Home | History | Annotate | Line # | Download | only in kern
uipc_mbuf.c revision 1.100.2.14
      1 /*	$NetBSD: uipc_mbuf.c,v 1.100.2.14 2007/09/03 14:58:03 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 1.100.2.14 2007/09/03 14:58:03 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)	simple_lock(&(m)->m_ext.ext_lock)
    149 #define	MEXT_UNLOCK(m)	simple_unlock(&(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 	int s;
    156 
    157 	s = splvm();
    158 	MEXT_LOCK(m);
    159 	m->m_ext.ext_refcnt++;
    160 	MEXT_UNLOCK(m);
    161 	splx(s);
    162 }
    163 
    164 static inline bool
    165 mcl_dec_and_test_reference(struct mbuf *m)
    166 {
    167 	bool gotzero;
    168 	int s;
    169 
    170 	s = splvm();
    171 	MEXT_LOCK(m);
    172 	KASSERT(m->m_ext.ext_refcnt > 0);
    173 	m->m_ext.ext_refcnt--;
    174 	gotzero = (m->m_ext.ext_refcnt == 0);
    175 	MEXT_UNLOCK(m);
    176 	splx(s);
    177 
    178 	return gotzero;
    179 }
    180 
    181 #define	MCLADDREFERENCE(o, n)						\
    182 do {									\
    183 	KASSERT(((o)->m_flags & M_EXT) != 0);				\
    184 	KASSERT(((n)->m_flags & M_EXT) == 0);				\
    185 	KASSERT((o)->m_ext.ext_refcnt >= 1);				\
    186 	(n)->m_flags |= ((o)->m_flags & M_EXTCOPYFLAGS);		\
    187 	mcl_inc_reference((o));						\
    188 	(n)->m_ext_ref = (o)->m_ext_ref;				\
    189 	_MOWNERREF((n), (n)->m_flags);					\
    190 	MCLREFDEBUGN((n), __FILE__, __LINE__);				\
    191 } while (/* CONSTCOND */ 0)
    192 
    193 /*
    194  * Initialize the mbuf allocator.
    195  */
    196 void
    197 mbinit(void)
    198 {
    199 
    200 	KASSERT(sizeof(struct _m_ext) <= MHLEN);
    201 	KASSERT(sizeof(struct mbuf) == MSIZE);
    202 
    203 	mclpool_allocator.pa_backingmap = mb_map;
    204 	pool_init(&mbpool, msize, 0, 0, 0, "mbpl", NULL, IPL_VM);
    205 	pool_init(&mclpool, mclbytes, 0, 0, 0, "mclpl", &mclpool_allocator,
    206 	    IPL_VM);
    207 
    208 	pool_set_drain_hook(&mbpool, m_reclaim, NULL);
    209 	pool_set_drain_hook(&mclpool, m_reclaim, NULL);
    210 
    211 	pool_cache_init(&mbpool_cache, &mbpool, mb_ctor, NULL, NULL);
    212 	pool_cache_init(&mclpool_cache, &mclpool, NULL, NULL, NULL);
    213 
    214 	/*
    215 	 * Set the hard limit on the mclpool to the number of
    216 	 * mbuf clusters the kernel is to support.  Log the limit
    217 	 * reached message max once a minute.
    218 	 */
    219 	pool_sethardlimit(&mclpool, nmbclusters, mclpool_warnmsg, 60);
    220 
    221 	/*
    222 	 * Set a low water mark for both mbufs and clusters.  This should
    223 	 * help ensure that they can be allocated in a memory starvation
    224 	 * situation.  This is important for e.g. diskless systems which
    225 	 * must allocate mbufs in order for the pagedaemon to clean pages.
    226 	 */
    227 	pool_setlowat(&mbpool, mblowat);
    228 	pool_setlowat(&mclpool, mcllowat);
    229 
    230 #ifdef MBUFTRACE
    231 	{
    232 		/*
    233 		 * Attach the unknown mowners.
    234 		 */
    235 		int i;
    236 		MOWNER_ATTACH(&revoked_mowner);
    237 		for (i = sizeof(unknown_mowners)/sizeof(unknown_mowners[0]);
    238 		     i-- > 0; )
    239 			MOWNER_ATTACH(&unknown_mowners[i]);
    240 	}
    241 #endif
    242 }
    243 
    244 /*
    245  * sysctl helper routine for the kern.mbuf subtree.  nmbclusters may
    246  * or may not be writable, and mblowat and mcllowat need range
    247  * checking and pool tweaking after being reset.
    248  */
    249 static int
    250 sysctl_kern_mbuf(SYSCTLFN_ARGS)
    251 {
    252 	int error, newval;
    253 	struct sysctlnode node;
    254 
    255 	node = *rnode;
    256 	node.sysctl_data = &newval;
    257 	switch (rnode->sysctl_num) {
    258 	case MBUF_NMBCLUSTERS:
    259 		if (mb_map != NULL) {
    260 			node.sysctl_flags &= ~CTLFLAG_READWRITE;
    261 			node.sysctl_flags |= CTLFLAG_READONLY;
    262 		}
    263 		/* FALLTHROUGH */
    264 	case MBUF_MBLOWAT:
    265 	case MBUF_MCLLOWAT:
    266 		newval = *(int*)rnode->sysctl_data;
    267 		break;
    268 	default:
    269 		return (EOPNOTSUPP);
    270 	}
    271 
    272 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
    273 	if (error || newp == NULL)
    274 		return (error);
    275 	if (newval < 0)
    276 		return (EINVAL);
    277 
    278 	switch (node.sysctl_num) {
    279 	case MBUF_NMBCLUSTERS:
    280 		if (newval < nmbclusters)
    281 			return (EINVAL);
    282 		nmbclusters = newval;
    283 		pool_sethardlimit(&mclpool, nmbclusters, mclpool_warnmsg, 60);
    284 		break;
    285 	case MBUF_MBLOWAT:
    286 		mblowat = newval;
    287 		pool_setlowat(&mbpool, mblowat);
    288 		break;
    289 	case MBUF_MCLLOWAT:
    290 		mcllowat = newval;
    291 		pool_setlowat(&mclpool, mcllowat);
    292 		break;
    293 	}
    294 
    295 	return (0);
    296 }
    297 
    298 #ifdef MBUFTRACE
    299 static int
    300 sysctl_kern_mbuf_mowners(SYSCTLFN_ARGS)
    301 {
    302 	struct mowner *mo;
    303 	size_t len = 0;
    304 	int error = 0;
    305 
    306 	if (namelen != 0)
    307 		return (EINVAL);
    308 	if (newp != NULL)
    309 		return (EPERM);
    310 
    311 	LIST_FOREACH(mo, &mowners, mo_link) {
    312 		if (oldp != NULL) {
    313 			if (*oldlenp - len < sizeof(*mo)) {
    314 				error = ENOMEM;
    315 				break;
    316 			}
    317 			error = copyout(mo, (char *)oldp + len, sizeof(*mo));
    318 			if (error)
    319 				break;
    320 		}
    321 		len += sizeof(*mo);
    322 	}
    323 
    324 	if (error == 0)
    325 		*oldlenp = len;
