Home | History | Annotate | Line # | Download | only in kern
uipc_mbuf.c revision 1.62
      1 /*	$NetBSD: uipc_mbuf.c,v 1.62 2003/01/31 04:55:52 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 1.62 2003/01/31 04:55:52 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(struct pool *, int);
    110 void	mclpool_release(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(void)
    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(int *name, u_int namelen, void *oldp, size_t *oldlenp,
    156     void *newp, size_t newlen)
    157 {
    158 	int error, newval;
    159 
    160 	/* All sysctl names at this level are terminal. */
    161 	if (namelen != 1)
    162 		return (ENOTDIR);		/* overloaded */
    163 
    164 	switch (name[0]) {
    165 	case MBUF_MSIZE:
    166 		return (sysctl_rdint(oldp, oldlenp, newp, msize));
    167 	case MBUF_MCLBYTES:
    168 		return (sysctl_rdint(oldp, oldlenp, newp, mclbytes));
    169 	case MBUF_NMBCLUSTERS:
    170 		/*
    171 		 * If we have direct-mapped pool pages, we can adjust this
    172 		 * number on the fly.  If not, we're limited by the size
    173 		 * of mb_map, and cannot change this value.
    174 		 *
    175 		 * Note: we only allow the value to be increased, never
    176 		 * decreased.
    177 		 */
    178 		if (mb_map == NULL) {
    179 			newval = nmbclusters;
    180 			error = sysctl_int(oldp, oldlenp, newp, newlen,
    181 			    &newval);
    182 			if (error != 0)
    183 				return (error);
    184 			if (newp != NULL) {
    185 				if (newval >= nmbclusters) {
    186 					nmbclusters = newval;
    187 					pool_sethardlimit(&mclpool,
    188 					    nmbclusters, mclpool_warnmsg, 60);
    189 				} else
    190 					error = EINVAL;
    191 			}
    192 			return (error);
    193 		} else
    194 			return (sysctl_rdint(oldp, oldlenp, newp, nmbclusters));
    195 	case MBUF_MBLOWAT:
    196 	case MBUF_MCLLOWAT:
    197 		/* New value must be >= 0. */
    198 		newval = (name[0] == MBUF_MBLOWAT) ? mblowat : mcllowat;
    199 		error = sysctl_int(oldp, oldlenp, newp, newlen, &newval);
    200 		if (error != 0)
    201 			return (error);
    202 		if (newp != NULL) {
    203 			if (newval >= 0) {
    204 				if (name[0] == MBUF_MBLOWAT) {
    205 					mblowat = newval;
    206 					pool_setlowat(&mbpool, newval);
    207 				} else {
    208 					mcllowat = newval;
    209 					pool_setlowat(&mclpool, newval);
    210 				}
    211 			} else
    212 				error = EINVAL;
    213 		}
    214 		return (error);
    215 	default:
    216 		return (EOPNOTSUPP);
    217 	}
    218 	/* NOTREACHED */
    219 }
    220 
    221 void *
    222 mclpool_alloc(struct pool *pp, int flags)
    223 {
    224 	boolean_t waitok = (flags & PR_WAITOK) ? TRUE : FALSE;
    225 
    226 	return ((void *)uvm_km_alloc_poolpage1(mb_map, NULL, waitok));
    227 }
    228 
    229 void
    230 mclpool_release(struct pool *pp, void *v)
    231 {
    232 
    233 	uvm_km_free_poolpage1(mb_map, (vaddr_t)v);
    234 }
    235 
    236 void
    237 m_reclaim(void *arg, int flags)
    238 {
    239 	struct domain *dp;
    240 	struct protosw *pr;
    241 	struct ifnet *ifp;
    242 	int s = splvm();
    243 
    244 	for (dp = domains; dp; dp = dp->dom_next)
    245 		for (pr = dp->dom_protosw;
    246 		     pr < dp->dom_protoswNPROTOSW; pr++)
    247 			if (pr->pr_drain)
    248 				(*pr->pr_drain)();
    249 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list))
    250 		if (ifp->if_drain)
    251 			(*ifp->if_drain)(ifp);
    252 	splx(s);
    253 	mbstat.m_drain++;
    254 }
    255 
    256 /*
    257  * Space allocation routines.
    258  * These are also available as macros
    259  * for critical paths.
