Home | History | Annotate | Line # | Download | only in kern
uipc_mbuf.c revision 1.42
      1 /*	$NetBSD: uipc_mbuf.c,v 1.42 1999/04/26 22:04:28 thorpej Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1999 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
      9  * NASA Ames Research Center.
     10  *
     11  * Redistribution and use in source and binary forms, with or without
     12  * modification, are permitted provided that the following conditions
     13  * are met:
     14  * 1. Redistributions of source code must retain the above copyright
     15  *    notice, this list of conditions and the following disclaimer.
     16  * 2. Redistributions in binary form must reproduce the above copyright
     17  *    notice, this list of conditions and the following disclaimer in the
     18  *    documentation and/or other materials provided with the distribution.
     19  * 3. All advertising materials mentioning features or use of this software
     20  *    must display the following acknowledgement:
     21  *	This product includes software developed by the NetBSD
     22  *	Foundation, Inc. and its contributors.
     23  * 4. Neither the name of The NetBSD Foundation nor the names of its
     24  *    contributors may be used to endorse or promote products derived
     25  *    from this software without specific prior written permission.
     26  *
     27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     37  * POSSIBILITY OF SUCH DAMAGE.
     38  */
     39 
     40 /*
     41  * Copyright (c) 1982, 1986, 1988, 1991, 1993
     42  *	The Regents of the University of California.  All rights reserved.
     43  *
     44  * Redistribution and use in source and binary forms, with or without
     45  * modification, are permitted provided that the following conditions
     46  * are met:
     47  * 1. Redistributions of source code must retain the above copyright
     48  *    notice, this list of conditions and the following disclaimer.
     49  * 2. Redistributions in binary form must reproduce the above copyright
     50  *    notice, this list of conditions and the following disclaimer in the
     51  *    documentation and/or other materials provided with the distribution.
     52  * 3. All advertising materials mentioning features or use of this software
     53  *    must display the following acknowledgement:
     54  *	This product includes software developed by the University of
     55  *	California, Berkeley and its contributors.
     56  * 4. Neither the name of the University nor the names of its contributors
     57  *    may be used to endorse or promote products derived from this software
     58  *    without specific prior written permission.
     59  *
     60  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     61  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     62  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     63  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     64  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     65  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     66  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     67  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     68  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     69  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     70  * SUCH DAMAGE.
     71  *
     72  *	@(#)uipc_mbuf.c	8.4 (Berkeley) 2/14/95
     73  */
     74 
     75 #include <sys/param.h>
     76 #include <sys/systm.h>
     77 #include <sys/proc.h>
     78 #include <sys/malloc.h>
     79 #include <sys/map.h>
     80 #define MBTYPES
     81 #include <sys/mbuf.h>
     82 #include <sys/kernel.h>
     83 #include <sys/syslog.h>
     84 #include <sys/domain.h>
     85 #include <sys/protosw.h>
     86 #include <sys/pool.h>
     87 #include <sys/socket.h>
     88 #include <net/if.h>
     89 
     90 #include <vm/vm.h>
     91 #include <vm/vm_kern.h>
     92 
     93 #include <uvm/uvm_extern.h>
     94 
     95 #include <sys/sysctl.h>
     96 
     97 struct	pool mbpool;		/* mbuf pool */
     98 struct	pool mclpool;		/* mbuf cluster pool */
     99 
    100 struct mbstat mbstat;
    101 int	max_linkhdr;
    102 int	max_protohdr;
    103 int	max_hdr;
    104 int	max_datalen;
    105 
    106 void	*mclpool_alloc __P((unsigned long, int, int));
    107 void	mclpool_release __P((void *, unsigned long, int));
    108 
    109 const char *mclpool_warnmsg =
    110     "WARNING: mclpool limit reached; increase NMBCLUSTERS";
    111 
    112 /*
    113  * Initialize the mbuf allcator.
    114  */
    115 void
    116 mbinit()
    117 {
    118 
    119 	pool_init(&mbpool, msize, 0, 0, 0, "mbpl", 0, NULL, NULL, 0);
    120 	pool_init(&mclpool, mclbytes, 0, 0, 0, "mclpl", 0, mclpool_alloc,
    121 	    mclpool_release, 0);
    122 
    123 	/*
    124 	 * Set the hard limit on the mclpool to the number of
    125 	 * mbuf clusters the kernel is to support.  Log the limit
    126 	 * reached message max once a minute.
