xref: /netbsd-src/sys/kern/uipc_mbuf.c (revision 7d62b00eb9ad855ffcd7da46b41e23feb5476fac)
1 /*	$NetBSD: uipc_mbuf.c,v 1.248 2023/02/24 11:02:27 riastradh Exp $	*/
2 
3 /*
4  * Copyright (c) 1999, 2001, 2018 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, and Maxime Villard.
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  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1982, 1986, 1988, 1991, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. Neither the name of the University nor the names of its contributors
46  *    may be used to endorse or promote products derived from this software
47  *    without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  *
61  *	@(#)uipc_mbuf.c	8.4 (Berkeley) 2/14/95
62  */
63 
64 #include <sys/cdefs.h>
65 __KERNEL_RCSID(0, "$NetBSD: uipc_mbuf.c,v 1.248 2023/02/24 11:02:27 riastradh Exp $");
66 
67 #ifdef _KERNEL_OPT
68 #include "opt_mbuftrace.h"
69 #include "opt_nmbclusters.h"
70 #include "opt_ddb.h"
71 #include "ether.h"
72 #endif
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/atomic.h>
77 #include <sys/cpu.h>
78 #include <sys/proc.h>
79 #include <sys/mbuf.h>
80 #include <sys/kernel.h>
81 #include <sys/syslog.h>
82 #include <sys/domain.h>
83 #include <sys/protosw.h>
84 #include <sys/percpu.h>
85 #include <sys/pool.h>
86 #include <sys/socket.h>
87 #include <sys/sysctl.h>
88 
89 #include <net/if.h>
90 
91 pool_cache_t mb_cache;	/* mbuf cache */
92 static pool_cache_t mcl_cache;	/* mbuf cluster cache */
93 
94 struct mbstat mbstat;
95 int max_linkhdr;
96 int max_protohdr;
97 int max_hdr;
98 int max_datalen;
99 
100 static void mb_drain(void *, int);
101 static int mb_ctor(void *, void *, int);
102 
103 static void sysctl_kern_mbuf_setup(void);
104 
105 static struct sysctllog *mbuf_sysctllog;
106 
107 static struct mbuf *m_copy_internal(struct mbuf *, int, int, int, bool);
108 static struct mbuf *m_split_internal(struct mbuf *, int, int, bool);
109 static int m_copyback_internal(struct mbuf **, int, int, const void *,
110     int, int);
111 
112 /* Flags for m_copyback_internal. */
113 #define	CB_COPYBACK	0x0001	/* copyback from cp */
114 #define	CB_PRESERVE	0x0002	/* preserve original data */
115 #define	CB_COW		0x0004	/* do copy-on-write */
116 #define	CB_EXTEND	0x0008	/* extend chain */
117 
118 static const char mclpool_warnmsg[] =
119     "WARNING: mclpool limit reached; increase kern.mbuf.nmbclusters";
120 
121 MALLOC_DEFINE(M_MBUF, "mbuf", "mbuf");
122 
123 static percpu_t *mbstat_percpu;
124 
125 #ifdef MBUFTRACE
126 struct mownerhead mowners = LIST_HEAD_INITIALIZER(mowners);
127 struct mowner unknown_mowners[] = {
128 	MOWNER_INIT("unknown", "free"),
129 	MOWNER_INIT("unknown", "data"),
130 	MOWNER_INIT("unknown", "header"),
131 	MOWNER_INIT("unknown", "soname"),
132 	MOWNER_INIT("unknown", "soopts"),
133 	MOWNER_INIT("unknown", "ftable"),
134 	MOWNER_INIT("unknown", "control"),
135 	MOWNER_INIT("unknown", "oobdata"),
136 };
137 struct mowner revoked_mowner = MOWNER_INIT("revoked", "");
138 #endif
139 
140 #define	MEXT_ISEMBEDDED(m) ((m)->m_ext_ref == (m))
141 
142 #define	MCLADDREFERENCE(o, n)						\
143 do {									\
144 	KASSERT(((o)->m_flags & M_EXT) != 0);				\
145 	KASSERT(((n)->m_flags & M_EXT) == 0);				\
146 	KASSERT((o)->m_ext.ext_refcnt >= 1);				\
147 	(n)->m_flags |= ((o)->m_flags & M_EXTCOPYFLAGS);		\
148 	atomic_inc_uint(&(o)->m_ext.ext_refcnt);			\
149 	(n)->m_ext_ref = (o)->m_ext_ref;				\
150 	mowner_ref((n), (n)->m_flags);					\
151 } while (/* CONSTCOND */ 0)
152 
153 static int
154 nmbclusters_limit(void)
155 {
156 #if defined(PMAP_MAP_POOLPAGE)
157 	/* direct mapping, doesn't use space in kmem_arena */
158 	vsize_t max_size = physmem / 4;
159 #else
160 	vsize_t max_size = MIN(physmem / 4, nkmempages / 4);
161 #endif
162 
163 	max_size = max_size * PAGE_SIZE / MCLBYTES;
164 #ifdef NMBCLUSTERS_MAX
165 	max_size = MIN(max_size, NMBCLUSTERS_MAX);
166 #endif
167 
168 	return max_size;
169 }
170 
171 /*
172  * Initialize the mbuf allocator.
173  */
174 void
175 mbinit(void)
176 {
177 
178 	CTASSERT(sizeof(struct _m_ext) <= MHLEN);
179 	CTASSERT(sizeof(struct mbuf) == MSIZE);
180 
181 	sysctl_kern_mbuf_setup();
182 
183 	mb_cache = pool_cache_init(msize, 0, 0, 0, "mbpl",
184 	    NULL, IPL_VM, mb_ctor, NULL, NULL);
185 	KASSERT(mb_cache != NULL);
186 
187 	mcl_cache = pool_cache_init(mclbytes, COHERENCY_UNIT, 0, 0, "mclpl",
188 	    NULL, IPL_VM, NULL, NULL, NULL);
189 	KASSERT(mcl_cache != NULL);
190 
191 	pool_cache_set_drain_hook(mb_cache, mb_drain, NULL);
192 	pool_cache_set_drain_hook(mcl_cache, mb_drain, NULL);
193 
194 	/*
195 	 * Set an arbitrary default limit on the number of mbuf clusters.
196 	 */
197 #ifdef NMBCLUSTERS
198 	nmbclusters = MIN(NMBCLUSTERS, nmbclusters_limit());
199 #else
200 	nmbclusters = MAX(1024,
201 	    (vsize_t)physmem * PAGE_SIZE / MCLBYTES / 16);
202 	nmbclusters = MIN(nmbclusters, nmbclusters_limit());
203 #endif
204 
205 	/*
206 	 * Set the hard limit on the mclpool to the number of
207 	 * mbuf clusters the kernel is to support.  Log the limit
208 	 * reached message max once a minute.
209 	 */
210 	pool_cache_sethardlimit(mcl_cache, nmbclusters, mclpool_warnmsg, 60);
211 
212 	mbstat_percpu = percpu_alloc(sizeof(struct mbstat_cpu));
213 
214 	/*
215 	 * Set a low water mark for both mbufs and clusters.  This should
216 	 * help ensure that they can be allocated in a memory starvation
217 	 * situation.  This is important for e.g. diskless systems which
218 	 * must allocate mbufs in order for the pagedaemon to clean pages.
219 	 */
220 	pool_cache_setlowat(mb_cache, mblowat);
221 	pool_cache_setlowat(mcl_cache, mcllowat);
222 
223 #ifdef MBUFTRACE
224 	{
225 		/*
226 		 * Attach the unknown mowners.
227 		 */
228 		int i;
229 		MOWNER_ATTACH(&revoked_mowner);
230 		for (i = sizeof(unknown_mowners)/sizeof(unknown_mowners[0]);
231 		     i-- > 0; )
232 			MOWNER_ATTACH(&unknown_mowners[i]);
233 	}
234 #endif
235 }
236 
237 static void
238 mb_drain(void *arg, int flags)
239 {
240 	struct domain *dp;
241 	const struct protosw *pr;
242 	struct ifnet *ifp;
243 	int s;
244 
245 	KERNEL_LOCK(1, NULL);
246 	s = splvm();
247 	DOMAIN_FOREACH(dp) {
248 		for (pr = dp->dom_protosw;
249 		     pr < dp->dom_protoswNPROTOSW; pr++)
250 			if (pr->pr_drain)
251 				(*pr->pr_drain)();
252 	}
253 	/* XXX we cannot use psref in H/W interrupt */
254 	if (!cpu_intr_p()) {
255 		int bound = curlwp_bind();
256 		IFNET_READER_FOREACH(ifp) {
257 			struct psref psref;
258 
259 			if_acquire(ifp, &psref);
260 
261 			if (ifp->if_drain)
262 				(*ifp->if_drain)(ifp);
263 
264 			if_release(ifp, &psref);
265 		}
266 		curlwp_bindx(bound);
267 	}
268 	splx(s);
269 	mbstat.m_drain++;
270 	KERNEL_UNLOCK_ONE(NULL);
271 }
272 
273 /*
274  * sysctl helper routine for the kern.mbuf subtree.
275  * nmbclusters, mblowat and mcllowat need range
276  * checking and pool tweaking after being reset.
