 /* "number of clusters of pages" */
76 #define NCL_INIT	1
77 
78 #define NMB_INIT	16
79 
80 /* ARGSUSED*/
81 static void
82 mbinit(dummy)
83 	void *dummy;
84 {
85 	int s;
86 
87 	mmbfree = NULL; mclfree = NULL;
88 	mbstat.m_msize = MSIZE;
89 	mbstat.m_mclbytes = MCLBYTES;
90 	mbstat.m_minclsize = MINCLSIZE;
91 	mbstat.m_mlen = MLEN;
92 	mbstat.m_mhlen = MHLEN;
93 
94 	s = splimp();
95 	if (m_mballoc(NMB_INIT, M_DONTWAIT) == 0)
96 		goto bad;
97 #if MCLBYTES <= PAGE_SIZE
98 	if (m_clalloc(NCL_INIT, M_DONTWAIT) == 0)
99 		goto bad;
100 #else
101 	/* It's OK to call contigmalloc in this context. */
102 	if (m_clalloc(16, M_WAIT) == 0)
103 		goto bad;
104 #endif
105 	splx(s);
106 	return;
107 bad:
108 	panic("mbinit");
109 }
110 
111 /*
112  * Allocate at least nmb mbufs and place on mbuf free list.
113  * Must be called at splimp.
114  */
115 /* ARGSUSED */
116 int
117 m_mballoc(nmb, how)
118 	register int nmb;
119 	int how;
120 {
121 	register caddr_t p;
122 	register int i;
123 	int nbytes;
124 
125 	/* Once we run out of map space, it will be impossible to get
126 	 * any more (nothing is ever freed back to the map) (XXX which
127 	 * is dumb). (however you are not dead as m_reclaim might
128 	 * still be able to free a substantial amount of space).
129 	 */
130 	if (mb_map_full)
131 		return (0);
132 
133 	nbytes = round_page(nmb * MSIZE);
134 	p = (caddr_t)kmem_malloc(mb_map, nbytes, M_NOWAIT);
135 	if (p == 0 && how == M_WAIT) {
136 		mbstat.m_wait++;
137 		p = (caddr_t)kmem_malloc(mb_map, nbytes, M_WAITOK);
138 	}
139 
140 	/*
141 	 * Either the map is now full, or `how' is M_NOWAIT and there
142 	 * are no pages left.
143 	 */
144 	if (p == NULL)
145 		return (0);
146 
147 	nmb = nbytes / MSIZE;
148 	for (i = 0; i < nmb; i++) {
149 		((struct mbuf *)p)->m_next = mmbfree;
150 		mmbfree = (struct mbuf *)p;
151 		p += MSIZE;
152 	}
153 	mbstat.m_mbufs += nmb;
154 	return (1);
155 }
156 
157 #if MCLBYTES > PAGE_SIZE
158 static int i_want_my_mcl;
159 
160 static void
161 kproc_mclalloc(void)
162 {
163 	int status;
164 
165 	while (1) {
166 		tsleep(&i_want_my_mcl, PVM, "mclalloc", 0);
167 
168 		for (; i_want_my_mcl; i_want_my_mcl--) {
169 			if (m_clalloc(1, M_WAIT) == 0)
170 				printf("m_clalloc failed even in process context!\n");
171 		}
172 	}
173 }
174 
175 static struct proc *mclallocproc;
176 static struct kproc_desc mclalloc_kp = {
177 	"mclalloc",
178 	kproc_mclalloc,
179 	&mclallocproc
180 };
181 SYSINIT(mclallocproc, SI_SUB_KTHREAD_UPDATE, SI_ORDER_ANY, kproc_start,
182 	   &mclalloc_kp);
183 #endif
184 
185 /*
186  * Allocate some number of mbuf clusters
187  * and place on cluster free list.
188  * Must be called at splimp.
189  */
190 /* ARGSUSED */
191 int
192 m_clalloc(ncl, how)
193 	register int ncl;
194 	int how;
195 {
196 	register caddr_t p;
197 	register int i;
198 	int npg;
199 
200 	/*
201 	 * Once we run out of map space, it will be impossible
202 	 * to get any more (nothing is ever freed back to the
203 	 * map).