    127 	 */
    128 	pool_sethardlimit(&mclpool, nmbclusters, mclpool_warnmsg, 60);
    129 
    130 	/*
    131 	 * Set a low water mark for both mbufs and clusters.  This should
    132 	 * help ensure that they can be allocated in a memory starvation
    133 	 * situation.  This is important for e.g. diskless systems which
    134 	 * must allocate mbufs in order for the pagedaemon to clean pages.
    135 	 */
    136 	pool_setlowat(&mbpool, mblowat);
    137 	pool_setlowat(&mclpool, mcllowat);
    138 }
    139 
    140 int
    141 sysctl_dombuf(name, namelen, oldp, oldlenp, newp, newlen)
    142 	int *name;
    143 	u_int namelen;
    144 	void *oldp;
    145 	size_t *oldlenp;
    146 	void *newp;
    147 	size_t newlen;
    148 {
    149 	int error, newval;
    150 
    151 	/* All sysctl names at this level are terminal. */
    152 	if (namelen != 1)
    153 		return (ENOTDIR);		/* overloaded */
    154 
    155 	switch (name[0]) {
    156 	case MBUF_MSIZE:
    157 		return (sysctl_rdint(oldp, oldlenp, newp, msize));
    158 	case MBUF_MCLBYTES:
    159 		return (sysctl_rdint(oldp, oldlenp, newp, mclbytes));
    160 	case MBUF_NMBCLUSTERS:
    161 		/*
    162 		 * If we have direct-mapped pool pages, we can adjust this
    163 		 * number on the fly.  If not, we're limited by the size
    164 		 * of mb_map, and cannot change this value.
    165 		 *
    166 		 * Note: we only allow the value to be increased, never
    167 		 * decreased.
    168 		 */
    169 		if (mb_map == NULL) {
    170 			newval = nmbclusters;
    171 			error = sysctl_int(oldp, oldlenp, newp, newlen,
    172 			    &newval);
    173 			if (error != 0)
    174 				return (error);
    175 			if (newp != NULL) {
    176 				if (newval >= nmbclusters) {
    177 					nmbclusters = newval;
    178 					pool_sethardlimit(&mclpool,
    179 					    nmbclusters, mclpool_warnmsg, 60);
    180 				} else
    181 					error = EINVAL;
    182 			}
    183 			return (error);
    184 		} else
    185 			return (sysctl_rdint(oldp, oldlenp, newp, nmbclusters));
    186 	case MBUF_MBLOWAT:
    187 	case MBUF_MCLLOWAT:
    188 		/* New value must be >= 0. */
    189 		newval = (name[0] == MBUF_MBLOWAT) ? mblowat : mcllowat;
    190 		error = sysctl_int(oldp, oldlenp, newp, newlen, &newval);
    191 		if (error != 0)
    192 			return (error);
    193 		if (newp != NULL) {
    194 			if (newval >= 0) {
    195 				if (name[0] == MBUF_MBLOWAT) {
    196 					mblowat = newval;
    197 					pool_setlowat(&mbpool, newval);
    198 				} else {
    199 					mcllowat = newval;
    200 					pool_setlowat(&mclpool, newval);
    201 				}
    202 			} else
    203 				error = EINVAL;
    204 		}
    205 		return (error);
    206 	default:
    207 		return (EOPNOTSUPP);
    208 	}
    209 	/* NOTREACHED */
    210 }
    211 
    212 void *
    213 mclpool_alloc(sz, flags, mtype)
    214 	unsigned long sz;
    215 	int flags;
    216 	int mtype;
    217 {
    218 	boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
    219 
    220 	return ((void *)uvm_km_alloc_poolpage1(mb_map, uvmexp.mb_object,
    221 	    waitok));
    222 }
    223 
    224 void
    225 mclpool_release(v, sz, mtype)
    226 	void *v;
    227 	unsigned long sz;
    228 	int mtype;
    229 {
    230 
    231 	uvm_km_free_poolpage1(mb_map, (vaddr_t)v);
    232 }
    233 
    234 /*
    235  * When MGET failes, ask protocols to free space when short of memory,
    236  * then re-attempt to allocate an mbuf.