    326 
    327 	return (error);
    328 }
    329 #endif /* MBUFTRACE */
    330 
    331 SYSCTL_SETUP(sysctl_kern_mbuf_setup, "sysctl kern.mbuf subtree setup")
    332 {
    333 
    334 	sysctl_createv(clog, 0, NULL, NULL,
    335 		       CTLFLAG_PERMANENT,
    336 		       CTLTYPE_NODE, "kern", NULL,
    337 		       NULL, 0, NULL, 0,
    338 		       CTL_KERN, CTL_EOL);
    339 	sysctl_createv(clog, 0, NULL, NULL,
    340 		       CTLFLAG_PERMANENT,
    341 		       CTLTYPE_NODE, "mbuf",
    342 		       SYSCTL_DESCR("mbuf control variables"),
    343 		       NULL, 0, NULL, 0,
    344 		       CTL_KERN, KERN_MBUF, CTL_EOL);
    345 
    346 	sysctl_createv(clog, 0, NULL, NULL,
    347 		       CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
    348 		       CTLTYPE_INT, "msize",
    349 		       SYSCTL_DESCR("mbuf base size"),
    350 		       NULL, msize, NULL, 0,
    351 		       CTL_KERN, KERN_MBUF, MBUF_MSIZE, CTL_EOL);
    352 	sysctl_createv(clog, 0, NULL, NULL,
    353 		       CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
    354 		       CTLTYPE_INT, "mclbytes",
    355 		       SYSCTL_DESCR("mbuf cluster size"),
    356 		       NULL, mclbytes, NULL, 0,
    357 		       CTL_KERN, KERN_MBUF, MBUF_MCLBYTES, CTL_EOL);
    358 	sysctl_createv(clog, 0, NULL, NULL,
    359 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    360 		       CTLTYPE_INT, "nmbclusters",
    361 		       SYSCTL_DESCR("Limit on the number of mbuf clusters"),
    362 		       sysctl_kern_mbuf, 0, &nmbclusters, 0,
    363 		       CTL_KERN, KERN_MBUF, MBUF_NMBCLUSTERS, CTL_EOL);
    364 	sysctl_createv(clog, 0, NULL, NULL,
    365 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    366 		       CTLTYPE_INT, "mblowat",
    367 		       SYSCTL_DESCR("mbuf low water mark"),
    368 		       sysctl_kern_mbuf, 0, &mblowat, 0,
    369 		       CTL_KERN, KERN_MBUF, MBUF_MBLOWAT, CTL_EOL);
    370 	sysctl_createv(clog, 0, NULL, NULL,
    371 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    372 		       CTLTYPE_INT, "mcllowat",
    373 		       SYSCTL_DESCR("mbuf cluster low water mark"),
    374 		       sysctl_kern_mbuf, 0, &mcllowat, 0,
    375 		       CTL_KERN, KERN_MBUF, MBUF_MCLLOWAT, CTL_EOL);
    376 	sysctl_createv(clog, 0, NULL, NULL,
    377 		       CTLFLAG_PERMANENT,
    378 		       CTLTYPE_STRUCT, "stats",
    379 		       SYSCTL_DESCR("mbuf allocation statistics"),
    380 		       NULL, 0, &mbstat, sizeof(mbstat),
    381 		       CTL_KERN, KERN_MBUF, MBUF_STATS, CTL_EOL);
    382 #ifdef MBUFTRACE
    383 	sysctl_createv(clog, 0, NULL, NULL,
    384 		       CTLFLAG_PERMANENT,
    385 		       CTLTYPE_STRUCT, "mowners",
    386 		       SYSCTL_DESCR("Information about mbuf owners"),
    387 		       sysctl_kern_mbuf_mowners, 0, NULL, 0,
    388 		       CTL_KERN, KERN_MBUF, MBUF_MOWNERS, CTL_EOL);
    389 #endif /* MBUFTRACE */
    390 }
    391 
    392 static void *
    393 mclpool_alloc(struct pool *pp, int flags)
    394 {
    395 	bool waitok = (flags & PR_WAITOK) ? true : false;
    396 
    397 	return ((void *)uvm_km_alloc_poolpage(mb_map, waitok));
    398 }
    399 
    400 static void
    401 mclpool_release(struct pool *pp, void *v)
    402 {
    403 
    404 	uvm_km_free_poolpage(mb_map, (vaddr_t)v);
    405 }
    406 
    407 /*ARGSUSED*/
    408 static int
    409 mb_ctor(void *arg, void *object, int flags)
    410 {
    411 	struct mbuf *m = object;
    412 
    413 #ifdef POOL_VTOPHYS
    414 	m->m_paddr = POOL_VTOPHYS(m);
    415 #else
    416 	m->m_paddr = M_PADDR_INVALID;
    417 #endif
    418 	return (0);
    419 }
    420 
    421 void
    422 m_reclaim(void *arg, int flags)
    423 {
    424 	struct domain *dp;
    425 	const struct protosw *pr;
    426 	struct ifnet *ifp;
    427 	int s = splvm();
    428 
    429 	DOMAIN_FOREACH(dp) {
    430 		for (pr = dp->dom_protosw;
    431 		     pr < dp->dom_protoswNPROTOSW; pr++)
    432 			if (pr->pr_drain)
    433 				(*pr->pr_drain)();
    434 	}
    435 	IFNET_FOREACH(ifp) {
    436 		if (ifp->if_drain)
    437 			(*ifp->if_drain)(ifp);
    438 	}
    439 	splx(s);
    440 	mbstat.m_drain++;
    441 }
    442 
    443 /*
    444  * Space allocation routines.
    445  * These are also available as macros
    446  * for critical paths.
    447  */
    448 struct mbuf *
    449 m_get(int nowait, int type)
    450 {
    451 	struct mbuf *m;
    452 
    453 	MGET(m, nowait, type);
    454 	return (m);
    455 }
    456 
    457 struct mbuf *
    458 m_gethdr(int nowait, int type)
    459 {
    460 	struct mbuf *m;
    461 
    462 	MGETHDR(m, nowait, type);
    463 	return (m);
    464 }
    465 
    466 struct mbuf *
    467 m_getclr(int nowait, int type)
    468 {
    469 	struct mbuf *m;
    470 
    471 	MGET(m, nowait, type);
    472 	if (m == 0)
    473 		return (NULL);
    474 	memset(mtod(m, void *), 0, MLEN);
    475 	return (m);
    476 }
    477 
    478 void
    479 m_clget(struct mbuf *m, int nowait)
    480 {
    481 
    482 	MCLGET(m, nowait);
    483 }
    484 
    485 struct mbuf *
    486 m_free(struct mbuf *m)
    487 {
    488 	struct mbuf *n;
    489 
    490 	MFREE(m, n);
    491 	return (n);
    492 }
    493 
    494 void
    495 m_freem(struct mbuf *m)
    496 {
    497 	struct mbuf *n;
    498 
    499 	if (m == NULL)
    500 		return;
    501 	do {
    502 		MFREE(m, n);
    503 		m = n;
    504 	} while (m);
    505 }
    506 
    507 #ifdef MBUFTRACE
    508 /*
    509  * Walk a chain of mbufs, claiming ownership of each mbuf in the chain.
