Home | History | Annotate | Line # | Download | only in kern
uipc_mbuf.c revision 1.100.2.15
      1 /*	$NetBSD: uipc_mbuf.c,v 1.100.2.15 2007/10/27 11:40:55 yamt 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. Neither the name of the University nor the names of its contributors
     53  *    may be used to endorse or promote products derived from this software
     54  *    without specific prior written permission.
     55  *
     56  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     57  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     58  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     59  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     60  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     61  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     62  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     63  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     64  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     65  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     66  * SUCH DAMAGE.
     67  *
     68  *	@(#)uipc_mbuf.c	8.4 (Berkeley) 2/14/95
     69  */
     70 
     71 #include <sys/cdefs.h>
     72 __KERNEL_RCSID(0, "$NetBSD: uipc_mbuf.c,v 1.100.2.15 2007/10/27 11:40:55 yamt Exp $");
     73 
     74 #include "opt_mbuftrace.h"
     75 #include "opt_ddb.h"
     76 
     77 #include <sys/param.h>
     78 #include <sys/systm.h>
     79 #include <sys/proc.h>
     80 #include <sys/malloc.h>
     81 #define MBTYPES
     82 #include <sys/mbuf.h>
     83 #include <sys/kernel.h>
     84 #include <sys/syslog.h>
     85 #include <sys/domain.h>
     86 #include <sys/protosw.h>
     87 #include <sys/pool.h>
     88 #include <sys/socket.h>
     89 #include <sys/sysctl.h>
     90 
     91 #include <net/if.h>
     92 
     93 #include <uvm/uvm.h>
     94 
     95 
     96 struct	pool mbpool;		/* mbuf pool */
     97 struct	pool mclpool;		/* mbuf cluster pool */
     98 
     99 struct pool_cache mbpool_cache;
    100 struct pool_cache mclpool_cache;
    101 
    102 struct mbstat mbstat;
    103 int	max_linkhdr;
    104 int	max_protohdr;
    105 int	max_hdr;
    106 int	max_datalen;
    107 
    108 static int mb_ctor(void *, void *, int);
    109 
    110 static void	*mclpool_alloc(struct pool *, int);
    111 static void	mclpool_release(struct pool *, void *);
    112 
    113 static struct pool_allocator mclpool_allocator = {
    114 	.pa_alloc = mclpool_alloc,
    115 	.pa_free = mclpool_release,
    116 };
    117 
    118 static struct mbuf *m_copym0(struct mbuf *, int, int, int, int);
    119 static struct mbuf *m_split0(struct mbuf *, int, int, int);
    120 static int m_copyback0(struct mbuf **, int, int, const void *, int, int);
    121 
    122 /* flags for m_copyback0 */
    123 #define	M_COPYBACK0_COPYBACK	0x0001	/* copyback from cp */
    124 #define	M_COPYBACK0_PRESERVE	0x0002	/* preserve original data */
    125 #define	M_COPYBACK0_COW		0x0004	/* do copy-on-write */
    126 #define	M_COPYBACK0_EXTEND	0x0008	/* extend chain */
    127 
    128 static const char mclpool_warnmsg[] =
    129     "WARNING: mclpool limit reached; increase NMBCLUSTERS";
    130 
    131 MALLOC_DEFINE(M_MBUF, "mbuf", "mbuf");
    132 
    133 #ifdef MBUFTRACE
    134 struct mownerhead mowners = LIST_HEAD_INITIALIZER(mowners);
    135 struct mowner unknown_mowners[] = {
    136 	MOWNER_INIT("unknown", "free"),
    137 	MOWNER_INIT("unknown", "data"),
    138 	MOWNER_INIT("unknown", "header"),
    139 	MOWNER_INIT("unknown", "soname"),
    140 	MOWNER_INIT("unknown", "soopts"),
    141 	MOWNER_INIT("unknown", "ftable"),
    142 	MOWNER_INIT("unknown", "control"),
    143 	MOWNER_INIT("unknown", "oobdata"),
    144 };
    145 struct mowner revoked_mowner = MOWNER_INIT("revoked", "");
    146 #endif
    147 
    148 #define	MEXT_LOCK(m)	mutex_enter(&(m)->m_ext.ext_lock)
    149 #define	MEXT_UNLOCK(m)	mutex_exit(&(m)->m_ext.ext_lock)
    150 #define	MEXT_ISEMBEDDED(m) ((m)->m_ext_ref == (m))
    151 
    152 static inline void
    153 mcl_inc_reference(struct mbuf *m)
    154 {
    155 
    156 	MEXT_LOCK(m);
    157 	m->m_ext.ext_refcnt++;
    158 	MEXT_UNLOCK(m);
    159 }
    160 
    161 static inline bool
    162 mcl_dec_and_test_reference(struct mbuf *m)
    163 {
    164 	bool gotzero;
    165 
    166 	MEXT_LOCK(m);
    167 	KASSERT(m->m_ext.ext_refcnt > 0);
    168 	m->m_ext.ext_refcnt--;
    169 	gotzero = (m->m_ext.ext_refcnt == 0);
    170 	MEXT_UNLOCK(m);
    171 
    172 	return gotzero;
    173 }
    174 
    175 #define	MCLADDREFERENCE(o, n)						\
    176 do {									\
    177 	KASSERT(((o)->m_flags & M_EXT) != 0);				\
    178 	KASSERT(((n)->m_flags & M_EXT) == 0);				\
    179 	KASSERT((o)->m_ext.ext_refcnt >= 1);				\
    180 	(n)->m_flags |= ((o)->m_flags & M_EXTCOPYFLAGS);		\
    181 	mcl_inc_reference((o));						\
    182 	(n)->m_ext_ref = (o)->m_ext_ref;				\
    183 	_MOWNERREF((n), (n)->m_flags);					\
    184 	MCLREFDEBUGN((n), __FILE__, __LINE__);				\
    185 } while (/* CONSTCOND */ 0)
    186 
    187 /*
    188  * Initialize the mbuf allocator.
    189  */
    190 void
    191 mbinit(void)
    192 {
    193 
    194 	KASSERT(sizeof(struct _m_ext) <= MHLEN);
    195 	KASSERT(sizeof(struct mbuf) == MSIZE);
    196 
    197 	mclpool_allocator.pa_backingmap = mb_map;
    198 	pool_init(&mbpool, msize, 0, 0, 0, "mbpl", NULL, IPL_VM);
    199 	pool_init(&mclpool, mclbytes, 0, 0, 0, "mclpl", &mclpool_allocator,
    200 	    IPL_VM);
    201 
    202 	pool_set_drain_hook(&mbpool, m_reclaim, NULL);
    203 	pool_set_drain_hook(&mclpool, m_reclaim, NULL);
    204 
    205 	pool_cache_init(&mbpool_cache, &mbpool, mb_ctor, NULL, NULL);
    206 	pool_cache_init(&mclpool_cache, &mclpool, NULL, NULL, NULL);
    207 
    208 	/*
    209 	 * Set the hard limit on the mclpool to the number of
    210 	 * mbuf clusters the kernel is to support.  Log the limit
    211 	 * reached message max once a minute.
    212 	 */
    213 	pool_sethardlimit(&mclpool, nmbclusters, mclpool_warnmsg, 60);
    214 
    215 	/*
    216 	 * Set a low water mark for both mbufs and clusters.  This should
    217 	 * help ensure that they can be allocated in a memory starvation
    218 	 * situation.  This is important for e.g. diskless systems which
    219 	 * must allocate mbufs in order for the pagedaemon to clean pages.