    237  */
    238 struct mbuf *
    239 m_retry(i, t)
    240 	int i, t;
    241 {
    242 	struct mbuf *m;
    243 
    244 	m_reclaim(i);
    245 #define m_retry(i, t)	(struct mbuf *)0
    246 	MGET(m, i, t);
    247 #undef m_retry
    248 	if (m != NULL)
    249 		mbstat.m_wait++;
    250 	else
    251 		mbstat.m_drops++;
    252 	return (m);
    253 }
    254 
    255 /*
    256  * As above; retry an MGETHDR.
    257  */
    258 struct mbuf *
    259 m_retryhdr(i, t)
    260 	int i, t;
    261 {
    262 	struct mbuf *m;
    263 
    264 	m_reclaim(i);
    265 #define m_retryhdr(i, t) (struct mbuf *)0
    266 	MGETHDR(m, i, t);
    267 #undef m_retryhdr
    268 	if (m != NULL)
    269 		mbstat.m_wait++;
    270 	else
    271 		mbstat.m_drops++;
    272 	return (m);
    273 }
    274 
    275 void
    276 m_reclaim(how)
    277 	int how;
    278 {
    279 	struct domain *dp;
    280 	struct protosw *pr;
    281 	struct ifnet *ifp;
    282 	int s = splimp();
    283 
    284 	for (dp = domains; dp; dp = dp->dom_next)
    285 		for (pr = dp->dom_protosw;
    286 		     pr < dp->dom_protoswNPROTOSW; pr++)
    287 			if (pr->pr_drain)
    288 				(*pr->pr_drain)();
    289 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list))
    290 		if (ifp->if_drain)
    291 			(*ifp->if_drain)(ifp);
    292 	splx(s);
    293 	mbstat.m_drain++;
    294 }
    295 
    296 /*
    297  * Space allocation routines.
    298  * These are also available as macros
    299  * for critical paths.
    300  */
    301 struct mbuf *
    302 m_get(nowait, type)
    303 	int nowait, type;
    304 {
    305 	struct mbuf *m;
    306 
    307 	MGET(m, nowait, type);
    308 	return (m);
    309 }
    310 
    311 struct mbuf *
    312 m_gethdr(nowait, type)
    313 	int nowait, type;
    314 {
    315 	struct mbuf *m;
    316 
    317 	MGETHDR(m, nowait, type);
    318 	return (m);
    319 }
    320 
    321 struct mbuf *
    322 m_getclr(nowait, type)
    323 	int nowait, type;
    324 {
    325 	struct mbuf *m;
    326 
    327 	MGET(m, nowait, type);
    328 	if (m == 0)
    329 		return (0);
    330 	memset(mtod(m, caddr_t), 0, MLEN);
    331 	return (m);
    332 }
    333 
    334 struct mbuf *
    335 m_free(m)
    336 	struct mbuf *m;
    337 {
    338 	struct mbuf *n;
    339 
    340 	MFREE(m, n);
    341 	return (n);
    342 }
    343 
    344 void
    345 m_freem(m)
    346 	struct mbuf *m;
    347 {
    348 	struct mbuf *n;
    349 
    350 	if (m == NULL)
    351 		return;
    352 	do {
    353 		MFREE(m, n);
    354 		m = n;
    355 	} while (m);
    356 }
    357 
    358 /*
    359  * Mbuffer utility routines.
    360  */
    361 
    362 /*
    363  * Lesser-used path for M_PREPEND:
    364  * allocate new mbuf to prepend to chain,
    365  * copy junk along.
    366  */
    367 struct mbuf *
    368 m_prepend(m, len, how)
    369 	struct mbuf *m;
    370 	int len, how;
    371 {
    372 	struct mbuf *mn;
    373 
    374 	MGET(mn, how, m->m_type);
    375 	if (mn == (struct mbuf *)NULL) {
    376 		m_freem(m);
    377 		return ((struct mbuf *)NULL);
    378 	}
    379 	if (m->m_flags & M_PKTHDR) {
    380 		M_COPY_PKTHDR(mn, m);
    381 		m->m_flags &= ~M_PKTHDR;
    382 	}
    383 	mn->m_next = m;
    384 	m = mn;
    385 	if (len < MHLEN)
    386 		MH_ALIGN(m, len);
    387 	m->m_len = len;
    388 	return (m);
    389 }
    390 
    391 /*
    392  * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
    393  * continuing for "len" bytes.  If len is M_COPYALL, copy to end of mbuf.