    510  */
    511 void
    512 m_claimm(struct mbuf *m, struct mowner *mo)
    513 {
    514 
    515 	for (; m != NULL; m = m->m_next)
    516 		MCLAIM(m, mo);
    517 }
    518 #endif
    519 
    520 /*
    521  * Mbuffer utility routines.
    522  */
    523 
    524 /*
    525  * Lesser-used path for M_PREPEND:
    526  * allocate new mbuf to prepend to chain,
    527  * copy junk along.
    528  */
    529 struct mbuf *
    530 m_prepend(struct mbuf *m, int len, int how)
    531 {
    532 	struct mbuf *mn;
    533 
    534 	MGET(mn, how, m->m_type);
    535 	if (mn == (struct mbuf *)NULL) {
    536 		m_freem(m);
    537 		return ((struct mbuf *)NULL);
    538 	}
    539 	if (m->m_flags & M_PKTHDR) {
    540 		M_MOVE_PKTHDR(mn, m);
    541 	} else {
    542 		MCLAIM(mn, m->m_owner);
    543 	}
    544 	mn->m_next = m;
    545 	m = mn;
    546 	if (len < MHLEN)
    547 		MH_ALIGN(m, len);
    548 	m->m_len = len;
    549 	return (m);
    550 }
    551 
    552 /*
    553  * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
    554  * continuing for "len" bytes.  If len is M_COPYALL, copy to end of mbuf.
    555  * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
    556  */
    557 int MCFail;
    558 
    559 struct mbuf *
    560 m_copym(struct mbuf *m, int off0, int len, int wait)
    561 {
    562 
    563 	return m_copym0(m, off0, len, wait, 0);	/* shallow copy on M_EXT */
    564 }
    565 
    566 struct mbuf *
    567 m_dup(struct mbuf *m, int off0, int len, int wait)
    568 {
    569 
    570 	return m_copym0(m, off0, len, wait, 1);	/* deep copy */
    571 }
    572 
    573 static struct mbuf *
    574 m_copym0(struct mbuf *m, int off0, int len, int wait, int deep)
    575 {
    576 	struct mbuf *n, **np;
    577 	int off = off0;
    578 	struct mbuf *top;
    579 	int copyhdr = 0;
    580 
    581 	if (off < 0 || len < 0)
    582 		panic("m_copym: off %d, len %d", off, len);
    583 	if (off == 0 && m->m_flags & M_PKTHDR)
    584 		copyhdr = 1;
    585 	while (off > 0) {
    586 		if (m == 0)
    587 			panic("m_copym: m == 0, off %d", off);
    588 		if (off < m->m_len)
    589 			break;
    590 		off -= m->m_len;
    591 		m = m->m_next;
    592 	}
    593 	np = &top;
    594 	top = 0;
    595 	while (len > 0) {
    596 		if (m == 0) {
    597 			if (len != M_COPYALL)
    598 				panic("m_copym: m == 0, len %d [!COPYALL]",
    599 				    len);
    600 			break;
    601 		}
    602 		MGET(n, wait, m->m_type);
    603 		*np = n;
    604 		if (n == 0)
    605 			goto nospace;
    606 		MCLAIM(n, m->m_owner);
    607 		if (copyhdr) {
    608 			M_COPY_PKTHDR(n, m);
    609 			if (len == M_COPYALL)
    610 				n->m_pkthdr.len -= off0;
    611 			else
    612 				n->m_pkthdr.len = len;
    613 			copyhdr = 0;
    614 		}
    615 		n->m_len = min(len, m->m_len - off);
    616 		if (m->m_flags & M_EXT) {
    617 			if (!deep) {
    618 				n->m_data = m->m_data + off;
    619 				MCLADDREFERENCE(m, n);
    620 			} else {
    621 				/*
    622 				 * we are unsure about the way m was allocated.
    623 				 * copy into multiple MCLBYTES cluster mbufs.
    624 				 */
    625 				MCLGET(n, wait);
    626 				n->m_len = 0;
    627 				n->m_len = M_TRAILINGSPACE(n);
    628 				n->m_len = min(n->m_len, len);
    629 				n->m_len = min(n->m_len, m->m_len - off);
    630 				memcpy(mtod(n, void *), mtod(m, char *) + off,
    631 				    (unsigned)n->m_len);
    632 			}
    633 		} else
    634 			memcpy(mtod(n, void *), mtod(m, char *) + off,
    635 			    (unsigned)n->m_len);
    636 		if (len != M_COPYALL)
    637 			len -= n->m_len;
    638 		off += n->m_len;
    639 #ifdef DIAGNOSTIC
    640 		if (off > m->m_len)
    641 			panic("m_copym0 overrun");
    642 #endif
    643 		if (off == m->m_len) {
    644 			m = m->m_next;
    645 			off = 0;
    646 		}
    647 		np = &n->m_next;
    648 	}
    649 	if (top == 0)
    650 		MCFail++;
    651 	return (top);
    652 nospace:
    653 	m_freem(top);
    654 	MCFail++;
    655 	return (NULL);
    656 }
    657 
    658 /*
    659  * Copy an entire packet, including header (which must be present).
    660  * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
    661  */
    662 struct mbuf *
    663 m_copypacket(struct mbuf *m, int how)
    664 {
    665 	struct mbuf *top, *n, *o;
    666 
    667 	MGET(n, how, m->m_type);
    668 	top = n;
    669 	if (!n)
    670 		goto nospace;
    671 
    672 	MCLAIM(n, m->m_owner);
    673 	M_COPY_PKTHDR(n, m);
    674 	n->m_len = m->m_len;
    675 	if (m->m_flags & M_EXT) {
    676 		n->m_data = m->m_data;
    677 		MCLADDREFERENCE(m, n);
    678 	} else {
    679 		memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
    680 	}
    681 
    682 	m = m->m_next;
    683 	while (m) {
    684 		MGET(o, how, m->m_type);
    685 		if (!o)
    686 			goto nospace;
    687 
    688 		MCLAIM(o, m->m_owner);
    689 		n->m_next = o;
    690 		n = n->m_next;
    691 
    692 		n->m_len = m->m_len;
    693 		if (m->m_flags & M_EXT) {
    694 			n->m_data = m->m_data;
    695 			MCLADDREFERENCE(m, n);
    696 		} else {
    697 			memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
    698 		}
    699 
    700 		m = m->m_next;
    701 	}
    702 	return top;
    703 nospace:
    704 	m_freem(top);
    705 	MCFail++;
    706 	return NULL;
    707 }
    708 
    709 /*
    710  * Copy data from an mbuf chain starting "off" bytes from the beginning,
    711  * continuing for "len" bytes, into the indicated buffer.