277  */
278 static int
279 sysctl_kern_mbuf(SYSCTLFN_ARGS)
280 {
281 	int error, newval;
282 	struct sysctlnode node;
283 
284 	node = *rnode;
285 	node.sysctl_data = &newval;
286 	switch (rnode->sysctl_num) {
287 	case MBUF_NMBCLUSTERS:
288 	case MBUF_MBLOWAT:
289 	case MBUF_MCLLOWAT:
290 		newval = *(int*)rnode->sysctl_data;
291 		break;
292 	case MBUF_NMBCLUSTERS_LIMIT:
293 		newval = nmbclusters_limit();
294 		break;
295 	default:
296 		return EOPNOTSUPP;
297 	}
298 
299 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
300 	if (error || newp == NULL)
301 		return error;
302 	if (newval < 0)
303 		return EINVAL;
304 
305 	switch (node.sysctl_num) {
306 	case MBUF_NMBCLUSTERS:
307 		if (newval < nmbclusters)
308 			return EINVAL;
309 		if (newval > nmbclusters_limit())
310 			return EINVAL;
311 		nmbclusters = newval;
312 		pool_cache_sethardlimit(mcl_cache, nmbclusters,
313 		    mclpool_warnmsg, 60);
314 		break;
315 	case MBUF_MBLOWAT:
316 		mblowat = newval;
317 		pool_cache_setlowat(mb_cache, mblowat);
318 		break;
319 	case MBUF_MCLLOWAT:
320 		mcllowat = newval;
321 		pool_cache_setlowat(mcl_cache, mcllowat);
322 		break;
323 	}
324 
325 	return 0;
326 }
327 
328 #ifdef MBUFTRACE
329 static void
330 mowner_convert_to_user_cb(void *v1, void *v2, struct cpu_info *ci)
331 {
332 	struct mowner_counter *mc = v1;
333 	struct mowner_user *mo_user = v2;
334 	int i;
335 
336 	for (i = 0; i < MOWNER_COUNTER_NCOUNTERS; i++) {
337 		mo_user->mo_counter[i] += mc->mc_counter[i];
338 	}
339 }
340 
341 static void
342 mowner_convert_to_user(struct mowner *mo, struct mowner_user *mo_user)
343 {
344 
345 	memset(mo_user, 0, sizeof(*mo_user));
346 	CTASSERT(sizeof(mo_user->mo_name) == sizeof(mo->mo_name));
347 	CTASSERT(sizeof(mo_user->mo_descr) == sizeof(mo->mo_descr));
348 	memcpy(mo_user->mo_name, mo->mo_name, sizeof(mo->mo_name));
349 	memcpy(mo_user->mo_descr, mo->mo_descr, sizeof(mo->mo_descr));
350 	percpu_foreach(mo->mo_counters, mowner_convert_to_user_cb, mo_user);
351 }
352 
353 static int
354 sysctl_kern_mbuf_mowners(SYSCTLFN_ARGS)
355 {
356 	struct mowner *mo;
357 	size_t len = 0;
358 	int error = 0;
359 
360 	if (namelen != 0)
361 		return EINVAL;
362 	if (newp != NULL)
363 		return EPERM;
364 
365 	LIST_FOREACH(mo, &mowners, mo_link) {
366 		struct mowner_user mo_user;
367 
368 		mowner_convert_to_user(mo, &mo_user);
369 
370 		if (oldp != NULL) {
371 			if (*oldlenp - len < sizeof(mo_user)) {
372 				error = ENOMEM;
373 				break;
374 			}
375 			error = copyout(&mo_user, (char *)oldp + len,
376 			    sizeof(mo_user));
377 			if (error)
378 				break;
379 		}
380 		len += sizeof(mo_user);
381 	}
382 
383 	if (error == 0)
384 		*oldlenp = len;
385 
386 	return error;
387 }
388 #endif /* MBUFTRACE */
389 
390 void
391 mbstat_type_add(int type, int diff)
392 {
393 	struct mbstat_cpu *mb;
394 	int s;
395 
396 	s = splvm();
397 	mb = percpu_getref(mbstat_percpu);
398 	mb->m_mtypes[type] += diff;
399 	percpu_putref(mbstat_percpu);
400 	splx(s);
401 }
402 
403 static void
404 mbstat_convert_to_user_cb(void *v1, void *v2, struct cpu_info *ci)
405 {
406 	struct mbstat_cpu *mbsc = v1;
407 	struct mbstat *mbs = v2;
408 	int i;
409 
410 	for (i = 0; i < __arraycount(mbs->m_mtypes); i++) {
411 		mbs->m_mtypes[i] += mbsc->m_mtypes[i];
412 	}
413 }
414 
415 static void
416 mbstat_convert_to_user(struct mbstat *mbs)
417 {
418 
419 	memset(mbs, 0, sizeof(*mbs));
420 	mbs->m_drain = mbstat.m_drain;
421 	percpu_foreach(mbstat_percpu, mbstat_convert_to_user_cb, mbs);
422 }
423 
424 static int
425 sysctl_kern_mbuf_stats(SYSCTLFN_ARGS)
426 {
427 	struct sysctlnode node;
428 	struct mbstat mbs;
429 
430 	mbstat_convert_to_user(&mbs);
431 	node = *rnode;
432 	node.sysctl_data = &mbs;
433 	node.sysctl_size = sizeof(mbs);
434 	return sysctl_lookup(SYSCTLFN_CALL(&node));
435 }
436 
437 static void
438 sysctl_kern_mbuf_setup(void)
439 {
440 
441 	KASSERT(mbuf_sysctllog == NULL);
442 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
443 		       CTLFLAG_PERMANENT,
444 		       CTLTYPE_NODE, "mbuf",
445 		       SYSCTL_DESCR("mbuf control variables"),
446 		       NULL, 0, NULL, 0,
447 		       CTL_KERN, KERN_MBUF, CTL_EOL);
448 
449 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
450 		       CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
451 		       CTLTYPE_INT, "msize",
452 		       SYSCTL_DESCR("mbuf base size"),
453 		       NULL, msize, NULL, 0,
454 		       CTL_KERN, KERN_MBUF, MBUF_MSIZE, CTL_EOL);
455 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
456 		       CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
457 		       CTLTYPE_INT, "mclbytes",
458 		       SYSCTL_DESCR("mbuf cluster size"),
459 		       NULL, mclbytes, NULL, 0,
460 		       CTL_KERN, KERN_MBUF, MBUF_MCLBYTES, CTL_EOL);
461 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
462 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
463 		       CTLTYPE_INT, "nmbclusters",
464 		       SYSCTL_DESCR("Limit on the number of mbuf clusters"),
465 		       sysctl_kern_mbuf, 0, &nmbclusters, 0,
466 		       CTL_KERN, KERN_MBUF, MBUF_NMBCLUSTERS, CTL_EOL);
467 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
468 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
469 		       CTLTYPE_INT, "mblowat",
470 		       SYSCTL_DESCR("mbuf low water mark"),
471 		       sysctl_kern_mbuf, 0, &mblowat, 0,
472 		       CTL_KERN, KERN_MBUF, MBUF_MBLOWAT, CTL_EOL);
473 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
474 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
475 		       CTLTYPE_INT, "mcllowat",
476 		       SYSCTL_DESCR("mbuf cluster low water mark"),
477 		       sysctl_kern_mbuf, 0, &mcllowat, 0,
478 		       CTL_KERN, KERN_MBUF, MBUF_MCLLOWAT, CTL_EOL);
479 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
480 		       CTLFLAG_PERMANENT,
481 		       CTLTYPE_STRUCT, "stats",
482 		       SYSCTL_DESCR("mbuf allocation statistics"),
483 		       sysctl_kern_mbuf_stats, 0, NULL, 0,
484 		       CTL_KERN, KERN_MBUF, MBUF_STATS, CTL_EOL);
485 #ifdef MBUFTRACE
486 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
487 		       CTLFLAG_PERMANENT,
488 		       CTLTYPE_STRUCT, "mowners",
489 		       SYSCTL_DESCR("Information about mbuf owners"),
490 		       sysctl_kern_mbuf_mowners, 0, NULL, 0,
491 		       CTL_KERN, KERN_MBUF, MBUF_MOWNERS, CTL_EOL);
492 #endif
493 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
494 		       CTLFLAG_PERMANENT|CTLFLAG_READONLY,
495 		       CTLTYPE_INT, "nmbclusters_limit",
496 		       SYSCTL_DESCR("Limit of nmbclusters"),
497 		       sysctl_kern_mbuf, 0, NULL, 0,
498 		       CTL_KERN, KERN_MBUF, MBUF_NMBCLUSTERS_LIMIT, CTL_EOL);
499 }
500 
501 static int
502 mb_ctor(void *arg, void *object, int flags)
503 {
504 	struct mbuf *m = object;
505 
506 #ifdef POOL_VTOPHYS
507 	m->m_paddr = POOL_VTOPHYS(m);
508 #else
509 	m->m_paddr = M_PADDR_INVALID;
510 #endif
511 	return 0;
512 }
513 
514 /*
515  * Add mbuf to the end of a chain
516  */
517 struct mbuf *
518 m_add(struct mbuf *c, struct mbuf *m)
519 {
520 	struct mbuf *n;
521 
522 	if (c == NULL)
523 		return m;
524 
525 	for (n = c; n->m_next != NULL; n = n->m_next)
526 		continue;
527 	n->m_next = m;
528 	return c;
529 }
530 
531 struct mbuf *
532 m_get(int how, int type)
533 {
534 	struct mbuf *m;
535 
536 	KASSERT(type != MT_FREE);
537 
538 	m = pool_cache_get(mb_cache,
539 	    how == M_WAIT ? PR_WAITOK|PR_LIMITFAIL : PR_NOWAIT);
540 	if (m == NULL)
541 		return NULL;
542 	KASSERT(((vaddr_t)m->m_dat & PAGE_MASK) + MLEN <= PAGE_SIZE);
543 
544 	mbstat_type_add(type, 1);
545 
546 	mowner_init(m, type);
547 	m->m_ext_ref = m; /* default */
548 	m->m_type = type;
549 	m->m_len = 0;
550 	m->m_next = NULL;
551 	m->m_nextpkt = NULL; /* default */
552 	m->m_data = m->m_dat;
553 	m->m_flags = 0; /* default */
554 
555 	return m;
556 }
557 
558 struct mbuf *
559 m_gethdr(int how, int type)
560 {
561 	struct mbuf *m;
562 
563 	m = m_get(how, type);
564 	if (m == NULL)
565 		return NULL;
566 
567 	m->m_data = m->m_pktdat;
568 	m->m_flags = M_PKTHDR;
569 
570 	m_reset_rcvif(m);
571 	m->m_pkthdr.len = 0;
572 	m->m_pkthdr.csum_flags = 0;
573 	m->m_pkthdr.csum_data = 0;
574 	m->m_pkthdr.segsz = 0;
575 	m->m_pkthdr.ether_vtag = 0;
576 	m->m_pkthdr.pkthdr_flags = 0;
577 	SLIST_INIT(&m->m_pkthdr.tags);
578 
579 	m->m_pkthdr.pattr_class = NULL;
580 	m->m_pkthdr.pattr_af = AF_UNSPEC;
581 	m->m_pkthdr.pattr_hdr = NULL;
582 
583 	return m;
584 }
585 
586 void
587 m_clget(struct mbuf *m, int how)
588 {
589 	m->m_ext_storage.ext_buf = (char *)pool_cache_get_paddr(mcl_cache,
590 	    how == M_WAIT ? (PR_WAITOK|PR_LIMITFAIL) : PR_NOWAIT,
591 	    &m->m_ext_storage.ext_paddr);
592 
593 	if (m->m_ext_storage.ext_buf == NULL)
594 		return;
595 
596 	KASSERT(((vaddr_t)m->m_ext_storage.ext_buf & PAGE_MASK) + mclbytes
597 	    <= PAGE_SIZE);
598 
599 	MCLINITREFERENCE(m);
600 	m->m_data = m->m_ext.ext_buf;
601 	m->m_flags = (m->m_flags & ~M_EXTCOPYFLAGS) |
602 	    M_EXT|M_EXT_CLUSTER|M_EXT_RW;
603 	m->m_ext.ext_size = MCLBYTES;
604 	m->m_ext.ext_free = NULL;
605 	m->m_ext.ext_arg = NULL;
606 	/* ext_paddr initialized above */
607 
608 	mowner_ref(m, M_EXT|M_EXT_CLUSTER);
609 }
610 
611 struct mbuf *
612 m_getcl(int how, int type, int flags)
613 {
614 	struct mbuf *mp;
615 
616 	if ((flags & M_PKTHDR) != 0)
617 		mp = m_gethdr(how, type);
618 	else
619 		mp = m_get(how, type);
620 
621 	if (mp == NULL)
622 		return NULL;
623 
624 	MCLGET(mp, how);
625 	if ((mp->m_flags & M_EXT) != 0)
626 		return mp;
627 
628 	m_free(mp);
629 	return NULL;
630 }
631 
632 /*
633  * Utility function for M_PREPEND. Do *NOT* use it directly.