    394  * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
    395  */
    396 int MCFail;
    397 
    398 struct mbuf *
    399 m_copym(m, off0, len, wait)
    400 	struct mbuf *m;
    401 	int off0, wait;
    402 	int len;
    403 {
    404 	struct mbuf *n, **np;
    405 	int off = off0;
    406 	struct mbuf *top;
    407 	int copyhdr = 0;
    408 
    409 	if (off < 0 || len < 0)
    410 		panic("m_copym");
    411 	if (off == 0 && m->m_flags & M_PKTHDR)
    412 		copyhdr = 1;
    413 	while (off > 0) {
    414 		if (m == 0)
    415 			panic("m_copym");
    416 		if (off < m->m_len)
    417 			break;
    418 		off -= m->m_len;
    419 		m = m->m_next;
    420 	}
    421 	np = &top;
    422 	top = 0;
    423 	while (len > 0) {
    424 		if (m == 0) {
    425 			if (len != M_COPYALL)
    426 				panic("m_copym");
    427 			break;
    428 		}
    429 		MGET(n, wait, m->m_type);
    430 		*np = n;
    431 		if (n == 0)
    432 			goto nospace;
    433 		if (copyhdr) {
    434 			M_COPY_PKTHDR(n, m);
    435 			if (len == M_COPYALL)
    436 				n->m_pkthdr.len -= off0;
    437 			else
    438 				n->m_pkthdr.len = len;
    439 			copyhdr = 0;
    440 		}
    441 		n->m_len = min(len, m->m_len - off);
    442 		if (m->m_flags & M_EXT) {
    443 			n->m_data = m->m_data + off;
    444 			n->m_ext = m->m_ext;
    445 			MCLADDREFERENCE(m, n);
    446 		} else
    447 			memcpy(mtod(n, caddr_t), mtod(m, caddr_t)+off,
    448 			    (unsigned)n->m_len);
    449 		if (len != M_COPYALL)
    450 			len -= n->m_len;
    451 		off = 0;
    452 		m = m->m_next;
    453 		np = &n->m_next;
    454 	}
    455 	if (top == 0)
    456 		MCFail++;
    457 	return (top);
    458 nospace:
    459 	m_freem(top);
    460 	MCFail++;
    461 	return (0);
    462 }
    463 
    464 /*
    465  * Copy an entire packet, including header (which must be present).
    466  * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
    467  */
    468 struct mbuf *
    469 m_copypacket(m, how)
    470 	struct mbuf *m;
    471 	int how;
    472 {
    473 	struct mbuf *top, *n, *o;
    474 
    475 	MGET(n, how, m->m_type);
    476 	top = n;
    477 	if (!n)
    478 		goto nospace;
    479 
    480 	M_COPY_PKTHDR(n, m);
    481 	n->m_len = m->m_len;
    482 	if (m->m_flags & M_EXT) {
    483 		n->m_data = m->m_data;
    484 		n->m_ext = m->m_ext;
    485 		MCLADDREFERENCE(m, n);
    486 	} else {
    487 		memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
    488 	}
    489 
    490 	m = m->m_next;
    491 	while (m) {
    492 		MGET(o, how, m->m_type);
    493 		if (!o)
    494 			goto nospace;
    495 
    496 		n->m_next = o;
    497 		n = n->m_next;
    498 
    499 		n->m_len = m->m_len;
    500 		if (m->m_flags & M_EXT) {
    501 			n->m_data = m->m_data;
    502 			n->m_ext = m->m_ext;
    503 			MCLADDREFERENCE(m, n);
    504 		} else {
    505 			memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
    506 		}
    507 
    508 		m = m->m_next;
    509 	}
    510 	return top;
    511 nospace:
    512 	m_freem(top);
    513 	MCFail++;
    514 	return 0;
    515 }
    516 
    517 /*
    518  * Copy data from an mbuf chain starting "off" bytes from the beginning,
    519  * continuing for "len" bytes, into the indicated buffer.