    712  */
    713 void
    714 m_copydata(struct mbuf *m, int off, int len, void *vp)
    715 {
    716 	unsigned	count;
    717 	void *		cp = vp;
    718 
    719 	if (off < 0 || len < 0)
    720 		panic("m_copydata: off %d, len %d", off, len);
    721 	while (off > 0) {
    722 		if (m == NULL)
    723 			panic("m_copydata: m == NULL, off %d", off);
    724 		if (off < m->m_len)
    725 			break;
    726 		off -= m->m_len;
    727 		m = m->m_next;
    728 	}
    729 	while (len > 0) {
    730 		if (m == NULL)
    731 			panic("m_copydata: m == NULL, len %d", len);
    732 		count = min(m->m_len - off, len);
    733 		memcpy(cp, mtod(m, char *) + off, count);
    734 		len -= count;
    735 		cp = (char *)cp + count;
    736 		off = 0;
    737 		m = m->m_next;
    738 	}
    739 }
    740 
    741 /*
    742  * Concatenate mbuf chain n to m.
    743  * n might be copied into m (when n->m_len is small), therefore data portion of
    744  * n could be copied into an mbuf of different mbuf type.
    745  * Any m_pkthdr is not updated.
    746  */
    747 void
    748 m_cat(struct mbuf *m, struct mbuf *n)
    749 {
    750 
    751 	while (m->m_next)
    752 		m = m->m_next;
    753 	while (n) {
    754 		if (M_READONLY(m) || n->m_len > M_TRAILINGSPACE(m)) {
    755 			/* just join the two chains */
    756 			m->m_next = n;
    757 			return;
    758 		}
    759 		/* splat the data from one into the other */
    760 		memcpy(mtod(m, char *) + m->m_len, mtod(n, void *),
    761 		    (u_int)n->m_len);
    762 		m->m_len += n->m_len;
    763 		n = m_free(n);
    764 	}
    765 }
    766 
    767 void
    768 m_adj(struct mbuf *mp, int req_len)
    769 {
    770 	int len = req_len;
    771 	struct mbuf *m;
    772 	int count;
    773 
    774 	if ((m = mp) == NULL)
    775 		return;
    776 	if (len >= 0) {
    777 		/*
    778 		 * Trim from head.
    779 		 */
    780 		while (m != NULL && len > 0) {
    781 			if (m->m_len <= len) {
    782 				len -= m->m_len;
    783 				m->m_len = 0;
    784 				m = m->m_next;
    785 			} else {
    786 				m->m_len -= len;
    787 				m->m_data += len;
    788 				len = 0;
    789 			}
    790 		}
    791 		m = mp;
    792 		if (mp->m_flags & M_PKTHDR)
    793 			m->m_pkthdr.len -= (req_len - len);
    794 	} else {
    795 		/*
    796 		 * Trim from tail.  Scan the mbuf chain,
    797 		 * calculating its length and finding the last mbuf.
    798 		 * If the adjustment only affects this mbuf, then just
    799 		 * adjust and return.  Otherwise, rescan and truncate
    800 		 * after the remaining size.
    801 		 */
    802 		len = -len;
    803 		count = 0;
    804 		for (;;) {
    805 			count += m->m_len;
    806 			if (m->m_next == (struct mbuf *)0)
    807 				break;
    808 			m = m->m_next;
    809 		}
    810 		if (m->m_len >= len) {
    811 			m->m_len -= len;
    812 			if (mp->m_flags & M_PKTHDR)
    813 				mp->m_pkthdr.len -= len;
    814 			return;
    815 		}
    816 		count -= len;
    817 		if (count < 0)
    818 			count = 0;
    819 		/*
    820 		 * Correct length for chain is "count".
    821 		 * Find the mbuf with last data, adjust its length,
    822 		 * and toss data from remaining mbufs on chain.
    823 		 */
    824 		m = mp;
    825 		if (m->m_flags & M_PKTHDR)
    826 			m->m_pkthdr.len = count;
    827 		for (; m; m = m->m_next) {
    828 			if (m->m_len >= count) {
    829 				m->m_len = count;
    830 				break;
    831 			}
    832 			count -= m->m_len;
    833 		}
    834 		if (m)
    835 			while (m->m_next)
    836 				(m = m->m_next)->m_len = 0;
    837 	}
    838 }
    839 
    840 /*
    841  * Rearrange an mbuf chain so that len bytes are contiguous
    842  * and in the data area of an mbuf (so that mtod and dtom
    843  * will work for a structure of size len).  Returns the resulting
    844  * mbuf chain on success, frees it and returns null on failure.
    845  * If there is room, it will add up to max_protohdr-len extra bytes to the
    846  * contiguous region in an attempt to avoid being called next time.
    847  */
    848 int MPFail;
    849 
    850 struct mbuf *
    851 m_pullup(struct mbuf *n, int len)
    852 {
    853 	struct mbuf *m;
    854 	int count;
    855 	int space;
    856 
    857 	/*
    858 	 * If first mbuf has no cluster, and has room for len bytes
    859 	 * without shifting current data, pullup into it,
    860 	 * otherwise allocate a new mbuf to prepend to the chain.
    861 	 */
    862 	if ((n->m_flags & M_EXT) == 0 &&
    863 	    n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
    864 		if (n->m_len >= len)
    865 			return (n);
    866 		m = n;
    867 		n = n->m_next;
    868 		len -= m->m_len;
    869 	} else {
    870 		if (len > MHLEN)
    871 			goto bad;
    872 		MGET(m, M_DONTWAIT, n->m_type);
    873 		if (m == 0)
    874 			goto bad;
    875 		MCLAIM(m, n->m_owner);
    876 		m->m_len = 0;
    877 		if (n->m_flags & M_PKTHDR) {
    878 			M_MOVE_PKTHDR(m, n);
    879 		}
    880 	}
    881 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
    882 	do {
    883 		count = min(min(max(len, max_protohdr), space), n->m_len);
    884 		memcpy(mtod(m, char *) + m->m_len, mtod(n, void *),
    885 		  (unsigned)count);
    886 		len -= count;
    887 		m->m_len += count;
    888 		n->m_len -= count;
    889 		space -= count;
    890 		if (n->m_len)
    891 			n->m_data += count;
    892 		else
    893 			n = m_free(n);
    894 	} while (len > 0 && n);
    895 	if (len > 0) {
    896 		(void) m_free(m);
    897 		goto bad;
    898 	}
    899 	m->m_next = n;
    900 	return (m);
    901 bad:
    902 	m_freem(n);
    903 	MPFail++;
    904 	return (NULL);
    905 }
    906 
    907 /*
    908  * Like m_pullup(), except a new mbuf is always allocated, and we allow
    909  * the amount of empty space before the data in the new mbuf to be specified
    910  * (in the event that the caller expects to prepend later).