    220 	 */
    221 	pool_setlowat(&mbpool, mblowat);
    222 	pool_setlowat(&mclpool, mcllowat);
    223 
    224 #ifdef MBUFTRACE
    225 	{
    226 		/*
    227 		 * Attach the unknown mowners.
    228 		 */
    229 		int i;
    230 		MOWNER_ATTACH(&revoked_mowner);
    231 		for (i = sizeof(unknown_mowners)/sizeof(unknown_mowners[0]);
    232 		     i-- > 0; )
    233 			MOWNER_ATTACH(&unknown_mowners[i]);
    234 	}
    235 #endif
    236 }
    237 
    238 /*
    239  * sysctl helper routine for the kern.mbuf subtree.  nmbclusters may
    240  * or may not be writable, and mblowat and mcllowat need range
    241  * checking and pool tweaking after being reset.
    242  */
    243 static int
    244 sysctl_kern_mbuf(SYSCTLFN_ARGS)
    245 {
    246 	int error, newval;
    247 	struct sysctlnode node;
    248 
    249 	node = *rnode;
    250 	node.sysctl_data = &newval;
    251 	switch (rnode->sysctl_num) {
    252 	case MBUF_NMBCLUSTERS:
    253 		if (mb_map != NULL) {
    254 			node.sysctl_flags &= ~CTLFLAG_READWRITE;
    255 			node.sysctl_flags |= CTLFLAG_READONLY;
    256 		}
    257 		/* FALLTHROUGH */
    258 	case MBUF_MBLOWAT:
    259 	case MBUF_MCLLOWAT:
    260 		newval = *(int*)rnode->sysctl_data;
    261 		break;
    262 	default:
    263 		return (EOPNOTSUPP);
    264 	}
    265 
    266 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
    267 	if (error || newp == NULL)
    268 		return (error);
    269 	if (newval < 0)
    270 		return (EINVAL);
    271 
    272 	switch (node.sysctl_num) {
    273 	case MBUF_NMBCLUSTERS:
    274 		if (newval < nmbclusters)
    275 			return (EINVAL);
    276 		nmbclusters = newval;
    277 		pool_sethardlimit(&mclpool, nmbclusters, mclpool_warnmsg, 60);
    278 		break;
    279 	case MBUF_MBLOWAT:
    280 		mblowat = newval;
    281 		pool_setlowat(&mbpool, mblowat);
    282 		break;
    283 	case MBUF_MCLLOWAT:
    284 		mcllowat = newval;
    285 		pool_setlowat(&mclpool, mcllowat);
    286 		break;
    287 	}
    288 
    289 	return (0);
    290 }
    291 
    292 #ifdef MBUFTRACE
    293 static int
    294 sysctl_kern_mbuf_mowners(SYSCTLFN_ARGS)
    295 {
    296 	struct mowner *mo;
    297 	size_t len = 0;
    298 	int error = 0;
    299 
    300 	if (namelen != 0)
    301 		return (EINVAL);
    302 	if (newp != NULL)
    303 		return (EPERM);
    304 
    305 	LIST_FOREACH(mo, &mowners, mo_link) {
    306 		if (oldp != NULL) {
    307 			if (*oldlenp - len < sizeof(*mo)) {
    308 				error = ENOMEM;
    309 				break;
    310 			}
    311 			error = copyout(mo, (char *)oldp + len, sizeof(*mo));
    312 			if (error)
    313 				break;
    314 		}
    315 		len += sizeof(*mo);
    316 	}
    317 
    318 	if (error == 0)
    319 		*oldlenp = len;
    320 
    321 	return (error);
    322 }
    323 #endif /* MBUFTRACE */
    324 
    325 SYSCTL_SETUP(sysctl_kern_mbuf_setup, "sysctl kern.mbuf subtree setup")
    326 {
    327 
    328 	sysctl_createv(clog, 0, NULL, NULL,
    329 		       CTLFLAG_PERMANENT,
    330 		       CTLTYPE_NODE, "kern", NULL,
    331 		       NULL, 0, NULL, 0,
    332 		       CTL_KERN, CTL_EOL);
    333 	sysctl_createv(clog, 0, NULL, NULL,
    334 		       CTLFLAG_PERMANENT,
    335 		       CTLTYPE_NODE, "mbuf",
    336 		       SYSCTL_DESCR("mbuf control variables"),
    337 		       NULL, 0, NULL, 0,
    338 		       CTL_KERN, KERN_MBUF, CTL_EOL);
    339 
    340 	sysctl_createv(clog, 0, NULL, NULL,
    341 		       CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
    342 		       CTLTYPE_INT, "msize",
    343 		       SYSCTL_DESCR("mbuf base size"),
    344 		       NULL, msize, NULL, 0,
    345 		       CTL_KERN, KERN_MBUF, MBUF_MSIZE, CTL_EOL);
    346 	sysctl_createv(clog, 0, NULL, NULL,
    347 		       CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
    348 		       CTLTYPE_INT, "mclbytes",
    349 		       SYSCTL_DESCR("mbuf cluster size"),
    350 		       NULL, mclbytes, NULL, 0,
    351 		       CTL_KERN, KERN_MBUF, MBUF_MCLBYTES, CTL_EOL);
    352 	sysctl_createv(clog, 0, NULL, NULL,
    353 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    354 		       CTLTYPE_INT, "nmbclusters",
    355 		       SYSCTL_DESCR("Limit on the number of mbuf clusters"),
    356 		       sysctl_kern_mbuf, 0, &nmbclusters, 0,
    357 		       CTL_KERN, KERN_MBUF, MBUF_NMBCLUSTERS, CTL_EOL);
    358 	sysctl_createv(clog, 0, NULL, NULL,
    359 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    360 		       CTLTYPE_INT, "mblowat",
    361 		       SYSCTL_DESCR("mbuf low water mark"),
    362 		       sysctl_kern_mbuf, 0, &mblowat, 0,
    363 		       CTL_KERN, KERN_MBUF, MBUF_MBLOWAT, CTL_EOL);
    364 	sysctl_createv(clog, 0, NULL, NULL,
    365 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    366 		       CTLTYPE_INT, "mcllowat",
    367 		       SYSCTL_DESCR("mbuf cluster low water mark"),
    368 		       sysctl_kern_mbuf, 0, &mcllowat, 0,
    369 		       CTL_KERN, KERN_MBUF, MBUF_MCLLOWAT, CTL_EOL);
    370 	sysctl_createv(clog, 0, NULL, NULL,
    371 		       CTLFLAG_PERMANENT,
    372 		       CTLTYPE_STRUCT, "stats",
    373 		       SYSCTL_DESCR("mbuf allocation statistics"),
    374 		       NULL, 0, &mbstat, sizeof(mbstat),
    375 		       CTL_KERN, KERN_MBUF, MBUF_STATS, CTL_EOL);
    376 #ifdef MBUFTRACE
    377 	sysctl_createv(clog, 0, NULL, NULL,
    378 		       CTLFLAG_PERMANENT,
    379 		       CTLTYPE_STRUCT, "mowners",
    380 		       SYSCTL_DESCR("Information about mbuf owners"),
    381 		       sysctl_kern_mbuf_mowners, 0, NULL, 0,
    382 		       CTL_KERN, KERN_MBUF, MBUF_MOWNERS, CTL_EOL);
    383 #endif /* MBUFTRACE */
    384 }
    385 
    386 static void *
    387 mclpool_alloc(struct pool *pp, int flags)
    388 {
    389 	bool waitok = (flags & PR_WAITOK) ? true : false;
    390 
    391 	return ((void *)uvm_km_alloc_poolpage(mb_map, waitok));
    392 }
    393 
    394 static void
    395 mclpool_release(struct pool *pp, void *v)
    396 {
    397 
    398 	uvm_km_free_poolpage(mb_map, (vaddr_t)v);
    399 }
    400 
    401 /*ARGSUSED*/
    402 static int
    403 mb_ctor(void *arg, void *object, int flags)
    404 {
    405 	struct mbuf *m = object;
    406 
    407 #ifdef POOL_VTOPHYS
    408 	m->m_paddr = POOL_VTOPHYS(m);
    409 #else
    410 	m->m_paddr = M_PADDR_INVALID;
    411 #endif
    412 	return (0);
    413 }
    414 
    415 void
    416 m_reclaim(void *arg, int flags)
    417 {
    418 	struct domain *dp;
    419 	const struct protosw *pr;
    420 	struct ifnet *ifp;
    421 	int s = splvm();
    422 
    423 	DOMAIN_FOREACH(dp) {
    424 		for (pr = dp->dom_protosw;
    425 		     pr < dp->dom_protoswNPROTOSW; pr++)
    426 			if (pr->pr_drain)
    427 				(*pr->pr_drain)();
    428 	}
    429 	IFNET_FOREACH(ifp) {
    430 		if (ifp->if_drain)
    431 			(*ifp->if_drain)(ifp);
    432 	}
    433 	splx(s);
    434 	mbstat.m_drain++;
    435 }
    436 
    437 /*
    438  * Space allocation routines.