    520  */
    521 void
    522 m_copydata(m, off, len, cp)
    523 	struct mbuf *m;
    524 	int off;
    525 	int len;
    526 	caddr_t cp;
    527 {
    528 	unsigned count;
    529 
    530 	if (off < 0 || len < 0)
    531 		panic("m_copydata");
    532 	while (off > 0) {
    533 		if (m == 0)
    534 			panic("m_copydata");
    535 		if (off < m->m_len)
    536 			break;
    537 		off -= m->m_len;
    538 		m = m->m_next;
    539 	}
    540 	while (len > 0) {
    541 		if (m == 0)
    542 			panic("m_copydata");
    543 		count = min(m->m_len - off, len);
    544 		memcpy(cp, mtod(m, caddr_t) + off, count);
    545 		len -= count;
    546 		cp += count;
    547 		off = 0;
    548 		m = m->m_next;
    549 	}
    550 }
    551 
    552 /*
    553  * Concatenate mbuf chain n to m.
    554  * Both chains must be of the same type (e.g. MT_DATA).
    555  * Any m_pkthdr is not updated.
    556  */
    557 void
    558 m_cat(m, n)
    559 	struct mbuf *m, *n;
    560 {
    561 	while (m->m_next)
    562 		m = m->m_next;
    563 	while (n) {
    564 		if (m->m_flags & M_EXT ||
    565 		    m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
    566 			/* just join the two chains */
    567 			m->m_next = n;
    568 			return;
    569 		}
    570 		/* splat the data from one into the other */
    571 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
    572 		    (u_int)n->m_len);
    573 		m->m_len += n->m_len;
    574 		n = m_free(n);
    575 	}
    576 }
    577 
    578 void
    579 m_adj(mp, req_len)
    580 	struct mbuf *mp;
    581 	int req_len;
    582 {
    583 	int len = req_len;
    584 	struct mbuf *m;
    585 	int count;
    586 
    587 	if ((m = mp) == NULL)
    588 		return;
    589 	if (len >= 0) {
    590 		/*
    591 		 * Trim from head.
    592 		 */
    593 		while (m != NULL && len > 0) {
    594 			if (m->m_len <= len) {
    595 				len -= m->m_len;
    596 				m->m_len = 0;
    597 				m = m->m_next;
    598 			} else {
    599 				m->m_len -= len;
    600 				m->m_data += len;
    601 				len = 0;
    602 			}
    603 		}
    604 		m = mp;
    605 		if (mp->m_flags & M_PKTHDR)
    606 			m->m_pkthdr.len -= (req_len - len);
    607 	} else {
    608 		/*
    609 		 * Trim from tail.  Scan the mbuf chain,
    610 		 * calculating its length and finding the last mbuf.
    611 		 * If the adjustment only affects this mbuf, then just
    612 		 * adjust and return.  Otherwise, rescan and truncate
    613 		 * after the remaining size.
    614 		 */
    615 		len = -len;
    616 		count = 0;
    617 		for (;;) {
    618 			count += m->m_len;
    619 			if (m->m_next == (struct mbuf *)0)
    620 				break;
    621 			m = m->m_next;
    622 		}
    623 		if (m->m_len >= len) {
    624 			m->m_len -= len;
    625 			if (mp->m_flags & M_PKTHDR)
    626 				mp->m_pkthdr.len -= len;
    627 			return;
    628 		}
    629 		count -= len;
    630 		if (count < 0)
    631 			count = 0;
    632 		/*
    633 		 * Correct length for chain is "count".
    634 		 * Find the mbuf with last data, adjust its length,
    635 		 * and toss data from remaining mbufs on chain.
    636 		 */
    637 		m = mp;
    638 		if (m->m_flags & M_PKTHDR)
    639 			m->m_pkthdr.len = count;
    640 		for (; m; m = m->m_next) {
    641 			if (m->m_len >= count) {
    642 				m->m_len = count;
    643 				break;
    644 			}
    645 			count -= m->m_len;
    646 		}
    647 		while (m->m_next)
    648 			(m = m->m_next) ->m_len = 0;
    649 	}
    650 }
    651 
    652 /*
    653  * Rearange an mbuf chain so that len bytes are contiguous
    654  * and in the data area of an mbuf (so that mtod and dtom
    655  * will work for a structure of size len).  Returns the resulting
    656  * mbuf chain on success, frees it and returns null on failure.
    657  * If there is room, it will add up to max_protohdr-len extra bytes to the
    658  * contiguous region in an attempt to avoid being called next time.