634  */
635 struct mbuf *
636 m_prepend(struct mbuf *m, int len, int how)
637 {
638 	struct mbuf *mn;
639 
640 	if (__predict_false(len > MHLEN)) {
641 		panic("%s: len > MHLEN", __func__);
642 	}
643 
644 	KASSERT(len != M_COPYALL);
645 	mn = m_get(how, m->m_type);
646 	if (mn == NULL) {
647 		m_freem(m);
648 		return NULL;
649 	}
650 
651 	if (m->m_flags & M_PKTHDR) {
652 		m_move_pkthdr(mn, m);
653 	} else {
654 		MCLAIM(mn, m->m_owner);
655 	}
656 	mn->m_next = m;
657 	m = mn;
658 
659 	if (m->m_flags & M_PKTHDR) {
660 		if (len < MHLEN)
661 			m_align(m, len);
662 	} else {
663 		if (len < MLEN)
664 			m_align(m, len);
665 	}
666 
667 	m->m_len = len;
668 	return m;
669 }
670 
671 struct mbuf *
672 m_copym(struct mbuf *m, int off, int len, int wait)
673 {
674 	/* Shallow copy on M_EXT. */
675 	return m_copy_internal(m, off, len, wait, false);
676 }
677 
678 struct mbuf *
679 m_dup(struct mbuf *m, int off, int len, int wait)
680 {
681 	/* Deep copy. */
682 	return m_copy_internal(m, off, len, wait, true);
683 }
684 
685 static inline int
686 m_copylen(int len, int copylen)
687 {
688 	return (len == M_COPYALL) ? copylen : uimin(len, copylen);
689 }
690 
691 static struct mbuf *
692 m_copy_internal(struct mbuf *m, int off0, int len, int wait, bool deep)
693 {
694 	struct mbuf *n, **np;
695 	int off = off0;
696 	struct mbuf *top;
697 	int copyhdr = 0;
698 
699 	if (off < 0 || (len != M_COPYALL && len < 0))
700 		panic("%s: off %d, len %d", __func__, off, len);
701 	if (off == 0 && m->m_flags & M_PKTHDR)
702 		copyhdr = 1;
703 	while (off > 0) {
704 		if (m == NULL)
705 			panic("%s: m == NULL, off %d", __func__, off);
706 		if (off < m->m_len)
707 			break;
708 		off -= m->m_len;
709 		m = m->m_next;
710 	}
711 
712 	np = &top;
713 	top = NULL;
714 	while (len == M_COPYALL || len > 0) {
715 		if (m == NULL) {
716 			if (len != M_COPYALL)
717 				panic("%s: m == NULL, len %d [!COPYALL]",
718 				    __func__, len);
719 			break;
720 		}
721 
722 		n = m_get(wait, m->m_type);
723 		*np = n;
724 		if (n == NULL)
725 			goto nospace;
726 		MCLAIM(n, m->m_owner);
727 
728 		if (copyhdr) {
729 			m_copy_pkthdr(n, m);
730 			if (len == M_COPYALL)
731 				n->m_pkthdr.len -= off0;
732 			else
733 				n->m_pkthdr.len = len;
734 			copyhdr = 0;
735 		}
736 		n->m_len = m_copylen(len, m->m_len - off);
737 
738 		if (m->m_flags & M_EXT) {
739 			if (!deep) {
740 				n->m_data = m->m_data + off;
741 				MCLADDREFERENCE(m, n);
742 			} else {
743 				/*
744 				 * We don't care if MCLGET fails. n->m_len is
745 				 * recomputed and handles that.
746 				 */
747 				MCLGET(n, wait);
748 				n->m_len = 0;
749 				n->m_len = M_TRAILINGSPACE(n);
750 				n->m_len = m_copylen(len, n->m_len);
751 				n->m_len = uimin(n->m_len, m->m_len - off);
752 				memcpy(mtod(n, void *), mtod(m, char *) + off,
753 				    (unsigned)n->m_len);
754 			}
755 		} else {
756 			memcpy(mtod(n, void *), mtod(m, char *) + off,
757 			    (unsigned)n->m_len);
758 		}
759 
760 		if (len != M_COPYALL)
761 			len -= n->m_len;
762 		off += n->m_len;
763 
764 		KASSERT(off <= m->m_len);
765 
766 		if (off == m->m_len) {
767 			m = m->m_next;
768 			off = 0;
769 		}
770 		np = &n->m_next;
771 	}
772 
773 	return top;
774 
775 nospace:
776 	m_freem(top);
777 	return NULL;
778 }
779 
780 /*
781  * Copy an entire packet, including header (which must be present).
782  * An optimization of the common case 'm_copym(m, 0, M_COPYALL, how)'.
783  */
784 struct mbuf *
785 m_copypacket(struct mbuf *m, int how)
786 {
787 	struct mbuf *top, *n, *o;
788 
789 	if (__predict_false((m->m_flags & M_PKTHDR) == 0)) {
790 		panic("%s: no header (m = %p)", __func__, m);
791 	}
792 
793 	n = m_get(how, m->m_type);
794 	top = n;
795 	if (!n)
796 		goto nospace;
797 
798 	MCLAIM(n, m->m_owner);
799 	m_copy_pkthdr(n, m);
800 	n->m_len = m->m_len;
801 	if (m->m_flags & M_EXT) {
802 		n->m_data = m->m_data;
803 		MCLADDREFERENCE(m, n);
804 	} else {
805 		memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
806 	}
807 
808 	m = m->m_next;
809 	while (m) {
810 		o = m_get(how, m->m_type);
811 		if (!o)
812 			goto nospace;
813 
814 		MCLAIM(o, m->m_owner);
815 		n->m_next = o;
816 		n = n->m_next;
817 
818 		n->m_len = m->m_len;
819 		if (m->m_flags & M_EXT) {
820 			n->m_data = m->m_data;
821 			MCLADDREFERENCE(m, n);
822 		} else {
823 			memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
824 		}
825 
826 		m = m->m_next;
827 	}
828 	return top;
829 
830 nospace:
831 	m_freem(top);
832 	return NULL;
833 }
834 
835 void
836 m_copydata(struct mbuf *m, int off, int len, void *cp)
837 {
838 	unsigned int count;
839 	struct mbuf *m0 = m;
840 	int len0 = len;
841 	int off0 = off;
842 	void *cp0 = cp;
843 
844 	KASSERT(len != M_COPYALL);
845 	if (off < 0 || len < 0)
846 		panic("m_copydata: off %d, len %d", off, len);
847 	while (off > 0) {
848 		if (m == NULL)
849 			panic("m_copydata(%p,%d,%d,%p): m=NULL, off=%d (%d)",
850 			    m0, len0, off0, cp0, off, off0 - off);
851 		if (off < m->m_len)
852 			break;
853 		off -= m->m_len;
854 		m = m->m_next;
855 	}
856 	while (len > 0) {
857 		if (m == NULL)
858 			panic("m_copydata(%p,%d,%d,%p): "
859 			    "m=NULL, off=%d (%d), len=%d (%d)",
860 			    m0, len0, off0, cp0,
861 			    off, off0 - off, len, len0 - len);
862 		count = uimin(m->m_len - off, len);
863 		memcpy(cp, mtod(m, char *) + off, count);
864 		len -= count;
865 		cp = (char *)cp + count;
866 		off = 0;
867 		m = m->m_next;
868 	}
869 }
870 
871 /*
872  * Concatenate mbuf chain n to m.
873  * n might be copied into m (when n->m_len is small), therefore data portion of
874  * n could be copied into an mbuf of different mbuf type.
875  * Any m_pkthdr is not updated.
876  */
877 void
878 m_cat(struct mbuf *m, struct mbuf *n)
879 {
880 
881 	while (m->m_next)
882 		m = m->m_next;
883 	while (n) {
884 		if (M_READONLY(m) || n->m_len > M_TRAILINGSPACE(m)) {
885 			/* just join the two chains */
886 			m->m_next = n;
887 			return;
888 		}
889 		/* splat the data from one into the other */
890 		memcpy(mtod(m, char *) + m->m_len, mtod(n, void *),
891 		    (u_int)n->m_len);
892 		m->m_len += n->m_len;
893 		n = m_free(n);
894 	}
895 }
896 
897 void
898 m_adj(struct mbuf *mp, int req_len)
899 {
900 	int len = req_len;
901 	struct mbuf *m;
902 	int count;
903 
904 	if ((m = mp) == NULL)
905 		return;
906 	if (len >= 0) {
907 		/*
908 		 * Trim from head.
909 		 */
910 		while (m != NULL && len > 0) {
911 			if (m->m_len <= len) {
912 				len -= m->m_len;
913 				m->m_len = 0;
914 				m = m->m_next;
915 			} else {
916 				m->m_len -= len;
917 				m->m_data += len;
918 				len = 0;
919 			}
920 		}
921 		if (mp->m_flags & M_PKTHDR)
922 			mp->m_pkthdr.len -= (req_len - len);
923 	} else {
924 		/*
925 		 * Trim from tail.  Scan the mbuf chain,
926 		 * calculating its length and finding the last mbuf.
927 		 * If the adjustment only affects this mbuf, then just
928 		 * adjust and return.  Otherwise, rescan and truncate
929 		 * after the remaining size.