    439  * These are also available as macros
    440  * for critical paths.
    441  */
    442 struct mbuf *
    443 m_get(int nowait, int type)
    444 {
    445 	struct mbuf *m;
    446 
    447 	MGET(m, nowait, type);
    448 	return (m);
    449 }
    450 
    451 struct mbuf *
    452 m_gethdr(int nowait, int type)
    453 {
    454 	struct mbuf *m;
    455 
    456 	MGETHDR(m, nowait, type);
    457 	return (m);
    458 }
    459 
    460 struct mbuf *
    461 m_getclr(int nowait, int type)
    462 {
    463 	struct mbuf *m;
    464 
    465 	MGET(m, nowait, type);
    466 	if (m == 0)
    467 		return (NULL);
    468 	memset(mtod(m, void *), 0, MLEN);
    469 	return (m);
    470 }
    471 
    472 void
    473 m_clget(struct mbuf *m, int nowait)
    474 {
    475 
    476 	MCLGET(m, nowait);
    477 }
    478 
    479 struct mbuf *
    480 m_free(struct mbuf *m)
    481 {
    482 	struct mbuf *n;
    483 
    484 	MFREE(m, n);
    485 	return (n);
    486 }
    487 
    488 void
    489 m_freem(struct mbuf *m)
    490 {
    491 	struct mbuf *n;
    492 
    493 	if (m == NULL)
    494 		return;
    495 	do {
    496 		MFREE(m, n);
    497 		m = n;
    498 	} while (m);
    499 }
    500 
    501 #ifdef MBUFTRACE
    502 /*
    503  * Walk a chain of mbufs, claiming ownership of each mbuf in the chain.
    504  */
    505 void
    506 m_claimm(struct mbuf *m, struct mowner *mo)
    507 {
    508 
    509 	for (; m != NULL; m = m->m_next)
    510 		MCLAIM(m, mo);
    511 }
    512 #endif
    513 
    514 /*
    515  * Mbuffer utility routines.
    516  */
    517 
    518 /*
    519  * Lesser-used path for M_PREPEND:
    520  * allocate new mbuf to prepend to chain,
    521  * copy junk along.
    522  */
    523 struct mbuf *
    524 m_prepend(struct mbuf *m, int len, int how)
    525 {
    526 	struct mbuf *mn;
    527 
    528 	MGET(mn, how, m->m_type);
    529 	if (mn == (struct mbuf *)NULL) {
    530 		m_freem(m);
    531 		return ((struct mbuf *)NULL);
    532 	}
    533 	if (m->m_flags & M_PKTHDR) {
    534 		M_MOVE_PKTHDR(mn, m);
    535 	} else {
    536 		MCLAIM(mn, m->m_owner);
    537 	}
    538 	mn->m_next = m;
    539 	m = mn;
    540 	if (len < MHLEN)
    541 		MH_ALIGN(m, len);
    542 	m->m_len = len;
    543 	return (m);
    544 }
    545 
    546 /*
    547  * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
    548  * continuing for "len" bytes.  If len is M_COPYALL, copy to end of mbuf.
    549  * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
    550  */
    551 int MCFail;
    552 
    553 struct mbuf *
    554 m_copym(struct mbuf *m, int off0, int len, int wait)
    555 {
    556 
    557 	return m_copym0(m, off0, len, wait, 0);	/* shallow copy on M_EXT */
    558 }
    559 
    560 struct mbuf *
    561 m_dup(struct mbuf *m, int off0, int len, int wait)
    562 {
    563 
    564 	return m_copym0(m, off0, len, wait, 1);	/* deep copy */
    565 }
    566 
    567 static struct mbuf *
    568 m_copym0(struct mbuf *m, int off0, int len, int wait, int deep)
    569 {
    570 	struct mbuf *n, **np;
    571 	int off = off0;
    572 	struct mbuf *top;
    573 	int copyhdr = 0;
    574 
    575 	if (off < 0 || len < 0)
    576 		panic("m_copym: off %d, len %d", off, len);
    577 	if (off == 0 && m->m_flags & M_PKTHDR)
    578 		copyhdr = 1;
    579 	while (off > 0) {
    580 		if (m == 0)
    581 			panic("m_copym: m == 0, off %d", off);
    582 		if (off < m->m_len)
    583 			break;
    584 		off -= m->m_len;
    585 		m = m->m_next;
    586 	}
    587 	np = &top;
    588 	top = 0;
    589 	while (len > 0) {
    590 		if (m == 0) {
    591 			if (len != M_COPYALL)
    592 				panic("m_copym: m == 0, len %d [!COPYALL]",
    593 				    len);
    594 			break;
    595 		}
    596 		MGET(n, wait, m->m_type);
    597 		*np = n;
    598 		if (n == 0)
    599 			goto nospace;
    600 		MCLAIM(n, m->m_owner);
    601 		if (copyhdr) {
    602 			M_COPY_PKTHDR(n, m);
    603 			if (len == M_COPYALL)
    604 				n->m_pkthdr.len -= off0;
    605 			else
    606 				n->m_pkthdr.len = len;
    607 			copyhdr = 0;
    608 		}
    609 		n->m_len = min(len, m->m_len - off);
    610 		if (m->m_flags & M_EXT) {
    611 			if (!deep) {
    612 				n->m_data = m->m_data + off;
    613 				MCLADDREFERENCE(m, n);
    614 			} else {
    615 				/*
    616 				 * we are unsure about the way m was allocated.
    617 				 * copy into multiple MCLBYTES cluster mbufs.