204 	 */
205 	if (mb_map_full) {
206 		mbstat.m_drops++;
207 		return (0);
208 	}
209 
210 #if MCLBYTES > PAGE_SIZE
211 	if (how != M_WAIT) {
212 		i_want_my_mcl += ncl;
213 		wakeup(&i_want_my_mcl);
214 		mbstat.m_wait++;
215 		p = 0;
216 	} else {
217 		p = contigmalloc1(MCLBYTES * ncl, M_DEVBUF, M_WAITOK, 0ul,
218 				  ~0ul, PAGE_SIZE, 0, mb_map);
219 	}
220 #else
221 	npg = ncl;
222 	p = (caddr_t)kmem_malloc(mb_map, ctob(npg),
223 				 how != M_WAIT ? M_NOWAIT : M_WAITOK);
224 	ncl = ncl * PAGE_SIZE / MCLBYTES;
225 #endif
226 	/*
227 	 * Either the map is now full, or `how' is M_NOWAIT and there
228 	 * are no pages left.
229 	 */
230 	if (p == NULL) {
231 		mbstat.m_drops++;
232 		return (0);
233 	}
234 
235 	for (i = 0; i < ncl; i++) {
236 		((union mcluster *)p)->mcl_next = mclfree;
237 		mclfree = (union mcluster *)p;
238 		p += MCLBYTES;
239 		mbstat.m_clfree++;
240 	}
241 	mbstat.m_clusters += ncl;
242 	return (1);
243 }
244 
245 /*
246  * When MGET fails, ask protocols to free space when short of memory,
247  * then re-attempt to allocate an mbuf.
248  */
249 struct mbuf *
250 m_retry(i, t)
251 	int i, t;
252 {
253 	register struct mbuf *m;
254 
255 	/*
256 	 * Must only do the reclaim if not in an interrupt context.
257 	 */
258 	if (i == M_WAIT)
259 		m_reclaim();
260 #define m_retry(i, t)	(struct mbuf *)0
261 	MGET(m, i, t);
262 #undef m_retry
263 	if (m != NULL) {
264 		mbstat.m_wait++;
265 	} else {
266 		if (i == M_DONTWAIT)
267 			mbstat.m_drops++;
268 		else
269 			panic("Out of mbuf clusters");
270 	}
271 	return (m);
272 }
273 
274 /*
275  * As above; retry an MGETHDR.
276  */
277 struct mbuf *
278 m_retryhdr(i, t)
279 	int i, t;
280 {
281 	register struct mbuf *m;
282 
283 	/*
284 	 * Must only do the reclaim if not in an interrupt context.
285 	 */
286 	if (i == M_WAIT)
287 		m_reclaim();
288 #define m_retryhdr(i, t) (struct mbuf *)0
289 	MGETHDR(m, i, t);
290 #undef m_retryhdr
291 	if (m != NULL) {
292 		mbstat.m_wait++;
293 	} else {
294 		if (i == M_DONTWAIT)
295 			mbstat.m_drops++;
296 		else
297 			panic("Out of mbuf clusters");
298 	}
299 	return (m);
300 }
301 
302 static void
303 m_reclaim()
304 {
305 	register struct domain *dp;
306 	register struct protosw *pr;
307 	int s = splimp();
308 
309 	for (dp = domains; dp; dp = dp->dom_next)
310 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
311 			if (pr->pr_drain)
312 				(*pr->pr_drain)();
313 	splx(s);
314 	mbstat.m_drain++;
315 }
316 
317 /*
318  * Space allocation routines.
319  * These are also available as macros
320  * for critical paths.
321  */
322 struct mbuf *
323 m_get(how, type)
324 	int how, type;
325 {
326 	register struct mbuf *m;
327 
328 	MGET(m, how, type);
329 	return (m);
330 }
331 
332 struct mbuf *
333 m_gethdr(how, type)
334 	int how, type;
335 {
336 	register struct mbuf *m;
337 
338 	MGETHDR(m, how, type);
339 	return (m);
340 }
341 
342 struct mbuf *
343 m_getclr(how, type)
344 	int how, type;
345 {
346 	register struct mbuf *m;
347 
348 	MGET(m, how, type);
349 	if (m == 0)
350 		return (0);
351 	bzero(mtod(m, caddr_t), MLEN);
352 	return (m);
353 }
354 
355 struct mbuf *
356 m_free(m)
357 	struct mbuf *m;
358 {
359 	register struct mbuf *n;
360 
361 	MFREE(m, n);
362 	return (n);
363 }
364 
365 void
366 m_freem(m)
367 	register struct mbuf *m;
368 {
369 	register struct mbuf *n;
370 
371 	if (m == NULL)
372 		return;
373 	do {
374 		MFREE(m, n);
375 		m = n;
376 	} while (m);
377 }
378 
379 /*
380  * Mbuffer utility routines.