    911  */
    912 int MSFail;
    913 
    914 struct mbuf *
    915 m_copyup(struct mbuf *n, int len, int dstoff)
    916 {
    917 	struct mbuf *m;
    918 	int count, space;
    919 
    920 	if (len > (MHLEN - dstoff))
    921 		goto bad;
    922 	MGET(m, M_DONTWAIT, n->m_type);
    923 	if (m == NULL)
    924 		goto bad;
    925 	MCLAIM(m, n->m_owner);
    926 	m->m_len = 0;
    927 	if (n->m_flags & M_PKTHDR) {
    928 		M_MOVE_PKTHDR(m, n);
    929 	}
    930 	m->m_data += dstoff;
    931 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
    932 	do {
    933 		count = min(min(max(len, max_protohdr), space), n->m_len);
    934 		memcpy(mtod(m, char *) + m->m_len, mtod(n, void *),
    935 		    (unsigned)count);
    936 		len -= count;
    937 		m->m_len += count;
    938 		n->m_len -= count;
    939 		space -= count;
    940 		if (n->m_len)
    941 			n->m_data += count;
    942 		else
    943 			n = m_free(n);
    944 	} while (len > 0 && n);
    945 	if (len > 0) {
    946 		(void) m_free(m);
    947 		goto bad;
    948 	}
    949 	m->m_next = n;
    950 	return (m);
    951  bad:
    952 	m_freem(n);
    953 	MSFail++;
    954 	return (NULL);
    955 }
    956 
    957 /*
    958  * Partition an mbuf chain in two pieces, returning the tail --
    959  * all but the first len0 bytes.  In case of failure, it returns NULL and
    960  * attempts to restore the chain to its original state.
    961  */
    962 struct mbuf *
    963 m_split(struct mbuf *m0, int len0, int wait)
    964 {
    965 
    966 	return m_split0(m0, len0, wait, 1);
    967 }
    968 
    969 static struct mbuf *
    970 m_split0(struct mbuf *m0, int len0, int wait, int copyhdr)
    971 {
    972 	struct mbuf *m, *n;
    973 	unsigned len = len0, remain, len_save;
    974 
    975 	for (m = m0; m && len > m->m_len; m = m->m_next)
    976 		len -= m->m_len;
    977 	if (m == 0)
    978 		return (NULL);
    979 	remain = m->m_len - len;
    980 	if (copyhdr && (m0->m_flags & M_PKTHDR)) {
    981 		MGETHDR(n, wait, m0->m_type);
    982 		if (n == 0)
    983 			return (NULL);
    984 		MCLAIM(n, m0->m_owner);
    985 		n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
    986 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
    987 		len_save = m0->m_pkthdr.len;
    988 		m0->m_pkthdr.len = len0;
    989 		if (m->m_flags & M_EXT)
    990 			goto extpacket;
    991 		if (remain > MHLEN) {
    992 			/* m can't be the lead packet */
    993 			MH_ALIGN(n, 0);
    994 			n->m_next = m_split(m, len, wait);
    995 			if (n->m_next == 0) {
    996 				(void) m_free(n);
    997 				m0->m_pkthdr.len = len_save;
    998 				return (NULL);
    999 			} else
   1000 				return (n);
   1001 		} else
   1002 			MH_ALIGN(n, remain);
   1003 	} else if (remain == 0) {
   1004 		n = m->m_next;
   1005 		m->m_next = 0;
   1006 		return (n);
   1007 	} else {
   1008 		MGET(n, wait, m->m_type);
   1009 		if (n == 0)
   1010 			return (NULL);
   1011 		MCLAIM(n, m->m_owner);
   1012 		M_ALIGN(n, remain);
   1013 	}
   1014 extpacket:
   1015 	if (m->m_flags & M_EXT) {
   1016 		n->m_data = m->m_data + len;
   1017 		MCLADDREFERENCE(m, n);
   1018 	} else {
   1019 		memcpy(mtod(n, void *), mtod(m, char *) + len, remain);
   1020 	}
   1021 	n->m_len = remain;
   1022 	m->m_len = len;
   1023 	n->m_next = m->m_next;
   1024 	m->m_next = 0;
   1025 	return (n);
   1026 }
   1027 /*
   1028  * Routine to copy from device local memory into mbufs.
   1029  */
   1030 struct mbuf *
   1031 m_devget(char *buf, int totlen, int off0, struct ifnet *ifp,
   1032     void (*copy)(const void *from, void *to, size_t len))
   1033 {
   1034 	struct mbuf *m;
   1035 	struct mbuf *top = 0, **mp = &top;
   1036 	int off = off0, len;
   1037 	char *cp;
   1038 	char *epkt;
   1039 
   1040 	cp = buf;
   1041 	epkt = cp + totlen;
   1042 	if (off) {
   1043 		/*
   1044 		 * If 'off' is non-zero, packet is trailer-encapsulated,
   1045 		 * so we have to skip the type and length fields.
   1046 		 */
   1047 		cp += off + 2 * sizeof(uint16_t);
   1048 		totlen -= 2 * sizeof(uint16_t);
   1049 	}
   1050 	MGETHDR(m, M_DONTWAIT, MT_DATA);
   1051 	if (m == 0)
   1052 		return (NULL);
   1053 	m->m_pkthdr.rcvif = ifp;
   1054 	m->m_pkthdr.len = totlen;
   1055 	m->m_len = MHLEN;
   1056 
   1057 	while (totlen > 0) {
   1058 		if (top) {
   1059 			MGET(m, M_DONTWAIT, MT_DATA);
   1060 			if (m == 0) {
   1061 				m_freem(top);
   1062 				return (NULL);
   1063 			}
   1064 			m->m_len = MLEN;
   1065 		}
   1066 		len = min(totlen, epkt - cp);
   1067 		if (len >= MINCLSIZE) {
   1068 			MCLGET(m, M_DONTWAIT);
   1069 			if ((m->m_flags & M_EXT) == 0) {
   1070 				m_free(m);
   1071 				m_freem(top);
   1072 				return (NULL);
   1073 			}
   1074 			m->m_len = len = min(len, MCLBYTES);
   1075 		} else {
   1076 			/*
   1077 			 * Place initial small packet/header at end of mbuf.
   1078 			 */
   1079 			if (len < m->m_len) {
   1080 				if (top == 0 && len + max_linkhdr <= m->m_len)
   1081 					m->m_data += max_linkhdr;
   1082 				m->m_len = len;
   1083 			} else
   1084 				len = m->m_len;
   1085 		}
   1086 		if (copy)
   1087 			copy(cp, mtod(m, void *), (size_t)len);
   1088 		else
   1089 			memcpy(mtod(m, void *), cp, (size_t)len);
   1090 		cp += len;
   1091 		*mp = m;
   1092 		mp = &m->m_next;
   1093 		totlen -= len;
   1094 		if (cp == epkt)
   1095 			cp = buf;
   1096 	}
   1097 	return (top);
   1098 }
   1099 
   1100 /*
   1101  * Copy data from a buffer back into the indicated mbuf chain,
   1102  * starting "off" bytes from the beginning, extending the mbuf
   1103  * chain if necessary.