    618 				 */
    619 				MCLGET(n, wait);
    620 				n->m_len = 0;
    621 				n->m_len = M_TRAILINGSPACE(n);
    622 				n->m_len = min(n->m_len, len);
    623 				n->m_len = min(n->m_len, m->m_len - off);
    624 				memcpy(mtod(n, void *), mtod(m, char *) + off,
    625 				    (unsigned)n->m_len);
    626 			}
    627 		} else
    628 			memcpy(mtod(n, void *), mtod(m, char *) + off,
    629 			    (unsigned)n->m_len);
    630 		if (len != M_COPYALL)
    631 			len -= n->m_len;
    632 		off += n->m_len;
    633 #ifdef DIAGNOSTIC
    634 		if (off > m->m_len)
    635 			panic("m_copym0 overrun");
    636 #endif
    637 		if (off == m->m_len) {
    638 			m = m->m_next;
    639 			off = 0;
    640 		}
    641 		np = &n->m_next;
    642 	}
    643 	if (top == 0)
    644 		MCFail++;
    645 	return (top);
    646 nospace:
    647 	m_freem(top);
    648 	MCFail++;
    649 	return (NULL);
    650 }
    651 
    652 /*
    653  * Copy an entire packet, including header (which must be present).
    654  * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
    655  */
    656 struct mbuf *
    657 m_copypacket(struct mbuf *m, int how)
    658 {
    659 	struct mbuf *top, *n, *o;
    660 
    661 	MGET(n, how, m->m_type);
    662 	top = n;
    663 	if (!n)
    664 		goto nospace;
    665 
    666 	MCLAIM(n, m->m_owner);
    667 	M_COPY_PKTHDR(n, m);
    668 	n->m_len = m->m_len;
    669 	if (m->m_flags & M_EXT) {
    670 		n->m_data = m->m_data;
    671 		MCLADDREFERENCE(m, n);
    672 	} else {
    673 		memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
    674 	}
    675 
    676 	m = m->m_next;
    677 	while (m) {
    678 		MGET(o, how, m->m_type);
    679 		if (!o)
    680 			goto nospace;
    681 
    682 		MCLAIM(o, m->m_owner);
    683 		n->m_next = o;
    684 		n = n->m_next;
    685 
    686 		n->m_len = m->m_len;
    687 		if (m->m_flags & M_EXT) {
    688 			n->m_data = m->m_data;
    689 			MCLADDREFERENCE(m, n);
    690 		} else {
    691 			memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
    692 		}
    693 
    694 		m = m->m_next;
    695 	}
    696 	return top;
    697 nospace:
    698 	m_freem(top);
    699 	MCFail++;
    700 	return NULL;
    701 }
    702 
    703 /*
    704  * Copy data from an mbuf chain starting "off" bytes from the beginning,
    705  * continuing for "len" bytes, into the indicated buffer.
    706  */
    707 void
    708 m_copydata(struct mbuf *m, int off, int len, void *vp)
    709 {
    710 	unsigned	count;
    711 	void *		cp = vp;
    712 
    713 	if (off < 0 || len < 0)
    714 		panic("m_copydata: off %d, len %d", off, len);
    715 	while (off > 0) {
    716 		if (m == NULL)
    717 			panic("m_copydata: m == NULL, off %d", off);
    718 		if (off < m->m_len)
    719 			break;
    720 		off -= m->m_len;
    721 		m = m->m_next;
    722 	}
    723 	while (len > 0) {
    724 		if (m == NULL)
    725 			panic("m_copydata: m == NULL, len %d", len);
    726 		count = min(m->m_len - off, len);
    727 		memcpy(cp, mtod(m, char *) + off, count);
    728 		len -= count;
    729 		cp = (char *)cp + count;
    730 		off = 0;
    731 		m = m->m_next;
    732 	}
    733 }
    734 
    735 /*
    736  * Concatenate mbuf chain n to m.
    737  * n might be copied into m (when n->m_len is small), therefore data portion of
    738  * n could be copied into an mbuf of different mbuf type.
    739  * Any m_pkthdr is not updated.
    740  */
    741 void
    742 m_cat(struct mbuf *m, struct mbuf *n)
    743 {
    744 
    745 	while (m->m_next)
    746 		m = m->m_next;
    747 	while (n) {
    748 		if (M_READONLY(m) || n->m_len > M_TRAILINGSPACE(m)) {
    749 			/* just join the two chains */
    750 			m->m_next = n;
    751 			return;
    752 		}
    753 		/* splat the data from one into the other */
    754 		memcpy(mtod(m, char *) + m->m_len, mtod(n, void *),
    755 		    (u_int)n->m_len);
    756 		m->m_len += n->m_len;
    757 		n = m_free(n);
    758 	}
    759 }
    760 
    761 void
    762 m_adj(struct mbuf *mp, int req_len)
    763 {
    764 	int len = req_len;
    765 	struct mbuf *m;
    766 	int count;
    767 
    768 	if ((m = mp) == NULL)
    769 		return;
    770 	if (len >= 0) {
    771 		/*
    772 		 * Trim from head.
    773 		 */
    774 		while (m != NULL && len > 0) {
    775 			if (m->m_len <= len) {
    776 				len -= m->m_len;
    777 				m->m_len = 0;
    778 				m = m->m_next;
    779 			} else {
    780 				m->m_len -= len;
    781 				m->m_data += len;
    782 				len = 0;
    783 			}
    784 		}
    785 		m = mp;
    786 		if (mp->m_flags & M_PKTHDR)
    787 			m->m_pkthdr.len -= (req_len - len);
    788 	} else {
    789 		/*
    790 		 * Trim from tail.  Scan the mbuf chain,
    791 		 * calculating its length and finding the last mbuf.
    792 		 * If the adjustment only affects this mbuf, then just
    793 		 * adjust and return.  Otherwise, rescan and truncate
    794 		 * after the remaining size.
    795 		 */
    796 		len = -len;
    797 		count = 0;
    798 		for (;;) {
    799 			count += m->m_len;
    800 			if (m->m_next == (struct mbuf *)0)
    801 				break;
    802 			m = m->m_next;
    803 		}
    804 		if (m->m_len >= len) {
    805 			m->m_len -= len;
    806 			if (mp->m_flags & M_PKTHDR)
    807 				mp->m_pkthdr.len -= len;
    808 			return;
    809 		}
    810 		count -= len;
    811 		if (count < 0)
    812 			count = 0;
    813 		/*
    814 		 * Correct length for chain is "count".
    815 		 * Find the mbuf with last data, adjust its length,
    816 		 * and toss data from remaining mbufs on chain.
    817 		 */
    818 		m = mp;
    819 		if (m->m_flags & M_PKTHDR)
    820 			m->m_pkthdr.len = count;
    821 		for (; m; m = m->m_next) {
    822 			if (m->m_len >= count) {
    823 				m->m_len = count;
    824 				break;
    825 			}
    826 			count -= m->m_len;
    827 		}
    828 		if (m)
    829 			while (m->m_next)
    830 				(m = m->m_next)->m_len = 0;
    831 	}
    832 }
    833 
    834 /*
    835  * Rearrange an mbuf chain so that len bytes are contiguous
    836  * and in the data area of an mbuf (so that mtod and dtom
    837  * will work for a structure of size len).  Returns the resulting
    838  * mbuf chain on success, frees it and returns null on failure.
    839  * If there is room, it will add up to max_protohdr-len extra bytes to the
    840  * contiguous region in an attempt to avoid being called next time.
    841  */
    842 int MPFail;
    843 
    844 struct mbuf *
    845 m_pullup(struct mbuf *n, int len)
    846 {
    847 	struct mbuf *m;
    848 	int count;
    849 	int space;
    850 
    851 	/*
    852 	 * If first mbuf has no cluster, and has room for len bytes
    853 	 * without shifting current data, pullup into it,
    854 	 * otherwise allocate a new mbuf to prepend to the chain.