381  */
382 
383 /*
384  * Lesser-used path for M_PREPEND:
385  * allocate new mbuf to prepend to chain,
386  * copy junk along.
387  */
388 struct mbuf *
389 m_prepend(m, len, how)
390 	register struct mbuf *m;
391 	int len, how;
392 {
393 	struct mbuf *mn;
394 
395 	MGET(mn, how, m->m_type);
396 	if (mn == (struct mbuf *)NULL) {
397 		m_freem(m);
398 		return ((struct mbuf *)NULL);
399 	}
400 	if (m->m_flags & M_PKTHDR) {
401 		M_COPY_PKTHDR(mn, m);
402 		m->m_flags &= ~M_PKTHDR;
403 	}
404 	mn->m_next = m;
405 	m = mn;
406 	if (len < MHLEN)
407 		MH_ALIGN(m, len);
408 	m->m_len = len;
409 	return (m);
410 }
411 
412 /*
413  * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
414  * continuing for "len" bytes.  If len is M_COPYALL, copy to end of mbuf.
415  * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
416  */
417 #define MCFail (mbstat.m_mcfail)
418 
419 struct mbuf *
420 m_copym(m, off0, len, wait)
421 	register struct mbuf *m;
422 	int off0, wait;
423 	register int len;
424 {
425 	register struct mbuf *n, **np;
426 	register int off = off0;
427 	struct mbuf *top;
428 	int copyhdr = 0;
429 
430 	if (off < 0 || len < 0)
431 		panic("m_copym");
432 	if (off == 0 && m->m_flags & M_PKTHDR)
433 		copyhdr = 1;
434 	while (off > 0) {
435 		if (m == 0)
436 			panic("m_copym");
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");
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 			n->m_data = m->m_data + off;
465 			if(!m->m_ext.ext_ref)
466 				mclrefcnt[mtocl(m->m_ext.ext_buf)]++;
467 			else
468 				(*(m->m_ext.ext_ref))(m->m_ext.ext_buf,
469 							m->m_ext.ext_size);
470 			n->m_ext = m->m_ext;
471 			n->m_flags |= M_EXT;
472 		} else
473 			bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
474 			    (unsigned)n->m_len);
475 		if (len != M_COPYALL)
476 			len -= n->m_len;
477 		off = 0;
478 		m = m->m_next;
479 		np = &n->m_next;
480 	}
481 	if (top == 0)
482 		MCFail++;
483 	return (top);
484 nospace:
485 	m_freem(top);
486 	MCFail++;
487 	return (0);
488 }
489 
490 /*
491  * Copy an entire packet, including header (which must be present).
492  * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
493  */
494 struct mbuf *
495 m_copypacket(m, how)
496 	struct mbuf *m;
497 	int how;
498 {
499 	struct mbuf *top, *n, *o;
500 
501 	MGET(n, how, m->m_type);
502 	top = n;
503 	if (!n)
504 		goto nospace;
505 
506 	M_COPY_PKTHDR(n, m);
507 	n->m_len = m->m_len;
508 	if (m->m_flags & M_EXT) {
509 		n->m_data = m->m_data;
510 		if(!m->m_ext.ext_ref)
511 			mclrefcnt[mtocl(m->m_ext.ext_buf)]++;
512 		else
513 			(*(m->m_ext.ext_ref))(m->m_ext.ext_buf,
514 						m->m_ext.ext_size);
515 		n->m_ext = m->m_ext;
516 		n->m_flags |= M_EXT;
517 	} else {
518 		bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
519 	}
520 
521 	m = m->m_next;
522 	while (m) {
523 		MGET(o, how, m->m_type);
524 		if (!o)
525 			goto nospace;
526 
527 		n->m_next = o;
528 		n = n->m_next;
529 
530 		n->m_len = m->m_len;
531 		if (m->m_flags & M_EXT) {
532 			n->m_data = m->m_data;
533 			if(!m->m_ext.ext_ref)
534 				mclrefcnt[mtocl(m->m_ext.ext_buf)]++;
535 			else
536 				(*(m->m_ext.ext_ref))(m->m_ext.ext_buf,
537 							m->m_ext.ext_size);
538 			n->m_ext = m->m_ext;
539 			n->m_flags |= M_EXT;
540 		} else {
541 			bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
542 		}
543 
544 		m = m->m_next;
545 	}
546 	return top;
547 nospace:
548 	m_freem(top);
549 	MCFail++;
550 	return 0;
551 }
552 
553 /*
554  * Copy data from an mbuf chain starting "off" bytes from the beginning,
555  * continuing for "len" bytes, into the indicated buffer.