   1104  */
   1105 void
   1106 m_copyback(struct mbuf *m0, int off, int len, const void *cp)
   1107 {
   1108 #if defined(DEBUG)
   1109 	struct mbuf *origm = m0;
   1110 	int error;
   1111 #endif /* defined(DEBUG) */
   1112 
   1113 	if (m0 == NULL)
   1114 		return;
   1115 
   1116 #if defined(DEBUG)
   1117 	error =
   1118 #endif /* defined(DEBUG) */
   1119 	m_copyback0(&m0, off, len, cp,
   1120 	    M_COPYBACK0_COPYBACK|M_COPYBACK0_EXTEND, M_DONTWAIT);
   1121 
   1122 #if defined(DEBUG)
   1123 	if (error != 0 || (m0 != NULL && origm != m0))
   1124 		panic("m_copyback");
   1125 #endif /* defined(DEBUG) */
   1126 }
   1127 
   1128 struct mbuf *
   1129 m_copyback_cow(struct mbuf *m0, int off, int len, const void *cp, int how)
   1130 {
   1131 	int error;
   1132 
   1133 	/* don't support chain expansion */
   1134 	KDASSERT(off + len <= m_length(m0));
   1135 
   1136 	error = m_copyback0(&m0, off, len, cp,
   1137 	    M_COPYBACK0_COPYBACK|M_COPYBACK0_COW, how);
   1138 	if (error) {
   1139 		/*
   1140 		 * no way to recover from partial success.
   1141 		 * just free the chain.
   1142 		 */
   1143 		m_freem(m0);
   1144 		return NULL;
   1145 	}
   1146 	return m0;
   1147 }
   1148 
   1149 /*
   1150  * m_makewritable: ensure the specified range writable.
   1151  */
   1152 int
   1153 m_makewritable(struct mbuf **mp, int off, int len, int how)
   1154 {
   1155 	int error;
   1156 #if defined(DEBUG)
   1157 	struct mbuf *n;
   1158 	int origlen, reslen;
   1159 
   1160 	origlen = m_length(*mp);
   1161 #endif /* defined(DEBUG) */
   1162 
   1163 #if 0 /* M_COPYALL is large enough */
   1164 	if (len == M_COPYALL)
   1165 		len = m_length(*mp) - off; /* XXX */
   1166 #endif
   1167 
   1168 	error = m_copyback0(mp, off, len, NULL,
   1169 	    M_COPYBACK0_PRESERVE|M_COPYBACK0_COW, how);
   1170 
   1171 #if defined(DEBUG)
   1172 	reslen = 0;
   1173 	for (n = *mp; n; n = n->m_next)
   1174 		reslen += n->m_len;
   1175 	if (origlen != reslen)
   1176 		panic("m_makewritable: length changed");
   1177 	if (((*mp)->m_flags & M_PKTHDR) != 0 && reslen != (*mp)->m_pkthdr.len)
   1178 		panic("m_makewritable: inconsist");
   1179 #endif /* defined(DEBUG) */
   1180 
   1181 	return error;
   1182 }
   1183 
   1184 int
   1185 m_copyback0(struct mbuf **mp0, int off, int len, const void *vp, int flags,
   1186     int how)
   1187 {
   1188 	int mlen;
   1189 	struct mbuf *m, *n;
   1190 	struct mbuf **mp;
   1191 	int totlen = 0;
   1192 	const char *cp = vp;
   1193 
   1194 	KASSERT(mp0 != NULL);
   1195 	KASSERT(*mp0 != NULL);
   1196 	KASSERT((flags & M_COPYBACK0_PRESERVE) == 0 || cp == NULL);
   1197 	KASSERT((flags & M_COPYBACK0_COPYBACK) == 0 || cp != NULL);
   1198 
   1199 	/*
   1200 	 * we don't bother to update "totlen" in the case of M_COPYBACK0_COW,
   1201 	 * assuming that M_COPYBACK0_EXTEND and M_COPYBACK0_COW are exclusive.
   1202 	 */
   1203 
   1204 	KASSERT((~flags & (M_COPYBACK0_EXTEND|M_COPYBACK0_COW)) != 0);
   1205 
   1206 	mp = mp0;
   1207 	m = *mp;
   1208 	while (off > (mlen = m->m_len)) {
   1209 		off -= mlen;
   1210 		totlen += mlen;
   1211 		if (m->m_next == NULL) {
   1212 			int tspace;
   1213 extend:
   1214 			if ((flags & M_COPYBACK0_EXTEND) == 0)
   1215 				goto out;
   1216 
   1217 			/*
   1218 			 * try to make some space at the end of "m".
   1219 			 */
   1220 
   1221 			mlen = m->m_len;
   1222 			if (off + len >= MINCLSIZE &&
   1223 			    (m->m_flags & M_EXT) == 0 && m->m_len == 0) {
   1224 				MCLGET(m, how);
   1225 			}
   1226 			tspace = M_TRAILINGSPACE(m);
   1227 			if (tspace > 0) {
   1228 				tspace = min(tspace, off + len);
   1229 				KASSERT(tspace > 0);
   1230 				memset(mtod(m, char *) + m->m_len, 0,
   1231 				    min(off, tspace));
   1232 				m->m_len += tspace;
   1233 				off += mlen;
   1234 				totlen -= mlen;
   1235 				continue;
   1236 			}
   1237 
   1238 			/*
   1239 			 * need to allocate an mbuf.
   1240 			 */
   1241 
   1242 			if (off + len >= MINCLSIZE) {
   1243 				n = m_getcl(how, m->m_type, 0);
   1244 			} else {
   1245 				n = m_get(how, m->m_type);
   1246 			}
   1247 			if (n == NULL) {
   1248 				goto out;
   1249 			}
   1250 			n->m_len = 0;
   1251 			n->m_len = min(M_TRAILINGSPACE(n), off + len);
   1252 			memset(mtod(n, char *), 0, min(n->m_len, off));
   1253 			m->m_next = n;
   1254 		}
   1255 		mp = &m->m_next;
   1256 		m = m->m_next;
   1257 	}
   1258 	while (len > 0) {
   1259 		mlen = m->m_len - off;
   1260 		if (mlen != 0 && M_READONLY(m)) {
   1261 			char *datap;
   1262 			int eatlen;
   1263 
   1264 			/*
   1265 			 * this mbuf is read-only.
   1266 			 * allocate a new writable mbuf and try again.
   1267 			 */
   1268 
   1269 #if defined(DIAGNOSTIC)
   1270 			if ((flags & M_COPYBACK0_COW) == 0)
   1271 				panic("m_copyback0: read-only");
   1272 #endif /* defined(DIAGNOSTIC) */
   1273 
   1274 			/*
   1275 			 * if we're going to write into the middle of
   1276 			 * a mbuf, split it first.