930 		 */
931 		len = -len;
932 		count = 0;
933 		for (;;) {
934 			count += m->m_len;
935 			if (m->m_next == NULL)
936 				break;
937 			m = m->m_next;
938 		}
939 		if (m->m_len >= len) {
940 			m->m_len -= len;
941 			if (mp->m_flags & M_PKTHDR)
942 				mp->m_pkthdr.len -= len;
943 			return;
944 		}
945 
946 		count -= len;
947 		if (count < 0)
948 			count = 0;
949 
950 		/*
951 		 * Correct length for chain is "count".
952 		 * Find the mbuf with last data, adjust its length,
953 		 * and toss data from remaining mbufs on chain.
954 		 */
955 		m = mp;
956 		if (m->m_flags & M_PKTHDR)
957 			m->m_pkthdr.len = count;
958 		for (; m; m = m->m_next) {
959 			if (m->m_len >= count) {
960 				m->m_len = count;
961 				break;
962 			}
963 			count -= m->m_len;
964 		}
965 		if (m) {
966 			while (m->m_next)
967 				(m = m->m_next)->m_len = 0;
968 		}
969 	}
970 }
971 
972 /*
973  * m_ensure_contig: rearrange an mbuf chain that given length of bytes
974  * would be contiguous and in the data area of an mbuf (therefore, mtod()
975  * would work for a structure of given length).
976  *
977  * => On success, returns true and the resulting mbuf chain; false otherwise.
978  * => The mbuf chain may change, but is always preserved valid.
979  */
980 bool
981 m_ensure_contig(struct mbuf **m0, int len)
982 {
983 	struct mbuf *n = *m0, *m;
984 	size_t count, space;
985 
986 	KASSERT(len != M_COPYALL);
987 	/*
988 	 * If first mbuf has no cluster, and has room for len bytes
989 	 * without shifting current data, pullup into it,
990 	 * otherwise allocate a new mbuf to prepend to the chain.
991 	 */
992 	if ((n->m_flags & M_EXT) == 0 &&
993 	    n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
994 		if (n->m_len >= len) {
995 			return true;
996 		}
997 		m = n;
998 		n = n->m_next;
999 		len -= m->m_len;
1000 	} else {
1001 		if (len > MHLEN) {
1002 			return false;
1003 		}
1004 		m = m_get(M_DONTWAIT, n->m_type);
1005 		if (m == NULL) {
1006 			return false;
1007 		}
1008 		MCLAIM(m, n->m_owner);
1009 		if (n->m_flags & M_PKTHDR) {
1010 			m_move_pkthdr(m, n);
1011 		}
1012 	}
1013 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1014 	do {
1015 		count = MIN(MIN(MAX(len, max_protohdr), space), n->m_len);
1016 		memcpy(mtod(m, char *) + m->m_len, mtod(n, void *),
1017 		  (unsigned)count);
1018 		len -= count;
1019 		m->m_len += count;
1020 		n->m_len -= count;
1021 		space -= count;
1022 		if (n->m_len)
1023 			n->m_data += count;
1024 		else
1025 			n = m_free(n);
1026 	} while (len > 0 && n);
1027 
1028 	m->m_next = n;
1029 	*m0 = m;
1030 
1031 	return len <= 0;
1032 }
1033 
1034 /*
1035  * m_pullup: same as m_ensure_contig(), but destroys mbuf chain on error.
1036  */
1037 struct mbuf *
1038 m_pullup(struct mbuf *n, int len)
1039 {
1040 	struct mbuf *m = n;
1041 
1042 	KASSERT(len != M_COPYALL);
1043 	if (!m_ensure_contig(&m, len)) {
1044 		KASSERT(m != NULL);
1045 		m_freem(m);
1046 		m = NULL;
1047 	}
1048 	return m;
1049 }
1050 
1051 /*
1052  * ensure that [off, off + len) is contiguous on the mbuf chain "m".
1053  * packet chain before "off" is kept untouched.
1054  * if offp == NULL, the target will start at <retval, 0> on resulting chain.
1055  * if offp != NULL, the target will start at <retval, *offp> on resulting chain.
1056  *
1057  * on error return (NULL return value), original "m" will be freed.
1058  *
1059  * XXX M_TRAILINGSPACE/M_LEADINGSPACE on shared cluster (sharedcluster)
1060  */
1061 struct mbuf *
1062 m_pulldown(struct mbuf *m, int off, int len, int *offp)
1063 {
1064 	struct mbuf *n, *o;
1065 	int hlen, tlen, olen;
1066 	int sharedcluster;
1067 
1068 	/* Check invalid arguments. */
1069 	if (m == NULL)
1070 		panic("%s: m == NULL", __func__);
1071 	if (len > MCLBYTES) {
1072 		m_freem(m);
1073 		return NULL;
1074 	}
1075 
1076 	n = m;
1077 	while (n != NULL && off > 0) {
1078 		if (n->m_len > off)
1079 			break;
1080 		off -= n->m_len;
1081 		n = n->m_next;
1082 	}
1083 	/* Be sure to point non-empty mbuf. */
1084 	while (n != NULL && n->m_len == 0)
1085 		n = n->m_next;
1086 	if (!n) {
1087 		m_freem(m);
1088 		return NULL;	/* mbuf chain too short */
1089 	}
1090 
1091 	sharedcluster = M_READONLY(n);
1092 
1093 	/*
1094 	 * The target data is on <n, off>. If we got enough data on the mbuf
1095 	 * "n", we're done.
1096 	 */
1097 #ifdef __NO_STRICT_ALIGNMENT
1098 	if ((off == 0 || offp) && len <= n->m_len - off && !sharedcluster)
1099 #else
1100 	if ((off == 0 || offp) && len <= n->m_len - off && !sharedcluster &&
1101 	    ALIGNED_POINTER((mtod(n, char *) + off), uint32_t))
1102 #endif
1103 		goto ok;
1104 
1105 	/*
1106 	 * When (len <= n->m_len - off) and (off != 0), it is a special case.
1107 	 * Len bytes from <n, off> sit in single mbuf, but the caller does
1108 	 * not like the starting position (off).
1109 	 *
1110 	 * Chop the current mbuf into two pieces, set off to 0.
1111 	 */
1112 	if (len <= n->m_len - off) {
1113 		struct mbuf *mlast;
1114 
1115 		o = m_dup(n, off, n->m_len - off, M_DONTWAIT);
1116 		if (o == NULL) {
1117 			m_freem(m);
1118 			return NULL;	/* ENOBUFS */
1119 		}
1120 		KASSERT(o->m_len >= len);
1121 		for (mlast = o; mlast->m_next != NULL; mlast = mlast->m_next)
1122 			;
1123 		n->m_len = off;
1124 		mlast->m_next = n->m_next;
1125 		n->m_next = o;
1126 		n = o;
1127 		off = 0;
1128 		goto ok;
1129 	}
1130 
1131 	/*
1132 	 * We need to take hlen from <n, off> and tlen from <n->m_next, 0>,
1133 	 * and construct contiguous mbuf with m_len == len.
1134 	 *
1135 	 * Note that hlen + tlen == len, and tlen > 0.
1136 	 */
1137 	hlen = n->m_len - off;
1138 	tlen = len - hlen;
1139 
1140 	/*
1141 	 * Ensure that we have enough trailing data on mbuf chain. If not,
1142 	 * we can do nothing about the chain.
1143 	 */
1144 	olen = 0;
1145 	for (o = n->m_next; o != NULL; o = o->m_next)
1146 		olen += o->m_len;
1147 	if (hlen + olen < len) {
1148 		m_freem(m);
1149 		return NULL;	/* mbuf chain too short */
1150 	}
1151 
1152 	/*
1153 	 * Easy cases first. We need to use m_copydata() to get data from
1154 	 * <n->m_next, 0>.
1155 	 */
1156 	if ((off == 0 || offp) && M_TRAILINGSPACE(n) >= tlen &&
1157 	    !sharedcluster) {
1158 		m_copydata(n->m_next, 0, tlen, mtod(n, char *) + n->m_len);
1159 		n->m_len += tlen;
1160 		m_adj(n->m_next, tlen);
1161 		goto ok;
1162 	}
1163 	if ((off == 0 || offp) && M_LEADINGSPACE(n->m_next) >= hlen &&
1164 #ifndef __NO_STRICT_ALIGNMENT
1165 	    ALIGNED_POINTER((n->m_next->m_data - hlen), uint32_t) &&
1166 #endif
1167 	    !sharedcluster && n->m_next->m_len >= tlen) {
1168 		n->m_next->m_data -= hlen;
1169 		n->m_next->m_len += hlen;
1170 		memcpy(mtod(n->m_next, void *), mtod(n, char *) + off, hlen);
1171 		n->m_len -= hlen;
1172 		n = n->m_next;
1173 		off = 0;
1174 		goto ok;
1175 	}
1176 
1177 	/*
1178 	 * Now, we need to do the hard way. Don't copy as there's no room
1179 	 * on both ends.
1180 	 */
1181 	o = m_get(M_DONTWAIT, m->m_type);
1182 	if (o && len > MLEN) {
1183 		MCLGET(o, M_DONTWAIT);
1184 		if ((o->m_flags & M_EXT) == 0) {
1185 			m_free(o);
1186 			o = NULL;
1187 		}
1188 	}
1189 	if (!o) {
1190 		m_freem(m);
1191 		return NULL;	/* ENOBUFS */
1192 	}
1193 	/* get hlen from <n, off> into <o, 0> */
1194 	o->m_len = hlen;
1195 	memcpy(mtod(o, void *), mtod(n, char *) + off, hlen);
1196 	n->m_len -= hlen;
1197 	/* get tlen from <n->m_next, 0> into <o, hlen> */
1198 	m_copydata(n->m_next, 0, tlen, mtod(o, char *) + o->m_len);
1199 	o->m_len += tlen;
1200 	m_adj(n->m_next, tlen);
1201 	o->m_next = n->m_next;
1202 	n->m_next = o;
1203 	n = o;
1204 	off = 0;
1205 
1206 ok:
1207 	if (offp)
1208 		*offp = off;
1209 	return n;
1210 }
1211 
1212 /*
1213  * Like m_pullup(), except a new mbuf is always allocated, and we allow
1214  * the amount of empty space before the data in the new mbuf to be specified
1215  * (in the event that the caller expects to prepend later).