    855 	 */
    856 	if ((n->m_flags & M_EXT) == 0 &&
    857 	    n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
    858 		if (n->m_len >= len)
    859 			return (n);
    860 		m = n;
    861 		n = n->m_next;
    862 		len -= m->m_len;
    863 	} else {
    864 		if (len > MHLEN)
    865 			goto bad;
    866 		MGET(m, M_DONTWAIT, n->m_type);
    867 		if (m == 0)
    868 			goto bad;
    869 		MCLAIM(m, n->m_owner);
    870 		m->m_len = 0;
    871 		if (n->m_flags & M_PKTHDR) {
    872 			M_MOVE_PKTHDR(m, n);
    873 		}
    874 	}
    875 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
    876 	do {
    877 		count = min(min(max(len, max_protohdr), space), n->m_len);
    878 		memcpy(mtod(m, char *) + m->m_len, mtod(n, void *),
    879 		  (unsigned)count);
    880 		len -= count;
    881 		m->m_len += count;
    882 		n->m_len -= count;
    883 		space -= count;
    884 		if (n->m_len)
    885 			n->m_data += count;
    886 		else
    887 			n = m_free(n);
    888 	} while (len > 0 && n);
    889 	if (len > 0) {
    890 		(void) m_free(m);
    891 		goto bad;
    892 	}
    893 	m->m_next = n;
    894 	return (m);
    895 bad:
    896 	m_freem(n);
    897 	MPFail++;
    898 	return (NULL);
    899 }
    900 
    901 /*
    902  * Like m_pullup(), except a new mbuf is always allocated, and we allow
    903  * the amount of empty space before the data in the new mbuf to be specified
    904  * (in the event that the caller expects to prepend later).
    905  */
    906 int MSFail;
    907 
    908 struct mbuf *
    909 m_copyup(struct mbuf *n, int len, int dstoff)
    910 {
    911 	struct mbuf *m;
    912 	int count, space;
    913 
    914 	if (len > (MHLEN - dstoff))
    915 		goto bad;
    916 	MGET(m, M_DONTWAIT, n->m_type);
    917 	if (m == NULL)
    918 		goto bad;
    919 	MCLAIM(m, n->m_owner);
    920 	m->m_len = 0;
    921 	if (n->m_flags & M_PKTHDR) {
    922 		M_MOVE_PKTHDR(m, n);
    923 	}
    924 	m->m_data += dstoff;
    925 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
    926 	do {
    927 		count = min(min(max(len, max_protohdr), space), n->m_len);
    928 		memcpy(mtod(m, char *) + m->m_len, mtod(n, void *),
    929 		    (unsigned)count);
    930 		len -= count;
    931 		m->m_len += count;
    932 		n->m_len -= count;
    933 		space -= count;
    934 		if (n->m_len)
    935 			n->m_data += count;
    936 		else
    937 			n = m_free(n);
    938 	} while (len > 0 && n);
    939 	if (len > 0) {
    940 		(void) m_free(m);
    941 		goto bad;
    942 	}
    943 	m->m_next = n;
    944 	return (m);
    945  bad:
    946 	m_freem(n);
    947 	MSFail++;
    948 	return (NULL);
    949 }
    950 
    951 /*
    952  * Partition an mbuf chain in two pieces, returning the tail --
    953  * all but the first len0 bytes.  In case of failure, it returns NULL and
    954  * attempts to restore the chain to its original state.
    955  */
    956 struct mbuf *
    957 m_split(struct mbuf *m0, int len0, int wait)
    958 {
    959 
    960 	return m_split0(m0, len0, wait, 1);
    961 }
    962 
    963 static struct mbuf *
    964 m_split0(struct mbuf *m0, int len0, int wait, int copyhdr)
    965 {
    966 	struct mbuf *m, *n;
    967 	unsigned len = len0, remain, len_save;
    968 
    969 	for (m = m0; m && len > m->m_len; m = m->m_next)
    970 		len -= m->m_len;
    971 	if (m == 0)
    972 		return (NULL);
    973 	remain = m->m_len - len;
    974 	if (copyhdr && (m0->m_flags & M_PKTHDR)) {
    975 		MGETHDR(n, wait, m0->m_type);
    976 		if (n == 0)
    977 			return (NULL);
    978 		MCLAIM(n, m0->m_owner);
    979 		n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
    980 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
    981 		len_save = m0->m_pkthdr.len;
    982 		m0->m_pkthdr.len = len0;
    983 		if (m->m_flags & M_EXT)
    984 			goto extpacket;
    985 		if (remain > MHLEN) {
    986 			/* m can't be the lead packet */
    987 			MH_ALIGN(n, 0);
    988 			n->m_next = m_split(m, len, wait);
    989 			if (n->m_next == 0) {
    990 				(void) m_free(n);
    991 				m0->m_pkthdr.len = len_save;
    992 				return (NULL);
    993 			} else
    994 				return (n);
    995 		} else
    996 			MH_ALIGN(n, remain);
    997 	} else if (remain == 0) {
    998 		n = m->m_next;
    999 		m->m_next = 0;
   1000 		return (n);
   1001 	} else {
   1002 		MGET(n, wait, m->m_type);
   1003 		if (n == 0)
   1004 			return (NULL);
   1005 		MCLAIM(n, m->m_owner);
   1006 		M_ALIGN(n, remain);
   1007 	}
   1008 extpacket:
   1009 	if (m->m_flags & M_EXT) {
   1010 		n->m_data = m->m_data + len;
   1011 		MCLADDREFERENCE(m, n);
   1012 	} else {
   1013 		memcpy(mtod(n, void *), mtod(m, char *) + len, remain);
   1014 	}
   1015 	n->m_len = remain;
   1016 	m->m_len = len;
   1017 	n->m_next = m->m_next;
   1018 	m->m_next = 0;
   1019 	return (n);
   1020 }
   1021 /*
   1022  * Routine to copy from device local memory into mbufs.
   1023  */
   1024 struct mbuf *
   1025 m_devget(char *buf, int totlen, int off0, struct ifnet *ifp,
   1026     void (*copy)(const void *from, void *to, size_t len))
   1027 {
   1028 	struct mbuf *m;
   1029 	struct mbuf *top = 0, **mp = &top;
   1030 	int off = off0, len;
   1031 	char *cp;
   1032 	char *epkt;
   1033 
   1034 	cp = buf;
   1035 	epkt = cp + totlen;
   1036 	if (off) {
   1037 		/*
   1038 		 * If 'off' is non-zero, packet is trailer-encapsulated,
   1039 		 * so we have to skip the type and length fields.
   1040 		 */
   1041 		cp += off + 2 * sizeof(uint16_t);
   1042 		totlen -= 2 * sizeof(uint16_t);
   1043 	}
   1044 	MGETHDR(m, M_DONTWAIT, MT_DATA);
   1045 	if (m == 0)
   1046 		return (NULL);
   1047 	m->m_pkthdr.rcvif = ifp;
   1048 	m->m_pkthdr.len = totlen;
   1049 	m->m_len = MHLEN;
   1050 
   1051 	while (totlen > 0) {
   1052 		if (top) {
   1053 			MGET(m, M_DONTWAIT, MT_DATA);
   1054 			if (m == 0) {
   1055 				m_freem(top);
   1056 				return (NULL);
   1057 			}
   1058 			m->m_len = MLEN;
   1059 		}
   1060 		len = min(totlen, epkt - cp);
   1061 		if (len >= MINCLSIZE) {
   1062 			MCLGET(m, M_DONTWAIT);
   1063 			if ((m->m_flags & M_EXT) == 0) {
   1064 				m_free(m);
   1065 				m_freem(top);
   1066 				return (NULL);
   1067 			}
   1068 			m->m_len = len = min(len, MCLBYTES);
   1069 		} else {
   1070 			/*
   1071 			 * Place initial small packet/header at end of mbuf.