   1277 			 */
   1278 			if (off > 0 && len < mlen) {
   1279 				n = m_split0(m, off, how, 0);
   1280 				if (n == NULL)
   1281 					goto enobufs;
   1282 				m->m_next = n;
   1283 				mp = &m->m_next;
   1284 				m = n;
   1285 				off = 0;
   1286 				continue;
   1287 			}
   1288 
   1289 			/*
   1290 			 * XXX TODO coalesce into the trailingspace of
   1291 			 * the previous mbuf when possible.
   1292 			 */
   1293 
   1294 			/*
   1295 			 * allocate a new mbuf.  copy packet header if needed.
   1296 			 */
   1297 			MGET(n, how, m->m_type);
   1298 			if (n == NULL)
   1299 				goto enobufs;
   1300 			MCLAIM(n, m->m_owner);
   1301 			if (off == 0 && (m->m_flags & M_PKTHDR) != 0) {
   1302 				M_MOVE_PKTHDR(n, m);
   1303 				n->m_len = MHLEN;
   1304 			} else {
   1305 				if (len >= MINCLSIZE)
   1306 					MCLGET(n, M_DONTWAIT);
   1307 				n->m_len =
   1308 				    (n->m_flags & M_EXT) ? MCLBYTES : MLEN;
   1309 			}
   1310 			if (n->m_len > len)
   1311 				n->m_len = len;
   1312 
   1313 			/*
   1314 			 * free the region which has been overwritten.
   1315 			 * copying data from old mbufs if requested.
   1316 			 */
   1317 			if (flags & M_COPYBACK0_PRESERVE)
   1318 				datap = mtod(n, char *);
   1319 			else
   1320 				datap = NULL;
   1321 			eatlen = n->m_len;
   1322 			KDASSERT(off == 0 || eatlen >= mlen);
   1323 			if (off > 0) {
   1324 				KDASSERT(len >= mlen);
   1325 				m->m_len = off;
   1326 				m->m_next = n;
   1327 				if (datap) {
   1328 					m_copydata(m, off, mlen, datap);
   1329 					datap += mlen;
   1330 				}
   1331 				eatlen -= mlen;
   1332 				mp = &m->m_next;
   1333 				m = m->m_next;
   1334 			}
   1335 			while (m != NULL && M_READONLY(m) &&
   1336 			    n->m_type == m->m_type && eatlen > 0) {
   1337 				mlen = min(eatlen, m->m_len);
   1338 				if (datap) {
   1339 					m_copydata(m, 0, mlen, datap);
   1340 					datap += mlen;
   1341 				}
   1342 				m->m_data += mlen;
   1343 				m->m_len -= mlen;
   1344 				eatlen -= mlen;
   1345 				if (m->m_len == 0)
   1346 					*mp = m = m_free(m);
   1347 			}
   1348 			if (eatlen > 0)
   1349 				n->m_len -= eatlen;
   1350 			n->m_next = m;
   1351 			*mp = m = n;
   1352 			continue;
   1353 		}
   1354 		mlen = min(mlen, len);
   1355 		if (flags & M_COPYBACK0_COPYBACK) {
   1356 			memcpy(mtod(m, char *) + off, cp, (unsigned)mlen);
   1357 			cp += mlen;
   1358 		}
   1359 		len -= mlen;
   1360 		mlen += off;
   1361 		off = 0;
   1362 		totlen += mlen;
   1363 		if (len == 0)
   1364 			break;
   1365 		if (m->m_next == NULL) {
   1366 			goto extend;
   1367 		}
   1368 		mp = &m->m_next;
   1369 		m = m->m_next;
   1370 	}
   1371 out:	if (((m = *mp0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) {
   1372 		KASSERT((flags & M_COPYBACK0_EXTEND) != 0);
   1373 		m->m_pkthdr.len = totlen;
   1374 	}
   1375 
   1376 	return 0;
   1377 
   1378 enobufs:
   1379 	return ENOBUFS;
   1380 }
   1381 
   1382 void
   1383 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
   1384 {
   1385 
   1386 	KASSERT((to->m_flags & M_EXT) == 0);
   1387 	KASSERT((to->m_flags & M_PKTHDR) == 0 || m_tag_first(to) == NULL);
   1388 	KASSERT((from->m_flags & M_PKTHDR) != 0);
   1389 
   1390 	to->m_pkthdr = from->m_pkthdr;
   1391 	to->m_flags = from->m_flags & M_COPYFLAGS;
   1392 	to->m_data = to->m_pktdat;
   1393 
   1394 	from->m_flags &= ~M_PKTHDR;
   1395 }
   1396 
   1397 /*
   1398  * Apply function f to the data in an mbuf chain starting "off" bytes from the
   1399  * beginning, continuing for "len" bytes.
   1400  */
   1401 int
   1402 m_apply(struct mbuf *m, int off, int len,
   1403     int (*f)(void *, void *, unsigned int), void *arg)
   1404 {
   1405 	unsigned int count;
   1406 	int rval;
   1407 
   1408 	KASSERT(len >= 0);
   1409 	KASSERT(off >= 0);
   1410 
   1411 	while (off > 0) {
   1412 		KASSERT(m != NULL);
   1413 		if (off < m->m_len)
   1414 			break;
   1415 		off -= m->m_len;
   1416 		m = m->m_next;
   1417 	}
   1418 	while (len > 0) {
   1419 		KASSERT(m != NULL);
   1420 		count = min(m->m_len - off, len);
   1421 
   1422 		rval = (*f)(arg, mtod(m, char *) + off, count);
   1423 		if (rval)
   1424 			return (rval);
   1425 
   1426 		len -= count;
   1427 		off = 0;
   1428 		m = m->m_next;
   1429 	}
   1430 
   1431 	return (0);
   1432 }
   1433 
   1434 /*
   1435  * Return a pointer to mbuf/offset of location in mbuf chain.
   1436  */
   1437 struct mbuf *
   1438 m_getptr(struct mbuf *m, int loc, int *off)
   1439 {
   1440 
   1441 	while (loc >= 0) {
   1442 		/* Normal end of search */
   1443 		if (m->m_len > loc) {
   1444 	    		*off = loc;
   1445 	    		return (m);
   1446 		} else {
   1447 	    		loc -= m->m_len;
   1448 
   1449 	    		if (m->m_next == NULL) {
   1450 				if (loc == 0) {
   1451  					/* Point at the end of valid data */
   1452 		    			*off = m->m_len;
   1453 		    			return (m);
   1454 				} else
   1455 		  			return (NULL);
   1456 	    		} else
   1457 	      			m = m->m_next;
   1458 		}
   1459     	}
   1460 
   1461 	return (NULL);
   1462 }
   1463 
   1464 /*
   1465  * m_ext_free: release a reference to the mbuf external storage.
   1466  *
   1467  * => free the mbuf m itsself as well.
   1468  * => called at splvm.