1216  */
1217 struct mbuf *
1218 m_copyup(struct mbuf *n, int len, int dstoff)
1219 {
1220 	struct mbuf *m;
1221 	int count, space;
1222 
1223 	KASSERT(len != M_COPYALL);
1224 	if (len > ((int)MHLEN - dstoff))
1225 		goto bad;
1226 	m = m_get(M_DONTWAIT, n->m_type);
1227 	if (m == NULL)
1228 		goto bad;
1229 	MCLAIM(m, n->m_owner);
1230 	if (n->m_flags & M_PKTHDR) {
1231 		m_move_pkthdr(m, n);
1232 	}
1233 	m->m_data += dstoff;
1234 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1235 	do {
1236 		count = uimin(uimin(uimax(len, max_protohdr), space), n->m_len);
1237 		memcpy(mtod(m, char *) + m->m_len, mtod(n, void *),
1238 		    (unsigned)count);
1239 		len -= count;
1240 		m->m_len += count;
1241 		n->m_len -= count;
1242 		space -= count;
1243 		if (n->m_len)
1244 			n->m_data += count;
1245 		else
1246 			n = m_free(n);
1247 	} while (len > 0 && n);
1248 	if (len > 0) {
1249 		(void) m_free(m);
1250 		goto bad;
1251 	}
1252 	m->m_next = n;
1253 	return m;
1254  bad:
1255 	m_freem(n);
1256 	return NULL;
1257 }
1258 
1259 struct mbuf *
1260 m_split(struct mbuf *m0, int len, int wait)
1261 {
1262 	return m_split_internal(m0, len, wait, true);
1263 }
1264 
1265 static struct mbuf *
1266 m_split_internal(struct mbuf *m0, int len0, int wait, bool copyhdr)
1267 {
1268 	struct mbuf *m, *n;
1269 	unsigned len = len0, remain, len_save;
1270 
1271 	KASSERT(len0 != M_COPYALL);
1272 	for (m = m0; m && len > m->m_len; m = m->m_next)
1273 		len -= m->m_len;
1274 	if (m == NULL)
1275 		return NULL;
1276 
1277 	remain = m->m_len - len;
1278 	if (copyhdr && (m0->m_flags & M_PKTHDR)) {
1279 		n = m_gethdr(wait, m0->m_type);
1280 		if (n == NULL)
1281 			return NULL;
1282 
1283 		MCLAIM(n, m0->m_owner);
1284 		m_copy_rcvif(n, m0);
1285 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1286 		len_save = m0->m_pkthdr.len;
1287 		m0->m_pkthdr.len = len0;
1288 
1289 		if (m->m_flags & M_EXT)
1290 			goto extpacket;
1291 
1292 		if (remain > MHLEN) {
1293 			/* m can't be the lead packet */
1294 			m_align(n, 0);
1295 			n->m_len = 0;
1296 			n->m_next = m_split(m, len, wait);
1297 			if (n->m_next == NULL) {
1298 				(void)m_free(n);
1299 				m0->m_pkthdr.len = len_save;
1300 				return NULL;
1301 			}
1302 			return n;
1303 		} else {
1304 			m_align(n, remain);
1305 		}
1306 	} else if (remain == 0) {
1307 		n = m->m_next;
1308 		m->m_next = NULL;
1309 		return n;
1310 	} else {
1311 		n = m_get(wait, m->m_type);
1312 		if (n == NULL)
1313 			return NULL;
1314 		MCLAIM(n, m->m_owner);
1315 		m_align(n, remain);
1316 	}
1317 
1318 extpacket:
1319 	if (m->m_flags & M_EXT) {
1320 		n->m_data = m->m_data + len;
1321 		MCLADDREFERENCE(m, n);
1322 	} else {
1323 		memcpy(mtod(n, void *), mtod(m, char *) + len, remain);
1324 	}
1325 
1326 	n->m_len = remain;
1327 	m->m_len = len;
1328 	n->m_next = m->m_next;
1329 	m->m_next = NULL;
1330 	return n;
1331 }
1332 
1333 /*
1334  * Routine to copy from device local memory into mbufs.
1335  */
1336 struct mbuf *
1337 m_devget(char *buf, int totlen, int off, struct ifnet *ifp)
1338 {
1339 	struct mbuf *m;
1340 	struct mbuf *top = NULL, **mp = &top;
1341 	char *cp, *epkt;
1342 	int len;
1343 
1344 	cp = buf;
1345 	epkt = cp + totlen;
1346 	if (off) {
1347 		/*
1348 		 * If 'off' is non-zero, packet is trailer-encapsulated,
1349 		 * so we have to skip the type and length fields.
1350 		 */
1351 		cp += off + 2 * sizeof(uint16_t);
1352 		totlen -= 2 * sizeof(uint16_t);
1353 	}
1354 
1355 	m = m_gethdr(M_DONTWAIT, MT_DATA);
1356 	if (m == NULL)
1357 		return NULL;
1358 	m_set_rcvif(m, ifp);
1359 	m->m_pkthdr.len = totlen;
1360 	m->m_len = MHLEN;
1361 
1362 	while (totlen > 0) {
1363 		if (top) {
1364 			m = m_get(M_DONTWAIT, MT_DATA);
1365 			if (m == NULL) {
1366 				m_freem(top);
1367 				return NULL;
1368 			}
1369 			m->m_len = MLEN;
1370 		}
1371 
1372 		len = uimin(totlen, epkt - cp);
1373 
1374 		if (len >= MINCLSIZE) {
1375 			MCLGET(m, M_DONTWAIT);
1376 			if ((m->m_flags & M_EXT) == 0) {
1377 				m_free(m);
1378 				m_freem(top);
1379 				return NULL;
1380 			}
1381 			m->m_len = len = uimin(len, MCLBYTES);
1382 		} else {
1383 			/*
1384 			 * Place initial small packet/header at end of mbuf.
1385 			 */
1386 			if (len < m->m_len) {
1387 				if (top == 0 && len + max_linkhdr <= m->m_len)
1388 					m->m_data += max_linkhdr;
1389 				m->m_len = len;
1390 			} else
1391 				len = m->m_len;
1392 		}
1393 
1394 		memcpy(mtod(m, void *), cp, (size_t)len);
1395 
1396 		cp += len;
1397 		*mp = m;
1398 		mp = &m->m_next;
1399 		totlen -= len;
1400 		if (cp == epkt)
1401 			cp = buf;
1402 	}
1403 
1404 	return top;
1405 }
1406 
1407 /*
1408  * Copy data from a buffer back into the indicated mbuf chain,
1409  * starting "off" bytes from the beginning, extending the mbuf
1410  * chain if necessary.
1411  */
1412 void
1413 m_copyback(struct mbuf *m0, int off, int len, const void *cp)
1414 {
1415 #if defined(DEBUG)
1416 	struct mbuf *origm = m0;
1417 	int error;
1418 #endif
1419 
1420 	if (m0 == NULL)
1421 		return;
1422 
1423 #if defined(DEBUG)
1424 	error =
1425 #endif
1426 	m_copyback_internal(&m0, off, len, cp, CB_COPYBACK|CB_EXTEND,
1427 	    M_DONTWAIT);
1428 
1429 #if defined(DEBUG)
1430 	if (error != 0 || (m0 != NULL && origm != m0))
1431 		panic("m_copyback");
1432 #endif
1433 }
1434 
1435 struct mbuf *
1436 m_copyback_cow(struct mbuf *m0, int off, int len, const void *cp, int how)
1437 {
1438 	int error;
1439 
1440 	/* don't support chain expansion */
1441 	KASSERT(len != M_COPYALL);
1442 	KDASSERT(off + len <= m_length(m0));
1443 
1444 	error = m_copyback_internal(&m0, off, len, cp, CB_COPYBACK|CB_COW,
1445 	    how);
1446 	if (error) {
1447 		/*
1448 		 * no way to recover from partial success.
1449 		 * just free the chain.
1450 		 */
1451 		m_freem(m0);
1452 		return NULL;
1453 	}
1454 	return m0;
1455 }
1456 
1457 int
1458 m_makewritable(struct mbuf **mp, int off, int len, int how)
1459 {
1460 	int error;
1461 #if defined(DEBUG)
1462 	int origlen = m_length(*mp);
1463 #endif
1464 
1465 	error = m_copyback_internal(mp, off, len, NULL, CB_PRESERVE|CB_COW,
1466 	    how);
1467 	if (error)
1468 		return error;
1469 
1470 #if defined(DEBUG)
1471 	int reslen = 0;
1472 	for (struct mbuf *n = *mp; n; n = n->m_next)
1473 		reslen += n->m_len;
1474 	if (origlen != reslen)
1475 		panic("m_makewritable: length changed");
1476 	if (((*mp)->m_flags & M_PKTHDR) != 0 && reslen != (*mp)->m_pkthdr.len)
1477 		panic("m_makewritable: inconsist");
1478 #endif
1479 
1480 	return 0;
1481 }
1482 
1483 static int
1484 m_copyback_internal(struct mbuf **mp0, int off, int len, const void *vp,
1485     int flags, int how)
1486 {
1487 	int mlen;
1488 	struct mbuf *m, *n;
1489 	struct mbuf **mp;
1490 	int totlen = 0;
1491 	const char *cp = vp;
1492 
1493 	KASSERT(mp0 != NULL);
1494 	KASSERT(*mp0 != NULL);
1495 	KASSERT((flags & CB_PRESERVE) == 0 || cp == NULL);
1496 	KASSERT((flags & CB_COPYBACK) == 0 || cp != NULL);
1497 
1498 	if (len == M_COPYALL)
1499 		len = m_length(*mp0) - off;
1500 
1501 	/*
1502 	 * we don't bother to update "totlen" in the case of CB_COW,
1503 	 * assuming that CB_EXTEND and CB_COW are exclusive.
1504 	 */
1505 
1506 	KASSERT((~flags & (CB_EXTEND|CB_COW)) != 0);
1507 
1508 	mp = mp0;
1509 	m = *mp;
1510 	while (off > (mlen = m->m_len)) {
1511 		off -= mlen;
1512 		totlen += mlen;
1513 		if (m->m_next == NULL) {
1514 			int tspace;
1515 extend:
1516 			if ((flags & CB_EXTEND) == 0)
1517 				goto out;
1518 
1519 			/*
1520 			 * try to make some space at the end of "m".
1521 			 */
1522 
1523 			mlen = m->m_len;
1524 			if (off + len >= MINCLSIZE &&
1525 			    (m->m_flags & M_EXT) == 0 && m->m_len == 0) {
1526 				MCLGET(m, how);
1527 			}
1528 			tspace = M_TRAILINGSPACE(m);
1529 			if (tspace > 0) {
1530 				tspace = uimin(tspace, off + len);
1531 				KASSERT(tspace > 0);
1532 				memset(mtod(m, char *) + m->m_len, 0,
1533 				    uimin(off, tspace));
1534 				m->m_len += tspace;
1535 				off += mlen;
1536 				totlen -= mlen;
1537 				continue;
1538 			}
1539 
1540 			/*
1541 			 * need to allocate an mbuf.