    260  */
    261 struct mbuf *
    262 m_get(int nowait, int type)
    263 {
    264 	struct mbuf *m;
    265 
    266 	MGET(m, nowait, type);
    267 	return (m);
    268 }
    269 
    270 struct mbuf *
    271 m_gethdr(int nowait, int type)
    272 {
    273 	struct mbuf *m;
    274 
    275 	MGETHDR(m, nowait, type);
    276 	return (m);
    277 }
    278 
    279 struct mbuf *
    280 m_getclr(int nowait, int type)
    281 {
    282 	struct mbuf *m;
    283 
    284 	MGET(m, nowait, type);
    285 	if (m == 0)
    286 		return (0);
    287 	memset(mtod(m, caddr_t), 0, MLEN);
    288 	return (m);
    289 }
    290 
    291 struct mbuf *
    292 m_free(struct mbuf *m)
    293 {
    294 	struct mbuf *n;
    295 
    296 	MFREE(m, n);
    297 	return (n);
    298 }
    299 
    300 void
    301 m_freem(struct mbuf *m)
    302 {
    303 	struct mbuf *n;
    304 
    305 	if (m == NULL)
    306 		return;
    307 	do {
    308 		MFREE(m, n);
    309 		m = n;
    310 	} while (m);
    311 }
    312 
    313 /*
    314  * Mbuffer utility routines.
    315  */
    316 
    317 /*
    318  * Lesser-used path for M_PREPEND:
    319  * allocate new mbuf to prepend to chain,
    320  * copy junk along.
    321  */
    322 struct mbuf *
    323 m_prepend(struct mbuf *m, int len, int how)
    324 {
    325 	struct mbuf *mn;
    326 
    327 	MGET(mn, how, m->m_type);
    328 	if (mn == (struct mbuf *)NULL) {
    329 		m_freem(m);
    330 		return ((struct mbuf *)NULL);
    331 	}
    332 	if (m->m_flags & M_PKTHDR) {
    333 		M_COPY_PKTHDR(mn, m);
    334 		m->m_flags &= ~M_PKTHDR;
    335 	}
    336 	mn->m_next = m;
    337 	m = mn;
    338 	if (len < MHLEN)
    339 		MH_ALIGN(m, len);
    340 	m->m_len = len;
    341 	return (m);
    342 }
    343 
    344 /*
    345  * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
    346  * continuing for "len" bytes.  If len is M_COPYALL, copy to end of mbuf.
    347  * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
    348  */
    349 int MCFail;
    350 
    351 struct mbuf *
    352 m_copym(struct mbuf *m, int off0, int len, int wait)
    353 {
    354 	return m_copym0(m, off0, len, wait, 0);	/* shallow copy on M_EXT */
    355 }
    356 
    357 struct mbuf *
    358 m_dup(struct mbuf *m, int off0, int len, int wait)
    359 {
    360 	return m_copym0(m, off0, len, wait, 1);	/* deep copy */
    361 }
    362 
    363 static struct mbuf *
    364 m_copym0(struct mbuf *m, int off0, int len, int wait, int deep)
    365 {
    366 	struct mbuf *n, **np;
    367 	int off = off0;
    368 	struct mbuf *top;
    369 	int copyhdr = 0;
    370 
    371 	if (off < 0 || len < 0)
    372 		panic("m_copym: off %d, len %d", off, len);
    373 	if (off == 0 && m->m_flags & M_PKTHDR)
    374 		copyhdr = 1;
    375 	while (off > 0) {
    376 		if (m == 0)
    377 			panic("m_copym: m == 0");
    378 		if (off < m->m_len)
    379 			break;
    380 		off -= m->m_len;
    381 		m = m->m_next;
    382 	}
    383 	np = &top;
    384 	top = 0;
    385 	while (len > 0) {
    386 		if (m == 0) {
    387 			if (len != M_COPYALL)
    388 				panic("m_copym: m == 0 and not COPYALL");
    389 			break;
    390 		}
    391 		MGET(n, wait, m->m_type);
    392 		*np = n;
    393 		if (n == 0)
    394 			goto nospace;
    395 		if (copyhdr) {
    396 			M_COPY_PKTHDR(n, m);
    397 			if (len == M_COPYALL)
    398 				n->m_pkthdr.len -= off0;
    399 			else
    400 				n->m_pkthdr.len = len;
    401 			copyhdr = 0;
    402 		}
    403 		n->m_len = min(len, m->m_len - off);
    404 		if (m->m_flags & M_EXT) {
    405 			if (!deep) {
    406 				n->m_data = m->m_data + off;
    407 				n->m_ext = m->m_ext;
    408 				MCLADDREFERENCE(m, n);
    409 			} else {
    410 				/*
    411 				 * we are unsure about the way m was allocated.