   1072 			 */
   1073 			if (len < m->m_len) {
   1074 				if (top == 0 && len + max_linkhdr <= m->m_len)
   1075 					m->m_data += max_linkhdr;
   1076 				m->m_len = len;
   1077 			} else
   1078 				len = m->m_len;
   1079 		}
   1080 		if (copy)
   1081 			copy(cp, mtod(m, void *), (size_t)len);
   1082 		else
   1083 			memcpy(mtod(m, void *), cp, (size_t)len);
   1084 		cp += len;
   1085 		*mp = m;
   1086 		mp = &m->m_next;
   1087 		totlen -= len;
   1088 		if (cp == epkt)
   1089 			cp = buf;
   1090 	}
   1091 	return (top);
   1092 }
   1093 
   1094 /*
   1095  * Copy data from a buffer back into the indicated mbuf chain,
   1096  * starting "off" bytes from the beginning, extending the mbuf
   1097  * chain if necessary.
   1098  */
   1099 void
   1100 m_copyback(struct mbuf *m0, int off, int len, const void *cp)
   1101 {
   1102 #if defined(DEBUG)
   1103 	struct mbuf *origm = m0;
   1104 	int error;
   1105 #endif /* defined(DEBUG) */
   1106 
   1107 	if (m0 == NULL)
   1108 		return;
   1109 
   1110 #if defined(DEBUG)
   1111 	error =
   1112 #endif /* defined(DEBUG) */
   1113 	m_copyback0(&m0, off, len, cp,
   1114 	    M_COPYBACK0_COPYBACK|M_COPYBACK0_EXTEND, M_DONTWAIT);
   1115 
   1116 #if defined(DEBUG)
   1117 	if (error != 0 || (m0 != NULL && origm != m0))
   1118 		panic("m_copyback");
   1119 #endif /* defined(DEBUG) */
   1120 }
   1121 
   1122 struct mbuf *
   1123 m_copyback_cow(struct mbuf *m0, int off, int len, const void *cp, int how)
   1124 {
   1125 	int error;
   1126 
   1127 	/* don't support chain expansion */
   1128 	KDASSERT(off + len <= m_length(m0));
   1129 
   1130 	error = m_copyback0(&m0, off, len, cp,
   1131 	    M_COPYBACK0_COPYBACK|M_COPYBACK0_COW, how);
   1132 	if (error) {
   1133 		/*
   1134 		 * no way to recover from partial success.
   1135 		 * just free the chain.
   1136 		 */
   1137 		m_freem(m0);
   1138 		return NULL;
   1139 	}
   1140 	return m0;
   1141 }
   1142 
   1143 /*
   1144  * m_makewritable: ensure the specified range writable.
   1145  */
   1146 int
   1147 m_makewritable(struct mbuf **mp, int off, int len, int how)
   1148 {
   1149 	int error;
   1150 #if defined(DEBUG)
   1151 	struct mbuf *n;
   1152 	int origlen, reslen;
   1153 
   1154 	origlen = m_length(*mp);
   1155 #endif /* defined(DEBUG) */
   1156 
   1157 #if 0 /* M_COPYALL is large enough */
   1158 	if (len == M_COPYALL)
   1159 		len = m_length(*mp) - off; /* XXX */
   1160 #endif
   1161 
   1162 	error = m_copyback0(mp, off, len, NULL,
   1163 	    M_COPYBACK0_PRESERVE|M_COPYBACK0_COW, how);
   1164 
   1165 #if defined(DEBUG)
   1166 	reslen = 0;
   1167 	for (n = *mp; n; n = n->m_next)
   1168 		reslen += n->m_len;
   1169 	if (origlen != reslen)
   1170 		panic("m_makewritable: length changed");
   1171 	if (((*mp)->m_flags & M_PKTHDR) != 0 && reslen != (*mp)->m_pkthdr.len)
   1172 		panic("m_makewritable: inconsist");
   1173 #endif /* defined(DEBUG) */
   1174 
   1175 	return error;
   1176 }
   1177 
   1178 int
   1179 m_copyback0(struct mbuf **mp0, int off, int len, const void *vp, int flags,
   1180     int how)
   1181 {
   1182 	int mlen;
   1183 	struct mbuf *m, *n;
   1184 	struct mbuf **mp;
   1185 	int totlen = 0;
   1186 	const char *cp = vp;
   1187 
   1188 	KASSERT(mp0 != NULL);
   1189 	KASSERT(*mp0 != NULL);
   1190 	KASSERT((flags & M_COPYBACK0_PRESERVE) == 0 || cp == NULL);
   1191 	KASSERT((flags & M_COPYBACK0_COPYBACK) == 0 || cp != NULL);
   1192 
   1193 	/*
   1194 	 * we don't bother to update "totlen" in the case of M_COPYBACK0_COW,
   1195 	 * assuming that M_COPYBACK0_EXTEND and M_COPYBACK0_COW are exclusive.
   1196 	 */
   1197 
   1198 	KASSERT((~flags & (M_COPYBACK0_EXTEND|M_COPYBACK0_COW)) != 0);
   1199 
   1200 	mp = mp0;
   1201 	m = *mp;
   1202 	while (off > (mlen = m->m_len)) {
   1203 		off -= mlen;
   1204 		totlen += mlen;
   1205 		if (m->m_next == NULL) {
   1206 			int tspace;
   1207 extend:
   1208 			if ((flags & M_COPYBACK0_EXTEND) == 0)
   1209 				goto out;
   1210 
   1211 			/*
   1212 			 * try to make some space at the end of "m".
   1213 			 */
   1214 
   1215 			mlen = m->m_len;
   1216 			if (off + len >= MINCLSIZE &&
   1217 			    (m->m_flags & M_EXT) == 0 && m->m_len == 0) {
   1218 				MCLGET(m, how);
   1219 			}
   1220 			tspace = M_TRAILINGSPACE(m);
   1221 			if (tspace > 0) {
   1222 				tspace = min(tspace, off + len);
   1223 				KASSERT(tspace > 0);
   1224 				memset(mtod(m, char *) + m->m_len, 0,
   1225 				    min(off, tspace));
   1226 				m->m_len += tspace;
   1227 				off += mlen;
   1228 				totlen -= mlen;
   1229 				continue;
   1230 			}
   1231 
   1232 			/*
   1233 			 * need to allocate an mbuf.
   1234 			 */
   1235 
   1236 			if (off + len >= MINCLSIZE) {
   1237 				n = m_getcl(how, m->m_type, 0);
   1238 			} else {
   1239 				n = m_get(how, m->m_type);
   1240 			}
   1241 			if (n == NULL) {
   1242 				goto out;
   1243 			}
   1244 			n->m_len = 0;
   1245 			n->m_len = min(M_TRAILINGSPACE(n), off + len);
   1246 			memset(mtod(n, char *), 0, min(n->m_len, off));
   1247 			m->m_next = n;
   1248 		}
   1249 		mp = &m->m_next;
   1250 		m = m->m_next;
   1251 	}
   1252 	while (len > 0) {
   1253 		mlen = m->m_len - off;
   1254 		if (mlen != 0 && M_READONLY(m)) {
   1255 			char *datap;
   1256 			int eatlen;
   1257 
   1258 			/*
   1259 			 * this mbuf is read-only.