1542 			 */
1543 
1544 			if (off + len >= MINCLSIZE) {
1545 				n = m_getcl(how, m->m_type, 0);
1546 			} else {
1547 				n = m_get(how, m->m_type);
1548 			}
1549 			if (n == NULL) {
1550 				goto out;
1551 			}
1552 			n->m_len = uimin(M_TRAILINGSPACE(n), off + len);
1553 			memset(mtod(n, char *), 0, uimin(n->m_len, off));
1554 			m->m_next = n;
1555 		}
1556 		mp = &m->m_next;
1557 		m = m->m_next;
1558 	}
1559 	while (len > 0) {
1560 		mlen = m->m_len - off;
1561 		if (mlen != 0 && M_READONLY(m)) {
1562 			/*
1563 			 * This mbuf is read-only. Allocate a new writable
1564 			 * mbuf and try again.
1565 			 */
1566 			char *datap;
1567 			int eatlen;
1568 
1569 			KASSERT((flags & CB_COW) != 0);
1570 
1571 			/*
1572 			 * if we're going to write into the middle of
1573 			 * a mbuf, split it first.
1574 			 */
1575 			if (off > 0) {
1576 				n = m_split_internal(m, off, how, false);
1577 				if (n == NULL)
1578 					goto enobufs;
1579 				m->m_next = n;
1580 				mp = &m->m_next;
1581 				m = n;
1582 				off = 0;
1583 				continue;
1584 			}
1585 
1586 			/*
1587 			 * XXX TODO coalesce into the trailingspace of
1588 			 * the previous mbuf when possible.
1589 			 */
1590 
1591 			/*
1592 			 * allocate a new mbuf.  copy packet header if needed.
1593 			 */
1594 			n = m_get(how, m->m_type);
1595 			if (n == NULL)
1596 				goto enobufs;
1597 			MCLAIM(n, m->m_owner);
1598 			if (off == 0 && (m->m_flags & M_PKTHDR) != 0) {
1599 				m_move_pkthdr(n, m);
1600 				n->m_len = MHLEN;
1601 			} else {
1602 				if (len >= MINCLSIZE)
1603 					MCLGET(n, M_DONTWAIT);
1604 				n->m_len =
1605 				    (n->m_flags & M_EXT) ? MCLBYTES : MLEN;
1606 			}
1607 			if (n->m_len > len)
1608 				n->m_len = len;
1609 
1610 			/*
1611 			 * free the region which has been overwritten.
1612 			 * copying data from old mbufs if requested.
1613 			 */
1614 			if (flags & CB_PRESERVE)
1615 				datap = mtod(n, char *);
1616 			else
1617 				datap = NULL;
1618 			eatlen = n->m_len;
1619 			while (m != NULL && M_READONLY(m) &&
1620 			    n->m_type == m->m_type && eatlen > 0) {
1621 				mlen = uimin(eatlen, m->m_len);
1622 				if (datap) {
1623 					m_copydata(m, 0, mlen, datap);
1624 					datap += mlen;
1625 				}
1626 				m->m_data += mlen;
1627 				m->m_len -= mlen;
1628 				eatlen -= mlen;
1629 				if (m->m_len == 0)
1630 					*mp = m = m_free(m);
1631 			}
1632 			if (eatlen > 0)
1633 				n->m_len -= eatlen;
1634 			n->m_next = m;
1635 			*mp = m = n;
1636 			continue;
1637 		}
1638 		mlen = uimin(mlen, len);
1639 		if (flags & CB_COPYBACK) {
1640 			memcpy(mtod(m, char *) + off, cp, (unsigned)mlen);
1641 			cp += mlen;
1642 		}
1643 		len -= mlen;
1644 		mlen += off;
1645 		off = 0;
1646 		totlen += mlen;
1647 		if (len == 0)
1648 			break;
1649 		if (m->m_next == NULL) {
1650 			goto extend;
1651 		}
1652 		mp = &m->m_next;
1653 		m = m->m_next;
1654 	}
1655 
1656 out:
1657 	if (((m = *mp0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) {
1658 		KASSERT((flags & CB_EXTEND) != 0);
1659 		m->m_pkthdr.len = totlen;
1660 	}
1661 
1662 	return 0;
1663 
1664 enobufs:
1665 	return ENOBUFS;
1666 }
1667 
1668 /*
1669  * Compress the mbuf chain. Return the new mbuf chain on success, NULL on
1670  * failure. The first mbuf is preserved, and on success the pointer returned
1671  * is the same as the one passed.
1672  */
1673 struct mbuf *
1674 m_defrag(struct mbuf *m, int how)
1675 {
1676 	struct mbuf *m0, *mn, *n;
1677 	int sz;
1678 
1679 	KASSERT((m->m_flags & M_PKTHDR) != 0);
1680 
1681 	if (m->m_next == NULL)
1682 		return m;
1683 
1684 	/* Defrag to single mbuf if at all possible */
1685 	if ((m->m_flags & M_EXT) == 0 && m->m_pkthdr.len <= MCLBYTES) {
1686 		if (m->m_pkthdr.len <= MHLEN) {
1687 			if (M_TRAILINGSPACE(m) < (m->m_pkthdr.len - m->m_len)) {
1688 				KASSERTMSG(M_LEADINGSPACE(m) +
1689 				    M_TRAILINGSPACE(m) >=
1690 				    (m->m_pkthdr.len - m->m_len),
1691 				    "too small leading %d trailing %d ro? %d"
1692 				    " pkthdr.len %d mlen %d",
1693 				    (int)M_LEADINGSPACE(m),
1694 				    (int)M_TRAILINGSPACE(m),
1695 				    M_READONLY(m),
1696 				    m->m_pkthdr.len, m->m_len);
1697 
1698 				memmove(m->m_pktdat, m->m_data, m->m_len);
1699 				m->m_data = m->m_pktdat;
1700 
1701 				KASSERT(M_TRAILINGSPACE(m) >=
1702 				    (m->m_pkthdr.len - m->m_len));
1703 			}
1704 		} else {
1705 			/* Must copy data before adding cluster */
1706 			m0 = m_get(how, MT_DATA);
1707 			if (m0 == NULL)
1708 				return NULL;
1709 			KASSERT(m->m_len <= MHLEN);
1710 			m_copydata(m, 0, m->m_len, mtod(m0, void *));
1711 
1712 			MCLGET(m, how);
1713 			if ((m->m_flags & M_EXT) == 0) {
1714 				m_free(m0);
1715 				return NULL;
1716 			}
1717 			memcpy(m->m_data, mtod(m0, void *), m->m_len);
1718 			m_free(m0);
1719 		}
1720 		KASSERT(M_TRAILINGSPACE(m) >= (m->m_pkthdr.len - m->m_len));
1721 		m_copydata(m->m_next, 0, m->m_pkthdr.len - m->m_len,
1722 			    mtod(m, char *) + m->m_len);
1723 		m->m_len = m->m_pkthdr.len;
1724 		m_freem(m->m_next);
1725 		m->m_next = NULL;
1726 		return m;
1727 	}
1728 
1729 	m0 = m_get(how, MT_DATA);
1730 	if (m0 == NULL)
1731 		return NULL;
1732 	mn = m0;
1733 
1734 	sz = m->m_pkthdr.len - m->m_len;
1735 	KASSERT(sz >= 0);
1736 
1737 	do {
1738 		if (sz > MLEN) {
1739 			MCLGET(mn, how);
1740 			if ((mn->m_flags & M_EXT) == 0) {
1741 				m_freem(m0);
1742 				return NULL;
1743 			}
1744 		}
1745 
1746 		mn->m_len = MIN(sz, MCLBYTES);
1747 
1748 		m_copydata(m, m->m_pkthdr.len - sz, mn->m_len,
1749 		     mtod(mn, void *));
1750 
1751 		sz -= mn->m_len;
1752 
1753 		if (sz > 0) {
1754 			/* need more mbufs */
1755 			n = m_get(how, MT_DATA);
1756 			if (n == NULL) {
1757 				m_freem(m0);
1758 				return NULL;
1759 			}
1760 
1761 			mn->m_next = n;
1762 			mn = n;
1763 		}
1764 	} while (sz > 0);
1765 
1766 	m_freem(m->m_next);
1767 	m->m_next = m0;
1768 
1769 	return m;
1770 }
1771 
1772 void
1773 m_remove_pkthdr(struct mbuf *m)
1774 {
1775 	KASSERT(m->m_flags & M_PKTHDR);
1776 
1777 	m_tag_delete_chain(m);
1778 	m->m_flags &= ~M_PKTHDR;
1779 	memset(&m->m_pkthdr, 0, sizeof(m->m_pkthdr));
1780 }
1781 
1782 void
1783 m_copy_pkthdr(struct mbuf *to, struct mbuf *from)
1784 {
1785 	KASSERT((to->m_flags & M_EXT) == 0);
1786 	KASSERT((to->m_flags & M_PKTHDR) == 0 ||
1787 	    SLIST_FIRST(&to->m_pkthdr.tags) == NULL);
1788 	KASSERT((from->m_flags & M_PKTHDR) != 0);
1789 
1790 	to->m_pkthdr = from->m_pkthdr;
1791 	to->m_flags = from->m_flags & M_COPYFLAGS;
1792 	to->m_data = to->m_pktdat;
1793 
1794 	SLIST_INIT(&to->m_pkthdr.tags);
1795 	m_tag_copy_chain(to, from);
1796 }
1797 
1798 void
1799 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
1800 {
1801 	KASSERT((to->m_flags & M_EXT) == 0);
1802 	KASSERT((to->m_flags & M_PKTHDR) == 0 ||
1803 	    SLIST_FIRST(&to->m_pkthdr.tags) == NULL);
1804 	KASSERT((from->m_flags & M_PKTHDR) != 0);
1805 
1806 	to->m_pkthdr = from->m_pkthdr;
1807 	to->m_flags = from->m_flags & M_COPYFLAGS;
1808 	to->m_data = to->m_pktdat;
1809 
1810 	from->m_flags &= ~M_PKTHDR;
1811 }
1812 
1813 /*
1814  * Set the m_data pointer of a newly-allocated mbuf to place an object of the
1815  * specified size at the end of the mbuf, longword aligned.