    412 				 * copy into multiple MCLBYTES cluster mbufs.
    413 				 */
    414 				MCLGET(n, wait);
    415 				n->m_len = 0;
    416 				n->m_len = M_TRAILINGSPACE(n);
    417 				n->m_len = min(n->m_len, len);
    418 				n->m_len = min(n->m_len, m->m_len - off);
    419 				memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off,
    420 				    (unsigned)n->m_len);
    421 			}
    422 		} else
    423 			memcpy(mtod(n, caddr_t), mtod(m, caddr_t)+off,
    424 			    (unsigned)n->m_len);
    425 		if (len != M_COPYALL)
    426 			len -= n->m_len;
    427 		off += n->m_len;
    428 #ifdef DIAGNOSTIC
    429 		if (off > m->m_len)
    430 			panic("m_copym0 overrun");
    431 #endif
    432 		if (off == m->m_len) {
    433 			m = m->m_next;
    434 			off = 0;
    435 		}
    436 		np = &n->m_next;
    437 	}
    438 	if (top == 0)
    439 		MCFail++;
    440 	return (top);
    441 nospace:
    442 	m_freem(top);
    443 	MCFail++;
    444 	return (0);
    445 }
    446 
    447 /*
    448  * Copy an entire packet, including header (which must be present).
    449  * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
    450  */
    451 struct mbuf *
    452 m_copypacket(struct mbuf *m, int how)
    453 {
    454 	struct mbuf *top, *n, *o;
    455 
    456 	MGET(n, how, m->m_type);
    457 	top = n;
    458 	if (!n)
    459 		goto nospace;
    460 
    461 	M_COPY_PKTHDR(n, m);
    462 	n->m_len = m->m_len;
    463 	if (m->m_flags & M_EXT) {
    464 		n->m_data = m->m_data;
    465 		n->m_ext = m->m_ext;
    466 		MCLADDREFERENCE(m, n);
    467 	} else {
    468 		memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
    469 	}
    470 
    471 	m = m->m_next;
    472 	while (m) {
    473 		MGET(o, how, m->m_type);
    474 		if (!o)
    475 			goto nospace;
    476 
    477 		n->m_next = o;
    478 		n = n->m_next;
    479 
    480 		n->m_len = m->m_len;
    481 		if (m->m_flags & M_EXT) {
    482 			n->m_data = m->m_data;
    483 			n->m_ext = m->m_ext;
    484 			MCLADDREFERENCE(m, n);
    485 		} else {
    486 			memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
    487 		}
    488 
    489 		m = m->m_next;
    490 	}
    491 	return top;
    492 nospace:
    493 	m_freem(top);
    494 	MCFail++;
    495 	return 0;
    496 }
    497 
    498 /*
    499  * Copy data from an mbuf chain starting "off" bytes from the beginning,
    500  * continuing for "len" bytes, into the indicated buffer.
    501  */
    502 void
    503 m_copydata(struct mbuf *m, int off, int len, caddr_t cp)
    504 {
    505 	unsigned count;
    506 
    507 	if (off < 0 || len < 0)
    508 		panic("m_copydata");
    509 	while (off > 0) {
    510 		if (m == 0)
    511 			panic("m_copydata");
    512 		if (off < m->m_len)
    513 			break;
    514 		off -= m->m_len;
    515 		m = m->m_next;
    516 	}
    517 	while (len > 0) {
    518 		if (m == 0)
    519 			panic("m_copydata");
    520 		count = min(m->m_len - off, len);
    521 		memcpy(cp, mtod(m, caddr_t) + off, count);
    522 		len -= count;
    523 		cp += count;
    524 		off = 0;
    525 		m = m->m_next;
    526 	}
    527 }
    528 
    529 /*
    530  * Concatenate mbuf chain n to m.
    531  * Both chains must be of the same type (e.g. MT_DATA).
    532  * Any m_pkthdr is not updated.