556  */
557 void
558 m_copydata(m, off, len, cp)
559 	register struct mbuf *m;
560 	register int off;
561 	register int len;
562 	caddr_t cp;
563 {
564 	register unsigned count;
565 
566 	if (off < 0 || len < 0)
567 		panic("m_copydata");
568 	while (off > 0) {
569 		if (m == 0)
570 			panic("m_copydata");
571 		if (off < m->m_len)
572 			break;
573 		off -= m->m_len;
574 		m = m->m_next;
575 	}
576 	while (len > 0) {
577 		if (m == 0)
578 			panic("m_copydata");
579 		count = min(m->m_len - off, len);
580 		bcopy(mtod(m, caddr_t) + off, cp, count);
581 		len -= count;
582 		cp += count;
583 		off = 0;
584 		m = m->m_next;
585 	}
586 }
587 
588 /*
589  * Concatenate mbuf chain n to m.
590  * Both chains must be of the same type (e.g. MT_DATA).
591  * Any m_pkthdr is not updated.
592  */
593 void
594 m_cat(m, n)
595 	register struct mbuf *m, *n;
596 {
597 	while (m->m_next)
598 		m = m->m_next;
599 	while (n) {
600 		if (m->m_flags & M_EXT ||
601 		    m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
602 			/* just join the two chains */
603 			m->m_next = n;
604 			return;
605 		}
606 		/* splat the data from one into the other */
607 		bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
608 		    (u_int)n->m_len);
609 		m->m_len += n->m_len;
610 		n = m_free(n);
611 	}
612 }
613 
614 void
615 m_adj(mp, req_len)
616 	struct mbuf *mp;
617 	int req_len;
618 {
619 	register int len = req_len;
620 	register struct mbuf *m;
621 	register int count;
622 
623 	if ((m = mp) == NULL)
624 		return;
625 	if (len >= 0) {
626 		/*
627 		 * Trim from head.
628 		 */
629 		while (m != NULL && len > 0) {
630 			if (m->m_len <= len) {
631 				len -= m->m_len;
632 				m->m_len = 0;
633 				m = m->m_next;
634 			} else {
635 				m->m_len -= len;
636 				m->m_data += len;
637 				len = 0;
638 			}
639 		}
640 		m = mp;
641 		if (mp->m_flags & M_PKTHDR)
642 			m->m_pkthdr.len -= (req_len - len);
643 	} else {
644 		/*
645 		 * Trim from tail.  Scan the mbuf chain,
646 		 * calculating its length and finding the last mbuf.
647 		 * If the adjustment only affects this mbuf, then just
648 		 * adjust and return.  Otherwise, rescan and truncate
649 		 * after the remaining size.
650 		 */
651 		len = -len;
652 		count = 0;
653 		for (;;) {
654 			count += m->m_len;
655 			if (m->m_next == (struct mbuf *)0)
656 				break;
657 			m = m->m_next;
658 		}
659 		if (m->m_len >= len) {
660 			m->m_len -= len;
661 			if (mp->m_flags & M_PKTHDR)
662 				mp->m_pkthdr.len -= len;
663 			return;
664 		}
665 		count -= len;
666 		if (count < 0)
667 			count = 0;
668 		/*
669 		 * Correct length for chain is "count".
670 		 * Find the mbuf with last data, adjust its length,
671 		 * and toss data from remaining mbufs on chain.