1816  */
1817 void
1818 m_align(struct mbuf *m, int len)
1819 {
1820 	int buflen, adjust;
1821 
1822 	KASSERT(len != M_COPYALL);
1823 	KASSERT(M_LEADINGSPACE(m) == 0);
1824 
1825 	buflen = M_BUFSIZE(m);
1826 
1827 	KASSERT(len <= buflen);
1828 	adjust = buflen - len;
1829 	m->m_data += adjust &~ (sizeof(long)-1);
1830 }
1831 
1832 /*
1833  * Apply function f to the data in an mbuf chain starting "off" bytes from the
1834  * beginning, continuing for "len" bytes.
1835  */
1836 int
1837 m_apply(struct mbuf *m, int off, int len,
1838     int (*f)(void *, void *, unsigned int), void *arg)
1839 {
1840 	unsigned int count;
1841 	int rval;
1842 
1843 	KASSERT(len != M_COPYALL);
1844 	KASSERT(len >= 0);
1845 	KASSERT(off >= 0);
1846 
1847 	while (off > 0) {
1848 		KASSERT(m != NULL);
1849 		if (off < m->m_len)
1850 			break;
1851 		off -= m->m_len;
1852 		m = m->m_next;
1853 	}
1854 	while (len > 0) {
1855 		KASSERT(m != NULL);
1856 		count = uimin(m->m_len - off, len);
1857 
1858 		rval = (*f)(arg, mtod(m, char *) + off, count);
1859 		if (rval)
1860 			return rval;
1861 
1862 		len -= count;
1863 		off = 0;
1864 		m = m->m_next;
1865 	}
1866 
1867 	return 0;
1868 }
1869 
1870 /*
1871  * Return a pointer to mbuf/offset of location in mbuf chain.
1872  */
1873 struct mbuf *
1874 m_getptr(struct mbuf *m, int loc, int *off)
1875 {
1876 
1877 	while (loc >= 0) {
1878 		/* Normal end of search */
1879 		if (m->m_len > loc) {
1880 			*off = loc;
1881 			return m;
1882 		}
1883 
1884 		loc -= m->m_len;
1885 
1886 		if (m->m_next == NULL) {
1887 			if (loc == 0) {
1888 				/* Point at the end of valid data */
1889 				*off = m->m_len;
1890 				return m;
1891 			}
1892 			return NULL;
1893 		} else {
1894 			m = m->m_next;
1895 		}
1896 	}
1897 
1898 	return NULL;
1899 }
1900 
1901 /*
1902  * Release a reference to the mbuf external storage.
1903  *
1904  * => free the mbuf m itself as well.
1905  */
1906 static void
1907 m_ext_free(struct mbuf *m)
1908 {
1909 	const bool embedded = MEXT_ISEMBEDDED(m);
1910 	bool dofree = true;
1911 	u_int refcnt;
1912 
1913 	KASSERT((m->m_flags & M_EXT) != 0);
1914 	KASSERT(MEXT_ISEMBEDDED(m->m_ext_ref));
1915 	KASSERT((m->m_ext_ref->m_flags & M_EXT) != 0);
1916 	KASSERT((m->m_flags & M_EXT_CLUSTER) ==
1917 	    (m->m_ext_ref->m_flags & M_EXT_CLUSTER));
1918 
1919 	if (__predict_false(m->m_type == MT_FREE)) {
1920 		panic("mbuf %p already freed", m);
1921 	}
1922 
1923 	if (__predict_true(m->m_ext.ext_refcnt == 1)) {
1924 		refcnt = m->m_ext.ext_refcnt = 0;
1925 	} else {
1926 		membar_release();
1927 		refcnt = atomic_dec_uint_nv(&m->m_ext.ext_refcnt);
1928 	}
1929 
1930 	if (refcnt > 0) {
1931 		if (embedded) {
1932 			/*
1933 			 * other mbuf's m_ext_ref still points to us.
1934 			 */
1935 			dofree = false;
1936 		} else {
1937 			m->m_ext_ref = m;
1938 		}
1939 	} else {
1940 		/*
1941 		 * dropping the last reference
1942 		 */
1943 		membar_acquire();
1944 		if (!embedded) {
1945 			m->m_ext.ext_refcnt++; /* XXX */
1946 			m_ext_free(m->m_ext_ref);
1947 			m->m_ext_ref = m;
1948 		} else if ((m->m_flags & M_EXT_CLUSTER) != 0) {
1949 			pool_cache_put_paddr(mcl_cache,
1950 			    m->m_ext.ext_buf, m->m_ext.ext_paddr);
1951 		} else if (m->m_ext.ext_free) {
1952 			(*m->m_ext.ext_free)(m,
1953 			    m->m_ext.ext_buf, m->m_ext.ext_size,
1954 			    m->m_ext.ext_arg);
1955 			/*
1956 			 * 'm' is already freed by the ext_free callback.
1957 			 */
1958 			dofree = false;
1959 		} else {
1960 			free(m->m_ext.ext_buf, 0);
1961 		}
1962 	}
1963 
1964 	if (dofree) {
1965 		m->m_type = MT_FREE;
1966 		m->m_data = NULL;
1967 		pool_cache_put(mb_cache, m);
1968 	}
1969 }
1970 
1971 /*
1972  * Free a single mbuf and associated external storage. Return the
1973  * successor, if any.
1974  */
1975 struct mbuf *
1976 m_free(struct mbuf *m)
1977 {
1978 	struct mbuf *n;
1979 
1980 	mowner_revoke(m, 1, m->m_flags);
1981 	mbstat_type_add(m->m_type, -1);
1982 
1983 	if (m->m_flags & M_PKTHDR)
1984 		m_tag_delete_chain(m);
1985 
1986 	n = m->m_next;
1987 
1988 	if (m->m_flags & M_EXT) {
1989 		m_ext_free(m);
1990 	} else {
1991 		if (__predict_false(m->m_type == MT_FREE)) {
1992 			panic("mbuf %p already freed", m);
1993 		}
1994 		m->m_type = MT_FREE;
1995 		m->m_data = NULL;
1996 		pool_cache_put(mb_cache, m);
1997 	}
1998 
1999 	return n;
2000 }
2001 
2002 void
2003 m_freem(struct mbuf *m)
2004 {
2005 	if (m == NULL)
2006 		return;
2007 	do {
2008 		m = m_free(m);
2009 	} while (m);
2010 }
2011 
2012 #if defined(DDB)
2013 void
2014 m_print(const struct mbuf *m, const char *modif, void (*pr)(const char *, ...))
2015 {
2016 	char ch;
2017 	bool opt_c = false;
2018 	bool opt_d = false;
2019 #if NETHER > 0
2020 	bool opt_v = false;
2021 	const struct mbuf *m0 = NULL;
2022 #endif
2023 	int no = 0;
2024 	char buf[512];
2025 
2026 	while ((ch = *(modif++)) != '\0') {
2027 		switch (ch) {
2028 		case 'c':
2029 			opt_c = true;
2030 			break;
2031 		case 'd':
2032 			opt_d = true;
2033 			break;
2034 #if NETHER > 0
2035 		case 'v':
2036 			opt_v = true;
2037 			m0 = m;
2038 			break;
2039 #endif
2040 		default:
2041 			break;
2042 		}
2043 	}
2044 
2045 nextchain:
2046 	(*pr)("MBUF(%d) %p\n", no, m);
2047 	snprintb(buf, sizeof(buf), M_FLAGS_BITS, (u_int)m->m_flags);
2048 	(*pr)("  data=%p, len=%d, type=%d, flags=%s\n",
2049 	    m->m_data, m->m_len, m->m_type, buf);
2050 	if (opt_d) {
2051 		int i;
2052 		unsigned char *p = m->m_data;
2053 
2054 		(*pr)("  data:");
2055 
2056 		for (i = 0; i < m->m_len; i++) {
2057 			if (i % 16 == 0)
2058 				(*pr)("\n");
2059 			(*pr)(" %02x", p[i]);
2060 		}
2061 
2062 		(*pr)("\n");
2063 	}
2064 	(*pr)("  owner=%p, next=%p, nextpkt=%p\n", m->m_owner, m->m_next,
2065 	    m->m_nextpkt);
2066 	(*pr)("  leadingspace=%u, trailingspace=%u, readonly=%u\n",
2067 	    (int)M_LEADINGSPACE(m), (int)M_TRAILINGSPACE(m),
2068 	    (int)M_READONLY(m));
2069 	if ((m->m_flags & M_PKTHDR) != 0) {
2070 		snprintb(buf, sizeof(buf), M_CSUM_BITS, m->m_pkthdr.csum_flags);
2071 		(*pr)("  pktlen=%d, rcvif=%p, csum_flags=%s, csum_data=0x%"
2072 		    PRIx32 ", segsz=%u\n",
2073 		    m->m_pkthdr.len, m_get_rcvif_NOMPSAFE(m),
2074 		    buf, m->m_pkthdr.csum_data, m->m_pkthdr.segsz);
2075 	}
2076 	if ((m->m_flags & M_EXT)) {
2077 		(*pr)("  ext_refcnt=%u, ext_buf=%p, ext_size=%zd, "
2078 		    "ext_free=%p, ext_arg=%p\n",
2079 		    m->m_ext.ext_refcnt,
2080 		    m->m_ext.ext_buf, m->m_ext.ext_size,
2081 		    m->m_ext.ext_free, m->m_ext.ext_arg);
2082 	}
2083 	if ((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0) {
2084 		vaddr_t sva = (vaddr_t)m->m_ext.ext_buf;
2085 		vaddr_t eva = sva + m->m_ext.ext_size;
2086 		int n = (round_page(eva) - trunc_page(sva)) >> PAGE_SHIFT;
2087 		int i;
2088 
2089 		(*pr)("  pages:");
2090 		for (i = 0; i < n; i ++) {
2091 			(*pr)(" %p", m->m_ext.ext_pgs[i]);
2092 		}
2093 		(*pr)("\n");
2094 	}
2095 
2096 	if (opt_c) {
2097 		m = m->m_next;
2098 		if (m != NULL) {
2099 			no++;
2100 			goto nextchain;
2101 		}
2102 	}
2103 
2104 #if NETHER > 0
2105 	if (opt_v && m0)
2106 		m_examine(m0, AF_ETHER, modif, pr);
2107 #endif
2108 }
2109 #endif /* defined(DDB) */
2110 
2111 #if defined(MBUFTRACE)
2112 void
2113 mowner_init_owner(struct mowner *mo, const char *name, const char *descr)
2114 {
2115 	memset(mo, 0, sizeof(*mo));
2116 	strlcpy(mo->mo_name, name, sizeof(mo->mo_name));