   1260 			 * allocate a new writable mbuf and try again.
   1261 			 */
   1262 
   1263 #if defined(DIAGNOSTIC)
   1264 			if ((flags & M_COPYBACK0_COW) == 0)
   1265 				panic("m_copyback0: read-only");
   1266 #endif /* defined(DIAGNOSTIC) */
   1267 
   1268 			/*
   1269 			 * if we're going to write into the middle of
   1270 			 * a mbuf, split it first.
   1271 			 */
   1272 			if (off > 0 && len < mlen) {
   1273 				n = m_split0(m, off, how, 0);
   1274 				if (n == NULL)
   1275 					goto enobufs;
   1276 				m->m_next = n;
   1277 				mp = &m->m_next;
   1278 				m = n;
   1279 				off = 0;
   1280 				continue;
   1281 			}
   1282 
   1283 			/*
   1284 			 * XXX TODO coalesce into the trailingspace of
   1285 			 * the previous mbuf when possible.
   1286 			 */
   1287 
   1288 			/*
   1289 			 * allocate a new mbuf.  copy packet header if needed.
   1290 			 */
   1291 			MGET(n, how, m->m_type);
   1292 			if (n == NULL)
   1293 				goto enobufs;
   1294 			MCLAIM(n, m->m_owner);
   1295 			if (off == 0 && (m->m_flags & M_PKTHDR) != 0) {
   1296 				M_MOVE_PKTHDR(n, m);
   1297 				n->m_len = MHLEN;
   1298 			} else {
   1299 				if (len >= MINCLSIZE)
   1300 					MCLGET(n, M_DONTWAIT);
   1301 				n->m_len =
   1302 				    (n->m_flags & M_EXT) ? MCLBYTES : MLEN;
   1303 			}
   1304 			if (n->m_len > len)
   1305 				n->m_len = len;
   1306 
   1307 			/*
   1308 			 * free the region which has been overwritten.
   1309 			 * copying data from old mbufs if requested.
   1310 			 */
   1311 			if (flags & M_COPYBACK0_PRESERVE)
   1312 				datap = mtod(n, char *);
   1313 			else
   1314 				datap = NULL;
   1315 			eatlen = n->m_len;
   1316 			KDASSERT(off == 0 || eatlen >= mlen);
   1317 			if (off > 0) {
   1318 				KDASSERT(len >= mlen);
   1319 				m->m_len = off;
   1320 				m->m_next = n;
   1321 				if (datap) {
   1322 					m_copydata(m, off, mlen, datap);
   1323 					datap += mlen;
   1324 				}
   1325 				eatlen -= mlen;
   1326 				mp = &m->m_next;
   1327 				m = m->m_next;
   1328 			}
   1329 			while (m != NULL && M_READONLY(m) &&
   1330 			    n->m_type == m->m_type && eatlen > 0) {
   1331 				mlen = min(eatlen, m->m_len);
   1332 				if (datap) {
   1333 					m_copydata(m, 0, mlen, datap);
   1334 					datap += mlen;
   1335 				}
   1336 				m->m_data += mlen;
   1337 				m->m_len -= mlen;
   1338 				eatlen -= mlen;
   1339 				if (m->m_len == 0)
   1340 					*mp = m = m_free(m);
   1341 			}
   1342 			if (eatlen > 0)
   1343 				n->m_len -= eatlen;
   1344 			n->m_next = m;
   1345 			*mp = m = n;
   1346 			continue;
   1347 		}
   1348 		mlen = min(mlen, len);
   1349 		if (flags & M_COPYBACK0_COPYBACK) {
   1350 			memcpy(mtod(m, char *) + off, cp, (unsigned)mlen);
   1351 			cp += mlen;
   1352 		}
   1353 		len -= mlen;
   1354 		mlen += off;
   1355 		off = 0;
   1356 		totlen += mlen;
   1357 		if (len == 0)
   1358 			break;
   1359 		if (m->m_next == NULL) {
   1360 			goto extend;
   1361 		}
   1362 		mp = &m->m_next;
   1363 		m = m->m_next;
   1364 	}
   1365 out:	if (((m = *mp0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) {
   1366 		KASSERT((flags & M_COPYBACK0_EXTEND) != 0);
   1367 		m->m_pkthdr.len = totlen;
   1368 	}
   1369 
   1370 	return 0;
   1371 
   1372 enobufs:
   1373 	return ENOBUFS;
   1374 }
   1375 
   1376 void
   1377 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
   1378 {
   1379 
   1380 	KASSERT((to->m_flags & M_EXT) == 0);
   1381 	KASSERT((to->m_flags & M_PKTHDR) == 0 || m_tag_first(to) == NULL);
   1382 	KASSERT((from->m_flags & M_PKTHDR) != 0);
   1383 
   1384 	to->m_pkthdr = from->m_pkthdr;
   1385 	to->m_flags = from->m_flags & M_COPYFLAGS;
   1386 	to->m_data = to->m_pktdat;
   1387 
   1388 	from->m_flags &= ~M_PKTHDR;
   1389 }
   1390 
   1391 /*
   1392  * Apply function f to the data in an mbuf chain starting "off" bytes from the
   1393  * beginning, continuing for "len" bytes.
   1394  */
   1395 int
   1396 m_apply(struct mbuf *m, int off, int len,
   1397     int (*f)(void *, void *, unsigned int), void *arg)
   1398 {
   1399 	unsigned int count;
   1400 	int rval;
   1401 
   1402 	KASSERT(len >= 0);
   1403 	KASSERT(off >= 0);
   1404 
   1405 	while (off > 0) {
   1406 		KASSERT(m != NULL);
   1407 		if (off < m->m_len)
   1408 			break;
   1409 		off -= m->m_len;
   1410 		m = m->m_next;
   1411 	}
   1412 	while (len > 0) {
   1413 		KASSERT(m != NULL);
   1414 		count = min(m->m_len - off, len);
   1415 
   1416 		rval = (*f)(arg, mtod(m, char *) + off, count);
   1417 		if (rval)
   1418 			return (rval);
   1419 
   1420 		len -= count;
   1421 		off = 0;
   1422 		m = m->m_next;
   1423 	}
   1424 
   1425 	return (0);
   1426 }
   1427 
   1428 /*
   1429  * Return a pointer to mbuf/offset of location in mbuf chain.
   1430  */
   1431 struct mbuf *
   1432 m_getptr(struct mbuf *m, int loc, int *off)
   1433 {
   1434 
   1435 	while (loc >= 0) {
   1436 		/* Normal end of search */
   1437 		if (m->m_len > loc) {
   1438 	    		*off = loc;
   1439 	    		return (m);
   1440 		} else {
   1441 	    		loc -= m->m_len;
   1442 
   1443 	    		if (m->m_next == NULL) {
   1444 				if (loc == 0) {
   1445  					/* Point at the end of valid data */
   1446 		    			*off = m->m_len;
   1447 		    			return (m);
   1448 				} else
   1449 		  			return (NULL);
   1450 	    		} else
   1451 	      			m = m->m_next;
   1452 		}
   1453     	}
   1454 
   1455 	return (NULL);
   1456 }
   1457 
   1458 /*
   1459  * m_ext_free: release a reference to the mbuf external storage.