    659  */
    660 int MPFail;
    661 
    662 struct mbuf *
    663 m_pullup(n, len)
    664 	struct mbuf *n;
    665 	int len;
    666 {
    667 	struct mbuf *m;
    668 	int count;
    669 	int space;
    670 
    671 	/*
    672 	 * If first mbuf has no cluster, and has room for len bytes
    673 	 * without shifting current data, pullup into it,
    674 	 * otherwise allocate a new mbuf to prepend to the chain.
    675 	 */
    676 	if ((n->m_flags & M_EXT) == 0 &&
    677 	    n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
    678 		if (n->m_len >= len)
    679 			return (n);
    680 		m = n;
    681 		n = n->m_next;
    682 		len -= m->m_len;
    683 	} else {
    684 		if (len > MHLEN)
    685 			goto bad;
    686 		MGET(m, M_DONTWAIT, n->m_type);
    687 		if (m == 0)
    688 			goto bad;
    689 		m->m_len = 0;
    690 		if (n->m_flags & M_PKTHDR) {
    691 			M_COPY_PKTHDR(m, n);
    692 			n->m_flags &= ~M_PKTHDR;
    693 		}
    694 	}
    695 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
    696 	do {
    697 		count = min(min(max(len, max_protohdr), space), n->m_len);
    698 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
    699 		  (unsigned)count);
    700 		len -= count;
    701 		m->m_len += count;
    702 		n->m_len -= count;
    703 		space -= count;
    704 		if (n->m_len)
    705 			n->m_data += count;
    706 		else
    707 			n = m_free(n);
    708 	} while (len > 0 && n);
    709 	if (len > 0) {
    710 		(void) m_free(m);
    711 		goto bad;
    712 	}
    713 	m->m_next = n;
    714 	return (m);
    715 bad:
    716 	m_freem(n);
    717 	MPFail++;
    718 	return (0);
    719 }
    720 
    721 /*
    722  * Partition an mbuf chain in two pieces, returning the tail --
    723  * all but the first len0 bytes.  In case of failure, it returns NULL and
    724  * attempts to restore the chain to its original state.
    725  */
    726 struct mbuf *
    727 m_split(m0, len0, wait)
    728 	struct mbuf *m0;
    729 	int len0, wait;
    730 {
    731 	struct mbuf *m, *n;
    732 	unsigned len = len0, remain, len_save;
    733 
    734 	for (m = m0; m && len > m->m_len; m = m->m_next)
    735 		len -= m->m_len;
    736 	if (m == 0)
    737 		return (0);
    738 	remain = m->m_len - len;
    739 	if (m0->m_flags & M_PKTHDR) {
    740 		MGETHDR(n, wait, m0->m_type);
    741 		if (n == 0)
    742 			return (0);
    743 		n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
    744 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
    745 		len_save = m0->m_pkthdr.len;
    746 		m0->m_pkthdr.len = len0;
    747 		if (m->m_flags & M_EXT)
    748 			goto extpacket;
    749 		if (remain > MHLEN) {
    750 			/* m can't be the lead packet */
    751 			MH_ALIGN(n, 0);
    752 			n->m_next = m_split(m, len, wait);
    753 			if (n->m_next == 0) {
    754 				(void) m_free(n);
    755 				m0->m_pkthdr.len = len_save;
    756 				return (0);
    757 			} else
    758 				return (n);
    759 		} else
    760 			MH_ALIGN(n, remain);
    761 	} else if (remain == 0) {
    762 		n = m->m_next;
    763 		m->m_next = 0;
    764 		return (n);
    765 	} else {
    766 		MGET(n, wait, m->m_type);
    767 		if (n == 0)
    768 			return (0);
    769 		M_ALIGN(n, remain);
    770 	}
    771 extpacket:
    772 	if (m->m_flags & M_EXT) {
    773 		n->m_ext = m->m_ext;
    774 		MCLADDREFERENCE(m, n);
    775 		n->m_data = m->m_data + len;
    776 	} else {
    777 		memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + len, remain);
    778 	}
    779 	n->m_len = remain;
    780 	m->m_len = len;
    781 	n->m_next = m->m_next;
    782 	m->m_next = 0;
    783 	return (n);
    784 }
    785 /*
    786  * Routine to copy from device local memory into mbufs.