672 		 */
673 		m = mp;
674 		if (m->m_flags & M_PKTHDR)
675 			m->m_pkthdr.len = count;
676 		for (; m; m = m->m_next) {
677 			if (m->m_len >= count) {
678 				m->m_len = count;
679 				break;
680 			}
681 			count -= m->m_len;
682 		}
683 		while (m->m_next)
684 			(m = m->m_next) ->m_len = 0;
685 	}
686 }
687 
688 /*
689  * Rearange an mbuf chain so that len bytes are contiguous
690  * and in the data area of an mbuf (so that mtod and dtom
691  * will work for a structure of size len).  Returns the resulting
692  * mbuf chain on success, frees it and returns null on failure.
693  * If there is room, it will add up to max_protohdr-len extra bytes to the
694  * contiguous region in an attempt to avoid being called next time.
695  */
696 #define MPFail (mbstat.m_mpfail)
697 
698 struct mbuf *
699 m_pullup(n, len)
700 	register struct mbuf *n;
701 	int len;
702 {
703 	register struct mbuf *m;
704 	register int count;
705 	int space;
706 
707 	/*
708 	 * If first mbuf has no cluster, and has room for len bytes
709 	 * without shifting current data, pullup into it,
710 	 * otherwise allocate a new mbuf to prepend to the chain.
711 	 */
712 	if ((n->m_flags & M_EXT) == 0 &&
713 	    n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
714 		if (n->m_len >= len)
715 			return (n);
716 		m = n;
717 		n = n->m_next;
718 		len -= m->m_len;
719 	} else {
720 		if (len > MHLEN)
721 			goto bad;
722 		MGET(m, M_DONTWAIT, n->m_type);
723 		if (m == 0)
724 			goto bad;
725 		m->m_len = 0;
726 		if (n->m_flags & M_PKTHDR) {
727 			M_COPY_PKTHDR(m, n);
728 			n->m_flags &= ~M_PKTHDR;
729 		}
730 	}
731 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
732 	do {
733 		count = min(min(max(len, max_protohdr), space), n->m_len);
734 		bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
735 		  (unsigned)count);
736 		len -= count;
737 		m->m_len += count;
738 		n->m_len -= count;
739 		space -= count;
740 		if (n->m_len)
741 			n->m_data += count;
742 		else
743 			n = m_free(n);
744 	} while (len > 0 && n);
745 	if (len > 0) {
746 		(void) m_free(m);
747 		goto bad;
748 	}
749 	m->m_next = n;
750 	return (m);
751 bad:
752 	m_freem(n);
753 	MPFail++;
754 	return (0);
755 }
756 
757 /*
758  * Partition an mbuf chain in two pieces, returning the tail --
759  * all but the first len0 bytes.  In case of failure, it returns NULL and
760  * attempts to restore the chain to its original state.
761  */
762 struct mbuf *
763 m_split(m0, len0, wait)
764 	register struct mbuf *m0;
765 	int len0, wait;
766 {
767 	register struct mbuf *m, *n;
768 	unsigned len = len0, remain;
769 
770 	for (m = m0; m && len > m->m_len; m = m->m_next)
771 		len -= m->m_len;
772 	if (m == 0)
773 		return (0);
774 	remain = m->m_len - len;
775 	if (m0->m_flags & M_PKTHDR) {
776 		MGETHDR(n, wait, m0->m_type);
777 		if (n == 0)
778 			return (0);
779 		n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
780 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
781 		m0->m_pkthdr.len = len0;
782 		if (m->m_flags & M_EXT)
783 			goto extpacket;
784 		if (remain > MHLEN) {
785 			/* m can't be the lead packet */
786 			MH_ALIGN(n, 0);
787 			n->m_next = m_split(m, len, wait);
788 			if (n->m_next == 0) {
789 				(void) m_free(n);
790 				return (0);
791 			} else
792 				return (n);
793 		} else
794 			MH_ALIGN(n, remain);
795 	} else if (remain == 0) {
796 		n = m->m_next;
797 		m->m_next = 0;
798 		return (n);
799 	} else {
800 		MGET(n, wait, m->m_type);
801 		if (n == 0)
802 			return (0);
803 		M_ALIGN(n, remain);
804 	}
805 extpacket:
806 	if (m->m_flags & M_EXT) {
807 		n->m_flags |= M_EXT;
808 		n->m_ext = m->m_ext;
809 		if(!m->m_ext.ext_ref)
810 			mclrefcnt[mtocl(m->m_ext.ext_buf)]++;
811 		else
812 			(*(m->m_ext.ext_ref))(m->m_ext.ext_buf,
813 						m->m_ext.ext_size);
814 		m->m_ext.ext_size = 0; /* For Accounting XXXXXX danger */
815 		n->m_data = m->m_data + len;
816 	} else {
817 		bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
818 	}
819 	n->m_len = remain;
820 	m->m_len = len;
821 	n->m_next = m->m_next;
822 	m->m_next = 0;
823 	return (n);
824 }
825 /*
826  * Routine to copy from device local memory into mbufs.