    533  */
    534 void
    535 m_cat(struct mbuf *m, struct mbuf *n)
    536 {
    537 	while (m->m_next)
    538 		m = m->m_next;
    539 	while (n) {
    540 		if (m->m_flags & M_EXT ||
    541 		    m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
    542 			/* just join the two chains */
    543 			m->m_next = n;
    544 			return;
    545 		}
    546 		/* splat the data from one into the other */
    547 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
    548 		    (u_int)n->m_len);
    549 		m->m_len += n->m_len;
    550 		n = m_free(n);
    551 	}
    552 }
    553 
    554 void
    555 m_adj(struct mbuf *mp, int req_len)
    556 {
    557 	int len = req_len;
    558 	struct mbuf *m;
    559 	int count;
    560 
    561 	if ((m = mp) == NULL)
    562 		return;
    563 	if (len >= 0) {
    564 		/*
    565 		 * Trim from head.
    566 		 */
    567 		while (m != NULL && len > 0) {
    568 			if (m->m_len <= len) {
    569 				len -= m->m_len;
    570 				m->m_len = 0;
    571 				m = m->m_next;
    572 			} else {
    573 				m->m_len -= len;
    574 				m->m_data += len;
    575 				len = 0;
    576 			}
    577 		}
    578 		m = mp;
    579 		if (mp->m_flags & M_PKTHDR)
    580 			m->m_pkthdr.len -= (req_len - len);
    581 	} else {
    582 		/*
    583 		 * Trim from tail.  Scan the mbuf chain,
    584 		 * calculating its length and finding the last mbuf.
    585 		 * If the adjustment only affects this mbuf, then just
    586 		 * adjust and return.  Otherwise, rescan and truncate
    587 		 * after the remaining size.
    588 		 */
    589 		len = -len;
    590 		count = 0;
    591 		for (;;) {
    592 			count += m->m_len;
    593 			if (m->m_next == (struct mbuf *)0)
    594 				break;
    595 			m = m->m_next;
    596 		}
    597 		if (m->m_len >= len) {
    598 			m->m_len -= len;
    599 			if (mp->m_flags & M_PKTHDR)
    600 				mp->m_pkthdr.len -= len;
    601 			return;
    602 		}
    603 		count -= len;
    604 		if (count < 0)
    605 			count = 0;
    606 		/*
    607 		 * Correct length for chain is "count".
    608 		 * Find the mbuf with last data, adjust its length,
    609 		 * and toss data from remaining mbufs on chain.
    610 		 */
    611 		m = mp;
    612 		if (m->m_flags & M_PKTHDR)
    613 			m->m_pkthdr.len = count;
    614 		for (; m; m = m->m_next) {
    615 			if (m->m_len >= count) {
    616 				m->m_len = count;
    617 				break;
    618 			}
    619 			count -= m->m_len;
    620 		}
    621 		while (m->m_next)
    622 			(m = m->m_next) ->m_len = 0;
    623 	}
    624 }
    625 
    626 /*
    627  * Rearange an mbuf chain so that len bytes are contiguous
    628  * and in the data area of an mbuf (so that mtod and dtom
    629  * will work for a structure of size len).  Returns the resulting
    630  * mbuf chain on success, frees it and returns null on failure.
    631  * If there is room, it will add up to max_protohdr-len extra bytes to the
    632  * contiguous region in an attempt to avoid being called next time.
    633  */
    634 int MPFail;
    635 
    636 struct mbuf *
    637 m_pullup(struct mbuf *n, int len)
    638 {
    639 	struct mbuf *m;
    640 	int count;
    641 	int space;
    642 
    643 	/*
    644 	 * If first mbuf has no cluster, and has room for len bytes
    645 	 * without shifting current data, pullup into it,
    646 	 * otherwise allocate a new mbuf to prepend to the chain.
    647 	 */
    648 	if ((n->m_flags & M_EXT) == 0 &&
    649 	    n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
    650 		if (n->m_len >= len)
    651 			return (n);
    652 		m = n;
    653 		n = n->m_next;
    654 		len -= m->m_len;
    655 	} else {
    656 		if (len > MHLEN)
    657 			goto bad;
    658 		MGET(m, M_DONTWAIT, n->m_type);
    659 		if (m == 0)
    660 			goto bad;
    661 		m->m_len = 0;
    662 		if (n->m_flags & M_PKTHDR) {
    663 			M_COPY_PKTHDR(m, n);
    664 			n->m_flags &= ~M_PKTHDR;
    665 		}
    666 	}
    667 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
    668 	do {
    669 		count = min(min(max(len, max_protohdr), space), n->m_len);
    670 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
    671 		  (unsigned)count);
    672 		len -= count;
    673 		m->m_len += count;
    674 		n->m_len -= count;
    675 		space -= count;
    676 		if (n->m_len)
    677 			n->m_data += count;
    678 		else
    679 			n = m_free(n);
    680 	} while (len > 0 && n);
    681 	if (len > 0) {
    682 		(void) m_free(m);
    683 		goto bad;
    684 	}
    685 	m->m_next = n;
    686 	return (m);
    687 bad:
    688 	m_freem(n);
    689 	MPFail++;
    690 	return (0);
    691 }
    692 
    693 /*
    694  * Like m_pullup(), except a new mbuf is always allocated, and we allow
    695  * the amount of empty space before the data in the new mbuf to be specified
    696  * (in the event that the caller expects to prepend later).