   1460  *
   1461  * => free the mbuf m itsself as well.
   1462  * => called at splvm.
   1463  */
   1464 
   1465 void
   1466 m_ext_free(struct mbuf *m)
   1467 {
   1468 	bool embedded = MEXT_ISEMBEDDED(m);
   1469 	bool dofree = true;
   1470 
   1471 	KASSERT((m->m_flags & M_EXT) != 0);
   1472 	KASSERT(MEXT_ISEMBEDDED(m->m_ext_ref));
   1473 	KASSERT((m->m_ext_ref->m_flags & M_EXT) != 0);
   1474 	KASSERT((m->m_flags & M_EXT_CLUSTER) ==
   1475 	    (m->m_ext_ref->m_flags & M_EXT_CLUSTER));
   1476 
   1477 	if (!mcl_dec_and_test_reference(m)) {
   1478 		if (embedded) {
   1479 			/*
   1480 			 * other mbuf's m_ext_ref still points to us.
   1481 			 */
   1482 			dofree = false;
   1483 		} else {
   1484 			m->m_ext_ref = m;
   1485 		}
   1486 	} else {
   1487 		/*
   1488 		 * dropping the last reference
   1489 		 */
   1490 		if (!embedded) {
   1491 			mcl_inc_reference(m); /* XXX */
   1492 			m_ext_free(m->m_ext_ref);
   1493 			m->m_ext_ref = m;
   1494 		} else if ((m->m_flags & M_EXT_CLUSTER) != 0) {
   1495 			pool_cache_put_paddr((struct pool_cache *)
   1496 			    m->m_ext.ext_arg,
   1497 			    m->m_ext.ext_buf, m->m_ext.ext_paddr);
   1498 		} else if (m->m_ext.ext_free) {
   1499 			mutex_destroy(&m->m_ext.ext_lock);
   1500 			(*m->m_ext.ext_free)(m,
   1501 			    m->m_ext.ext_buf, m->m_ext.ext_size,
   1502 			    m->m_ext.ext_arg);
   1503 			/*
   1504 			 * 'm' is already freed by the ext_free callback.
   1505 			 */
   1506 			dofree = false;
   1507 		} else {
   1508 			free(m->m_ext.ext_buf, m->m_ext.ext_type);
   1509 		}
   1510 	}
   1511 	if (dofree) {
   1512 		if (embedded) {
   1513 			mutex_destroy(&m->m_ext.ext_lock);
   1514 		}
   1515 		pool_cache_put(&mbpool_cache, m);
   1516 	}
   1517 }
   1518 
   1519 #if defined(__HAVE_LAZY_MBUF) || defined(DEBUG)
   1520 char *
   1521 m_mapin(struct mbuf *m)
   1522 {
   1523 
   1524 #if defined(__HAVE_LAZY_MBUF)
   1525 	KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES|M_EXT_LAZY)) == 0);
   1526 
   1527 	MEXT_LOCK(m);
   1528 	if (m->m_ext.ext_flags & M_EXT_LAZY) {
   1529 		vaddr_t buf = (vaddr_t)m->m_ext.ext_buf;
   1530 		vsize_t size = (vsize_t)m->m_ext.ext_size;
   1531 		vaddr_t va, sva, eva;
   1532 		int i;
   1533 
   1534 		sva = trunc_page(buf);
   1535 		eva = round_page(buf + size);
   1536 
   1537 		for (i = 0, va = sva; va < eva; i++, va += PAGE_SIZE) {
   1538 			pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
   1539 			    VM_PROT_READ);
   1540 		}
   1541 		pmap_update(pmap_kernel());
   1542 		m->m_ext.ext_flags &= ~M_EXT_LAZY;
   1543 	}
   1544 	MEXT_UNLOCK(m);
   1545 
   1546 	m->m_flags &= ~M_EXT_LAZY;
   1547 	return m->m_data;
   1548 #else /* defined(__HAVE_LAZY_MBUF) */
   1549 	panic("m_mapin");
   1550 #endif /* defined(__HAVE_LAZY_MBUF) */
   1551 }
   1552 #endif /* defined(__HAVE_LAZY_MBUF) || defined(DEBUG) */
   1553 
   1554 #if defined(DDB)
   1555 void
   1556 m_print(const struct mbuf *m, const char *modif, void (*pr)(const char *, ...))
   1557 {
   1558 	char ch;
   1559 	bool opt_c = false;
   1560 	char buf[512];
   1561 
   1562 	while ((ch = *(modif++)) != '\0') {
   1563 		switch (ch) {
   1564 		case 'c':
   1565 			opt_c = true;
   1566 			break;
   1567 		}
   1568 	}
   1569 
   1570 nextchain:
   1571 	(*pr)("MBUF %p\n", m);
   1572 	bitmask_snprintf(m->m_flags, M_FLAGS_BITS, buf, sizeof(buf));
   1573 	(*pr)("  data=%p, len=%d, type=%d, flags=0x%s\n",
   1574 	    m->m_data, m->m_len, m->m_type, buf);
   1575 	(*pr)("  owner=%p, next=%p, nextpkt=%p\n", m->m_owner, m->m_next,
   1576 	    m->m_nextpkt);
   1577 	(*pr)("  leadingspace=%u, trailingspace=%u, readonly=%u\n",
   1578 	    (int)M_LEADINGSPACE(m), (int)M_TRAILINGSPACE(m),
   1579 	    (int)M_READONLY(m));
   1580 	if ((m->m_flags & M_PKTHDR) != 0) {
   1581 		bitmask_snprintf(m->m_pkthdr.csum_flags, M_CSUM_BITS, buf,
   1582 		    sizeof(buf));
   1583 		(*pr)("  pktlen=%d, rcvif=%p, csum_flags=0x%s, csum_data=0x%"
   1584 		    PRIx32 ", segsz=%u\n",
   1585 		    m->m_pkthdr.len, m->m_pkthdr.rcvif,
   1586 		    buf, m->m_pkthdr.csum_data, m->m_pkthdr.segsz);
   1587 	}
   1588 	if ((m->m_flags & M_EXT)) {
   1589 		(*pr)("  ext_refcnt=%u, ext_buf=%p, ext_size=%zd, "
   1590 		    "ext_free=%p, ext_arg=%p\n",
   1591 		    m->m_ext.ext_refcnt,
   1592 		    m->m_ext.ext_buf, m->m_ext.ext_size,
   1593 		    m->m_ext.ext_free, m->m_ext.ext_arg);
   1594 	}
   1595 	if ((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0) {
   1596 		vaddr_t sva = (vaddr_t)m->m_ext.ext_buf;
   1597 		vaddr_t eva = sva + m->m_ext.ext_size;
   1598 		int n = (round_page(eva) - trunc_page(sva)) >> PAGE_SHIFT;
   1599 		int i;
   1600 
   1601 		(*pr)("  pages:");
   1602 		for (i = 0; i < n; i ++) {
   1603 			(*pr)(" %p", m->m_ext.ext_pgs[i]);
   1604 		}
   1605 		(*pr)("\n");
   1606 	}
   1607 
   1608 	if (opt_c) {
   1609 		m = m->m_next;
   1610 		if (m != NULL) {
   1611 			goto nextchain;
   1612 		}
   1613 	}
   1614 }
   1615 #endif /* defined(DDB) */
   1616