   1469  */
   1470 
   1471 void
   1472 m_ext_free(struct mbuf *m)
   1473 {
   1474 	bool embedded = MEXT_ISEMBEDDED(m);
   1475 	bool dofree = true;
   1476 
   1477 	KASSERT((m->m_flags & M_EXT) != 0);
   1478 	KASSERT(MEXT_ISEMBEDDED(m->m_ext_ref));
   1479 	KASSERT((m->m_ext_ref->m_flags & M_EXT) != 0);
   1480 	KASSERT((m->m_flags & M_EXT_CLUSTER) ==
   1481 	    (m->m_ext_ref->m_flags & M_EXT_CLUSTER));
   1482 
   1483 	if (!mcl_dec_and_test_reference(m)) {
   1484 		if (embedded) {
   1485 			dofree = false;
   1486 		} else {
   1487 			m->m_ext_ref = m;
   1488 		}
   1489 	} else {
   1490 		/*
   1491 		 * dropping the last reference
   1492 		 */
   1493 		if (!embedded) {
   1494 			mcl_inc_reference(m); /* XXX */
   1495 			m_ext_free(m->m_ext_ref);
   1496 			m->m_ext_ref = m;
   1497 		} else if ((m->m_flags & M_EXT_CLUSTER) != 0) {
   1498 			pool_cache_put_paddr((struct pool_cache *)
   1499 			    m->m_ext.ext_arg,
   1500 			    m->m_ext.ext_buf, m->m_ext.ext_paddr);
   1501 		} else if (m->m_ext.ext_free) {
   1502 			(*m->m_ext.ext_free)(m,
   1503 			    m->m_ext.ext_buf, m->m_ext.ext_size,
   1504 			    m->m_ext.ext_arg);
   1505 			/*
   1506 			 * 'm' is already freed by the ext_free callback.
   1507 			 */
   1508 			dofree = false;
   1509 		} else {
   1510 			free(m->m_ext.ext_buf, m->m_ext.ext_type);
   1511 		}
   1512 	}
   1513 	if (dofree) {
   1514 		pool_cache_put(&mbpool_cache, m);
   1515 	}
   1516 }
   1517 
   1518 #if defined(__HAVE_LAZY_MBUF) || defined(DEBUG)
   1519 char *
   1520 m_mapin(struct mbuf *m)
   1521 {
   1522 #if defined(__HAVE_LAZY_MBUF)
   1523 	int s;
   1524 
   1525 	KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES|M_EXT_LAZY)) == 0);
   1526 
   1527 	s = splvm();
   1528 	MEXT_LOCK(m);
   1529 	if (m->m_ext.ext_flags & M_EXT_LAZY) {
   1530 		vaddr_t buf = (vaddr_t)m->m_ext.ext_buf;
   1531 		vsize_t size = (vsize_t)m->m_ext.ext_size;
   1532 		vaddr_t va, sva, eva;
   1533 		int i;
   1534 
   1535 		sva = trunc_page(buf);
   1536 		eva = round_page(buf + size);
   1537 
   1538 		for (i = 0, va = sva; va < eva; i++, va += PAGE_SIZE) {
   1539 			pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
   1540 			    VM_PROT_READ);
   1541 		}
   1542 		pmap_update(pmap_kernel());
   1543 		m->m_ext.ext_flags &= ~M_EXT_LAZY;
   1544 	}
   1545 	MEXT_UNLOCK(m);
   1546 	splx(s);
   1547 
   1548 	m->m_flags &= ~M_EXT_LAZY;
   1549 	return m->m_data;
   1550 #else /* defined(__HAVE_LAZY_MBUF) */
   1551 	panic("m_mapin");
   1552 #endif /* defined(__HAVE_LAZY_MBUF) */
   1553 }
   1554 #endif /* defined(__HAVE_LAZY_MBUF) || defined(DEBUG) */
   1555 
   1556 #if defined(DDB)
   1557 void
   1558 m_print(const struct mbuf *m, const char *modif, void (*pr)(const char *, ...))
   1559 {
   1560 	char ch;
   1561 	bool opt_c = false;
   1562 	char buf[512];
   1563 
   1564 	while ((ch = *(modif++)) != '\0') {
   1565 		switch (ch) {
   1566 		case 'c':
   1567 			opt_c = true;
   1568 			break;
   1569 		}
   1570 	}
   1571 
   1572 nextchain:
   1573 	(*pr)("MBUF %p\n", m);
   1574 	bitmask_snprintf(m->m_flags, M_FLAGS_BITS, buf, sizeof(buf));
   1575 	(*pr)("  data=%p, len=%d, type=%d, flags=0x%s\n",
   1576 	    m->m_data, m->m_len, m->m_type, buf);
   1577 	(*pr)("  owner=%p, next=%p, nextpkt=%p\n", m->m_owner, m->m_next,
   1578 	    m->m_nextpkt);
   1579 	(*pr)("  leadingspace=%u, trailingspace=%u, readonly=%u\n",
   1580 	    (int)M_LEADINGSPACE(m), (int)M_TRAILINGSPACE(m),
   1581 	    (int)M_READONLY(m));
   1582 	if ((m->m_flags & M_PKTHDR) != 0) {
   1583 		bitmask_snprintf(m->m_pkthdr.csum_flags, M_CSUM_BITS, buf,
   1584 		    sizeof(buf));
   1585 		(*pr)("  pktlen=%d, rcvif=%p, csum_flags=0x%s, csum_data=0x%"
   1586 		    PRIx32 ", segsz=%u\n",
   1587 		    m->m_pkthdr.len, m->m_pkthdr.rcvif,
   1588 		    buf, m->m_pkthdr.csum_data, m->m_pkthdr.segsz);
   1589 	}
   1590 	if ((m->m_flags & M_EXT)) {
   1591 		(*pr)("  ext_refcnt=%u, ext_buf=%p, ext_size=%zd, "
   1592 		    "ext_free=%p, ext_arg=%p\n",
   1593 		    m->m_ext.ext_refcnt,
   1594 		    m->m_ext.ext_buf, m->m_ext.ext_size,
   1595 		    m->m_ext.ext_free, m->m_ext.ext_arg);
   1596 	}
   1597 	if ((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0) {
   1598 		vaddr_t sva = (vaddr_t)m->m_ext.ext_buf;
   1599 		vaddr_t eva = sva + m->m_ext.ext_size;
   1600 		int n = (round_page(eva) - trunc_page(sva)) >> PAGE_SHIFT;
   1601 		int i;
   1602 
   1603 		(*pr)("  pages:");
   1604 		for (i = 0; i < n; i ++) {
   1605 			(*pr)(" %p", m->m_ext.ext_pgs[i]);
   1606 		}
   1607 		(*pr)("\n");
   1608 	}
   1609 
   1610 	if (opt_c) {
   1611 		m = m->m_next;
   1612 		if (m != NULL) {
   1613 			goto nextchain;
   1614 		}
   1615 	}
   1616 }
   1617 #endif /* defined(DDB) */
   1618