    697  */
    698 int MSFail;
    699 
    700 struct mbuf *
    701 m_copyup(struct mbuf *n, int len, int dstoff)
    702 {
    703 	struct mbuf *m;
    704 	int count, space;
    705 
    706 	if (len > (MHLEN - dstoff))
    707 		goto bad;
    708 	MGET(m, M_DONTWAIT, n->m_type);
    709 	if (m == NULL)
    710 		goto bad;
    711 	m->m_len = 0;
    712 	if (n->m_flags & M_PKTHDR) {
    713 		M_COPY_PKTHDR(m, n);
    714 		n->m_flags &= ~M_PKTHDR;
    715 	}
    716 	m->m_data += dstoff;
    717 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
    718 	do {
    719 		count = min(min(max(len, max_protohdr), space), n->m_len);
    720 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
    721 		    (unsigned)count);
    722 		len -= count;
    723 		m->m_len += count;
    724 		n->m_len -= count;
    725 		space -= count;
    726 		if (n->m_len)
    727 			n->m_data += count;
    728 		else
    729 			n = m_free(n);
    730 	} while (len > 0 && n);
    731 	if (len > 0) {
    732 		(void) m_free(m);
    733 		goto bad;
    734 	}
    735 	m->m_next = n;
    736 	return (m);
    737  bad:
    738 	m_freem(n);
    739 	MSFail++;
    740 	return (NULL);
    741 }
    742 
    743 /*
    744  * Partition an mbuf chain in two pieces, returning the tail --
    745  * all but the first len0 bytes.  In case of failure, it returns NULL and
    746  * attempts to restore the chain to its original state.
    747  */
    748 struct mbuf *
    749 m_split(struct mbuf *m0, int len0, int wait)
    750 {
    751 	struct mbuf *m, *n;
    752 	unsigned len = len0, remain, len_save;
    753 
    754 	for (m = m0; m && len > m->m_len; m = m->m_next)
    755 		len -= m->m_len;
    756 	if (m == 0)
    757 		return (0);
    758 	remain = m->m_len - len;
    759 	if (m0->m_flags & M_PKTHDR) {
    760 		MGETHDR(n, wait, m0->m_type);
    761 		if (n == 0)
    762 			return (0);
    763 		n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
    764 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
    765 		len_save = m0->m_pkthdr.len;
    766 		m0->m_pkthdr.len = len0;
    767 		if (m->m_flags & M_EXT)
    768 			goto extpacket;
    769 		if (remain > MHLEN) {
    770 			/* m can't be the lead packet */
    771 			MH_ALIGN(n, 0);
    772 			n->m_next = m_split(m, len, wait);
    773 			if (n->m_next == 0) {
    774 				(void) m_free(n);
    775 				m0->m_pkthdr.len = len_save;
    776 				return (0);
    777 			} else
    778 				return (n);
    779 		} else
    780 			MH_ALIGN(n, remain);
    781 	} else if (remain == 0) {
    782 		n = m->m_next;
    783 		m->m_next = 0;
    784 		return (n);
    785 	} else {
    786 		MGET(n, wait, m->m_type);
    787 		if (n == 0)
    788 			return (0);
    789 		M_ALIGN(n, remain);
    790 	}
    791 extpacket:
    792 	if (m->m_flags & M_EXT) {
    793 		n->m_ext = m->m_ext;
    794 		MCLADDREFERENCE(m, n);
    795 		n->m_data = m->m_data + len;
    796 	} else {
    797 		memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + len, remain);
    798 	}
    799 	n->m_len = remain;
    800 	m->m_len = len;
    801 	n->m_next = m->m_next;
    802 	m->m_next = 0;
    803 	return (n);
    804 }
    805 /*
    806  * Routine to copy from device local memory into mbufs.