827  */
828 struct mbuf *
829 m_devget(buf, totlen, off0, ifp, copy)
830 	char *buf;
831 	int totlen, off0;
832 	struct ifnet *ifp;
833 	void (*copy) __P((char *from, caddr_t to, u_int len));
834 {
835 	register struct mbuf *m;
836 	struct mbuf *top = 0, **mp = &top;
837 	register int off = off0, len;
838 	register char *cp;
839 	char *epkt;
840 
841 	cp = buf;
842 	epkt = cp + totlen;
843 	if (off) {
844 		cp += off + 2 * sizeof(u_short);
845 		totlen -= 2 * sizeof(u_short);
846 	}
847 	MGETHDR(m, M_DONTWAIT, MT_DATA);
848 	if (m == 0)
849 		return (0);
850 	m->m_pkthdr.rcvif = ifp;
851 	m->m_pkthdr.len = totlen;
852 	m->m_len = MHLEN;
853 
854 	while (totlen > 0) {
855 		if (top) {
856 			MGET(m, M_DONTWAIT, MT_DATA);
857 			if (m == 0) {
858 				m_freem(top);
859 				return (0);
860 			}
861 			m->m_len = MLEN;
862 		}
863 		len = min(totlen, epkt - cp);
864 		if (len >= MINCLSIZE) {
865 			MCLGET(m, M_DONTWAIT);
866 			if (m->m_flags & M_EXT)
867 				m->m_len = len = min(len, MCLBYTES);
868 			else
869 				len = m->m_len;
870 		} else {
871 			/*
872 			 * Place initial small packet/header at end of mbuf.
873 			 */
874 			if (len < m->m_len) {
875 				if (top == 0 && len + max_linkhdr <= m->m_len)
876 					m->m_data += max_linkhdr;
877 				m->m_len = len;
878 			} else
879 				len = m->m_len;
880 		}
881 		if (copy)
882 			copy(cp, mtod(m, caddr_t), (unsigned)len);
883 		else
884 			bcopy(cp, mtod(m, caddr_t), (unsigned)len);
885 		cp += len;
886 		*mp = m;
887 		mp = &m->m_next;
888 		totlen -= len;
889 		if (cp == epkt)
890 			cp = buf;
891 	}
892 	return (top);
893 }
894 
895 /*
896  * Copy data from a buffer back into the indicated mbuf chain,
897  * starting "off" bytes from the beginning, extending the mbuf
898  * chain if necessary.
899  */
900 void
901 m_copyback(m0, off, len, cp)
902 	struct	mbuf *m0;
903 	register int off;
904 	register int len;
905 	caddr_t cp;
906 {
907 	register int mlen;
908 	register struct mbuf *m = m0, *n;
909 	int totlen = 0;
910 
911 	if (m0 == 0)
912 		return;
913 	while (off > (mlen = m->m_len)) {
914 		off -= mlen;
915 		totlen += mlen;
916 		if (m->m_next == 0) {
917 			n = m_getclr(M_DONTWAIT, m->m_type);
918 			if (n == 0)
919 				goto out;
920 			n->m_len = min(MLEN, len + off);
921 			m->m_next = n;
922 		}
923 		m = m->m_next;
924 	}
925 	while (len > 0) {
926 		mlen = min (m->m_len - off, len);
927 		bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen);
928 		cp += mlen;
929 		len -= mlen;
930 		mlen += off;
931 		off = 0;
932 		totlen += mlen;
933 		if (len == 0)
934 			break;
935 		if (m->m_next == 0) {
936 			n = m_get(M_DONTWAIT, m->m_type);
937 			if (n == 0)
938 				break;
939 			n->m_len = min(MLEN, len);
940 			m->m_next = n;
941 		}
942 		m = m->m_next;
943 	}
944 out:	if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
945 		m->m_pkthdr.len = totlen;
946 }
947