    787  */
    788 struct mbuf *
    789 m_devget(buf, totlen, off0, ifp, copy)
    790 	char *buf;
    791 	int totlen, off0;
    792 	struct ifnet *ifp;
    793 	void (*copy) __P((const void *from, void *to, size_t len));
    794 {
    795 	struct mbuf *m;
    796 	struct mbuf *top = 0, **mp = &top;
    797 	int off = off0, len;
    798 	char *cp;
    799 	char *epkt;
    800 
    801 	cp = buf;
    802 	epkt = cp + totlen;
    803 	if (off) {
    804 		/*
    805 		 * If 'off' is non-zero, packet is trailer-encapsulated,
    806 		 * so we have to skip the type and length fields.
    807 		 */
    808 		cp += off + 2 * sizeof(u_int16_t);
    809 		totlen -= 2 * sizeof(u_int16_t);
    810 	}
    811 	MGETHDR(m, M_DONTWAIT, MT_DATA);
    812 	if (m == 0)
    813 		return (0);
    814 	m->m_pkthdr.rcvif = ifp;
    815 	m->m_pkthdr.len = totlen;
    816 	m->m_len = MHLEN;
    817 
    818 	while (totlen > 0) {
    819 		if (top) {
    820 			MGET(m, M_DONTWAIT, MT_DATA);
    821 			if (m == 0) {
    822 				m_freem(top);
    823 				return (0);
    824 			}
    825 			m->m_len = MLEN;
    826 		}
    827 		len = min(totlen, epkt - cp);
    828 		if (len >= MINCLSIZE) {
    829 			MCLGET(m, M_DONTWAIT);
    830 			if ((m->m_flags & M_EXT) == 0) {
    831 				m_free(m);
    832 				m_freem(top);
    833 				return (0);
    834 			}
    835 			m->m_len = len = min(len, MCLBYTES);
    836 		} else {
    837 			/*
    838 			 * Place initial small packet/header at end of mbuf.
    839 			 */
    840 			if (len < m->m_len) {
    841 				if (top == 0 && len + max_linkhdr <= m->m_len)
    842 					m->m_data += max_linkhdr;
    843 				m->m_len = len;
    844 			} else
    845 				len = m->m_len;
    846 		}
    847 		if (copy)
    848 			copy(cp, mtod(m, caddr_t), (size_t)len);
    849 		else
    850 			memcpy(mtod(m, caddr_t), cp, (size_t)len);
    851 		cp += len;
    852 		*mp = m;
    853 		mp = &m->m_next;
    854 		totlen -= len;
    855 		if (cp == epkt)
    856 			cp = buf;
    857 	}
    858 	return (top);
    859 }
    860 
    861 /*
    862  * Copy data from a buffer back into the indicated mbuf chain,
    863  * starting "off" bytes from the beginning, extending the mbuf
    864  * chain if necessary.
    865  */
    866 void
    867 m_copyback(m0, off, len, cp)
    868 	struct	mbuf *m0;
    869 	int off;
    870 	int len;
    871 	caddr_t cp;
    872 {
    873 	int mlen;
    874 	struct mbuf *m = m0, *n;
    875 	int totlen = 0;
    876 
    877 	if (m0 == 0)
    878 		return;
    879 	while (off > (mlen = m->m_len)) {
    880 		off -= mlen;
    881 		totlen += mlen;
    882 		if (m->m_next == 0) {
    883 			n = m_getclr(M_DONTWAIT, m->m_type);
    884 			if (n == 0)
    885 				goto out;
    886 			n->m_len = min(MLEN, len + off);
    887 			m->m_next = n;
    888 		}
    889 		m = m->m_next;
    890 	}
    891 	while (len > 0) {
    892 		mlen = min (m->m_len - off, len);
    893 		memcpy(mtod(m, caddr_t) + off, cp, (unsigned)mlen);
    894 		cp += mlen;
    895 		len -= mlen;
    896 		mlen += off;
    897 		off = 0;
    898 		totlen += mlen;
    899 		if (len == 0)
    900 			break;
    901 		if (m->m_next == 0) {
    902 			n = m_get(M_DONTWAIT, m->m_type);
    903 			if (n == 0)
    904 				break;
    905 			n->m_len = min(MLEN, len);
    906 			m->m_next = n;
    907 		}
    908 		m = m->m_next;
    909 	}
    910 out:	if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
    911 		m->m_pkthdr.len = totlen;
    912 }
    913