    807  */
    808 struct mbuf *
    809 m_devget(char *buf, int totlen, int off0, struct ifnet *ifp,
    810     void (*copy)(const void *from, void *to, size_t len))
    811 {
    812 	struct mbuf *m;
    813 	struct mbuf *top = 0, **mp = &top;
    814 	int off = off0, len;
    815 	char *cp;
    816 	char *epkt;
    817 
    818 	cp = buf;
    819 	epkt = cp + totlen;
    820 	if (off) {
    821 		/*
    822 		 * If 'off' is non-zero, packet is trailer-encapsulated,
    823 		 * so we have to skip the type and length fields.
    824 		 */
    825 		cp += off + 2 * sizeof(u_int16_t);
    826 		totlen -= 2 * sizeof(u_int16_t);
    827 	}
    828 	MGETHDR(m, M_DONTWAIT, MT_DATA);
    829 	if (m == 0)
    830 		return (0);
    831 	m->m_pkthdr.rcvif = ifp;
    832 	m->m_pkthdr.len = totlen;
    833 	m->m_len = MHLEN;
    834 
    835 	while (totlen > 0) {
    836 		if (top) {
    837 			MGET(m, M_DONTWAIT, MT_DATA);
    838 			if (m == 0) {
    839 				m_freem(top);
    840 				return (0);
    841 			}
    842 			m->m_len = MLEN;
    843 		}
    844 		len = min(totlen, epkt - cp);
    845 		if (len >= MINCLSIZE) {
    846 			MCLGET(m, M_DONTWAIT);
    847 			if ((m->m_flags & M_EXT) == 0) {
    848 				m_free(m);
    849 				m_freem(top);
    850 				return (0);
    851 			}
    852 			m->m_len = len = min(len, MCLBYTES);
    853 		} else {
    854 			/*
    855 			 * Place initial small packet/header at end of mbuf.
    856 			 */
    857 			if (len < m->m_len) {
    858 				if (top == 0 && len + max_linkhdr <= m->m_len)
    859 					m->m_data += max_linkhdr;
    860 				m->m_len = len;
    861 			} else
    862 				len = m->m_len;
    863 		}
    864 		if (copy)
    865 			copy(cp, mtod(m, caddr_t), (size_t)len);
    866 		else
    867 			memcpy(mtod(m, caddr_t), cp, (size_t)len);
    868 		cp += len;
    869 		*mp = m;
    870 		mp = &m->m_next;
    871 		totlen -= len;
    872 		if (cp == epkt)
    873 			cp = buf;
    874 	}
    875 	return (top);
    876 }
    877 
    878 /*
    879  * Copy data from a buffer back into the indicated mbuf chain,
    880  * starting "off" bytes from the beginning, extending the mbuf
    881  * chain if necessary.
    882  */
    883 void
    884 m_copyback(struct mbuf *m0, int off, int len, caddr_t cp)
    885 {
    886 	int mlen;
    887 	struct mbuf *m = m0, *n;
    888 	int totlen = 0;
    889 
    890 	if (m0 == 0)
    891 		return;
    892 	while (off > (mlen = m->m_len)) {
    893 		off -= mlen;
    894 		totlen += mlen;
    895 		if (m->m_next == 0) {
    896 			n = m_getclr(M_DONTWAIT, m->m_type);
    897 			if (n == 0)
    898 				goto out;
    899 			n->m_len = min(MLEN, len + off);
    900 			m->m_next = n;
    901 		}
    902 		m = m->m_next;
    903 	}
    904 	while (len > 0) {
    905 		mlen = min (m->m_len - off, len);
    906 		memcpy(mtod(m, caddr_t) + off, cp, (unsigned)mlen);
    907 		cp += mlen;
    908 		len -= mlen;
    909 		mlen += off;
    910 		off = 0;
    911 		totlen += mlen;
    912 		if (len == 0)
    913 			break;
    914 		if (m->m_next == 0) {
    915 			n = m_get(M_DONTWAIT, m->m_type);
    916 			if (n == 0)
    917 				break;
    918 			n->m_len = min(MLEN, len);
    919 			m->m_next = n;
    920 		}
    921 		m = m->m_next;
    922 	}
    923 out:	if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
    924 		m->m_pkthdr.len = totlen;
    925 }
    926