2117 	strlcpy(mo->mo_descr, descr, sizeof(mo->mo_descr));
2118 }
2119 
2120 void
2121 mowner_attach(struct mowner *mo)
2122 {
2123 
2124 	KASSERT(mo->mo_counters == NULL);
2125 	mo->mo_counters = percpu_alloc(sizeof(struct mowner_counter));
2126 
2127 	/* XXX lock */
2128 	LIST_INSERT_HEAD(&mowners, mo, mo_link);
2129 }
2130 
2131 void
2132 mowner_detach(struct mowner *mo)
2133 {
2134 
2135 	KASSERT(mo->mo_counters != NULL);
2136 
2137 	/* XXX lock */
2138 	LIST_REMOVE(mo, mo_link);
2139 
2140 	percpu_free(mo->mo_counters, sizeof(struct mowner_counter));
2141 	mo->mo_counters = NULL;
2142 }
2143 
2144 void
2145 mowner_init(struct mbuf *m, int type)
2146 {
2147 	struct mowner_counter *mc;
2148 	struct mowner *mo;
2149 	int s;
2150 
2151 	m->m_owner = mo = &unknown_mowners[type];
2152 	s = splvm();
2153 	mc = percpu_getref(mo->mo_counters);
2154 	mc->mc_counter[MOWNER_COUNTER_CLAIMS]++;
2155 	percpu_putref(mo->mo_counters);
2156 	splx(s);
2157 }
2158 
2159 void
2160 mowner_ref(struct mbuf *m, int flags)
2161 {
2162 	struct mowner *mo = m->m_owner;
2163 	struct mowner_counter *mc;
2164 	int s;
2165 
2166 	s = splvm();
2167 	mc = percpu_getref(mo->mo_counters);
2168 	if ((flags & M_EXT) != 0)
2169 		mc->mc_counter[MOWNER_COUNTER_EXT_CLAIMS]++;
2170 	if ((flags & M_EXT_CLUSTER) != 0)
2171 		mc->mc_counter[MOWNER_COUNTER_CLUSTER_CLAIMS]++;
2172 	percpu_putref(mo->mo_counters);
2173 	splx(s);
2174 }
2175 
2176 void
2177 mowner_revoke(struct mbuf *m, bool all, int flags)
2178 {
2179 	struct mowner *mo = m->m_owner;
2180 	struct mowner_counter *mc;
2181 	int s;
2182 
2183 	s = splvm();
2184 	mc = percpu_getref(mo->mo_counters);
2185 	if ((flags & M_EXT) != 0)
2186 		mc->mc_counter[MOWNER_COUNTER_EXT_RELEASES]++;
2187 	if ((flags & M_EXT_CLUSTER) != 0)
2188 		mc->mc_counter[MOWNER_COUNTER_CLUSTER_RELEASES]++;
2189 	if (all)
2190 		mc->mc_counter[MOWNER_COUNTER_RELEASES]++;
2191 	percpu_putref(mo->mo_counters);
2192 	splx(s);
2193 	if (all)
2194 		m->m_owner = &revoked_mowner;
2195 }
2196 
2197 static void
2198 mowner_claim(struct mbuf *m, struct mowner *mo)
2199 {
2200 	struct mowner_counter *mc;
2201 	int flags = m->m_flags;
2202 	int s;
2203 
2204 	s = splvm();
2205 	mc = percpu_getref(mo->mo_counters);
2206 	mc->mc_counter[MOWNER_COUNTER_CLAIMS]++;
2207 	if ((flags & M_EXT) != 0)
2208 		mc->mc_counter[MOWNER_COUNTER_EXT_CLAIMS]++;
2209 	if ((flags & M_EXT_CLUSTER) != 0)
2210 		mc->mc_counter[MOWNER_COUNTER_CLUSTER_CLAIMS]++;
2211 	percpu_putref(mo->mo_counters);
2212 	splx(s);
2213 	m->m_owner = mo;
2214 }
2215 
2216 void
2217 m_claim(struct mbuf *m, struct mowner *mo)
2218 {
2219 
2220 	if (m->m_owner == mo || mo == NULL)
2221 		return;
2222 
2223 	mowner_revoke(m, true, m->m_flags);
2224 	mowner_claim(m, mo);
2225 }
2226 
2227 void
2228 m_claimm(struct mbuf *m, struct mowner *mo)
2229 {
2230 
2231 	for (; m != NULL; m = m->m_next)
2232 		m_claim(m, mo);
2233 }
2234 #endif /* defined(MBUFTRACE) */
2235 
2236 #ifdef DIAGNOSTIC
2237 /*
2238  * Verify that the mbuf chain is not malformed. Used only for diagnostic.
2239  * Panics on error.
2240  */
2241 void
2242 m_verify_packet(struct mbuf *m)
2243 {
2244 	struct mbuf *n = m;
2245 	char *low, *high, *dat;
2246 	int totlen = 0, len;
2247 
2248 	if (__predict_false((m->m_flags & M_PKTHDR) == 0)) {
2249 		panic("%s: mbuf doesn't have M_PKTHDR", __func__);
2250 	}
2251 
2252 	while (n != NULL) {
2253 		if (__predict_false(n->m_type == MT_FREE)) {
2254 			panic("%s: mbuf already freed (n = %p)", __func__, n);
2255 		}
2256 #if 0
2257 		/*
2258 		 * This ought to be a rule of the mbuf API. Unfortunately,
2259 		 * many places don't respect that rule.
2260 		 */
2261 		if (__predict_false((n != m) && (n->m_flags & M_PKTHDR) != 0)) {
2262 			panic("%s: M_PKTHDR set on secondary mbuf", __func__);
2263 		}
2264 #endif
2265 		if (__predict_false(n->m_nextpkt != NULL)) {
2266 			panic("%s: m_nextpkt not null (m_nextpkt = %p)",
2267 			    __func__, n->m_nextpkt);
2268 		}
2269 
2270 		dat = n->m_data;
2271 		len = n->m_len;
2272 		if (__predict_false(len < 0)) {
2273 			panic("%s: incorrect length (len = %d)", __func__, len);
2274 		}
2275 
2276 		low = M_BUFADDR(n);
2277 		high = low + M_BUFSIZE(n);
2278 		if (__predict_false((dat < low) || (dat + len > high))) {
2279 			panic("%s: m_data not in packet"
2280 			    "(dat = %p, len = %d, low = %p, high = %p)",
2281 			    __func__, dat, len, low, high);
2282 		}
2283 
2284 		totlen += len;
2285 		n = n->m_next;
2286 	}
2287 
2288 	if (__predict_false(totlen != m->m_pkthdr.len)) {
2289 		panic("%s: inconsistent mbuf length (%d != %d)", __func__,
2290 		    totlen, m->m_pkthdr.len);
2291 	}
2292 }
2293 #endif
2294 
2295 struct m_tag *
2296 m_tag_get(int type, int len, int wait)
2297 {
2298 	struct m_tag *t;
2299 
2300 	if (len < 0)
2301 		return NULL;
2302 	t = malloc(len + sizeof(struct m_tag), M_PACKET_TAGS, wait);
2303 	if (t == NULL)
2304 		return NULL;
2305 	t->m_tag_id = type;
2306 	t->m_tag_len = len;
2307 	return t;
2308 }
2309 
2310 void
2311 m_tag_free(struct m_tag *t)
2312 {
2313 	free(t, M_PACKET_TAGS);
2314 }
2315 
2316 void
2317 m_tag_prepend(struct mbuf *m, struct m_tag *t)
2318 {
2319 	KASSERT((m->m_flags & M_PKTHDR) != 0);
2320 	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
2321 }
2322 
2323 void
2324 m_tag_unlink(struct mbuf *m, struct m_tag *t)
2325 {
2326 	KASSERT((m->m_flags & M_PKTHDR) != 0);
2327 	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
2328 }
2329 
2330 void
2331 m_tag_delete(struct mbuf *m, struct m_tag *t)
2332 {
2333 	m_tag_unlink(m, t);
2334 	m_tag_free(t);
2335 }
2336 
2337 void
2338 m_tag_delete_chain(struct mbuf *m)
2339 {
2340 	struct m_tag *p, *q;
2341 
2342 	KASSERT((m->m_flags & M_PKTHDR) != 0);
2343 
2344 	p = SLIST_FIRST(&m->m_pkthdr.tags);
2345 	if (p == NULL)
2346 		return;
2347 	while ((q = SLIST_NEXT(p, m_tag_link)) != NULL)
2348 		m_tag_delete(m, q);
2349 	m_tag_delete(m, p);
2350 }
2351 
2352 struct m_tag *
2353 m_tag_find(const struct mbuf *m, int type)
2354 {
2355 	struct m_tag *p;
2356 
2357 	KASSERT((m->m_flags & M_PKTHDR) != 0);
2358 
2359 	p = SLIST_FIRST(&m->m_pkthdr.tags);
2360 	while (p != NULL) {
2361 		if (p->m_tag_id == type)
2362 			return p;
2363 		p = SLIST_NEXT(p, m_tag_link);
2364 	}
2365 	return NULL;
2366 }
2367 
2368 struct m_tag *
2369 m_tag_copy(struct m_tag *t)
2370 {
2371 	struct m_tag *p;
2372 
2373 	p = m_tag_get(t->m_tag_id, t->m_tag_len, M_NOWAIT);
2374 	if (p == NULL)
2375 		return NULL;
2376 	memcpy(p + 1, t + 1, t->m_tag_len);
2377 	return p;
2378 }
2379 
2380 /*
2381  * Copy two tag chains. The destination mbuf (to) loses any attached
2382  * tags even if the operation fails. This should not be a problem, as
2383  * m_tag_copy_chain() is typically called with a newly-allocated
2384  * destination mbuf.
2385  */
2386 int
2387 m_tag_copy_chain(struct mbuf *to, struct mbuf *from)
2388 {
2389 	struct m_tag *p, *t, *tprev = NULL;
2390 
2391 	KASSERT((from->m_flags & M_PKTHDR) != 0);
2392 
2393 	m_tag_delete_chain(to);
2394 	SLIST_FOREACH(p, &from->m_pkthdr.tags, m_tag_link) {
2395 		t = m_tag_copy(p);
2396 		if (t == NULL) {
2397 			m_tag_delete_chain(to);
2398 			return 0;
2399 		}
2400 		if (tprev == NULL)
2401 			SLIST_INSERT_HEAD(&to->m_pkthdr.tags, t, m_tag_link);
2402 		else
2403 			SLIST_INSERT_AFTER(tprev, t, m_tag_link);
2404 		tprev = t;
2405 	}
2406 	return 1;
2407 }
2408