xref: /netbsd-src/sys/kern/uipc_mbuf.c (revision aef5eb5f59cdfe8314f1b5f78ac04eb144e44010)
1 /*	$NetBSD: uipc_mbuf.c,v 1.246 2022/04/09 23:38:33 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 1.246 2022/04/09 23:38:33 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 	default:
293 		return EOPNOTSUPP;
294 	}
295 
296 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
297 	if (error || newp == NULL)
298 		return error;
299 	if (newval < 0)
300 		return EINVAL;
301 
302 	switch (node.sysctl_num) {
303 	case MBUF_NMBCLUSTERS:
304 		if (newval < nmbclusters)
305 			return EINVAL;
306 		if (newval > nmbclusters_limit())
307 			return EINVAL;
308 		nmbclusters = newval;
309 		pool_cache_sethardlimit(mcl_cache, nmbclusters,
310 		    mclpool_warnmsg, 60);
311 		break;
312 	case MBUF_MBLOWAT:
313 		mblowat = newval;
314 		pool_cache_setlowat(mb_cache, mblowat);
315 		break;
316 	case MBUF_MCLLOWAT:
317 		mcllowat = newval;
318 		pool_cache_setlowat(mcl_cache, mcllowat);
319 		break;
320 	}
321 
322 	return 0;
323 }
324 
325 #ifdef MBUFTRACE
326 static void
327 mowner_convert_to_user_cb(void *v1, void *v2, struct cpu_info *ci)
328 {
329 	struct mowner_counter *mc = v1;
330 	struct mowner_user *mo_user = v2;
331 	int i;
332 
333 	for (i = 0; i < MOWNER_COUNTER_NCOUNTERS; i++) {
334 		mo_user->mo_counter[i] += mc->mc_counter[i];
335 	}
336 }
337 
338 static void
339 mowner_convert_to_user(struct mowner *mo, struct mowner_user *mo_user)
340 {
341 
342 	memset(mo_user, 0, sizeof(*mo_user));
343 	CTASSERT(sizeof(mo_user->mo_name) == sizeof(mo->mo_name));
344 	CTASSERT(sizeof(mo_user->mo_descr) == sizeof(mo->mo_descr));
345 	memcpy(mo_user->mo_name, mo->mo_name, sizeof(mo->mo_name));
346 	memcpy(mo_user->mo_descr, mo->mo_descr, sizeof(mo->mo_descr));
347 	percpu_foreach(mo->mo_counters, mowner_convert_to_user_cb, mo_user);
348 }
349 
350 static int
351 sysctl_kern_mbuf_mowners(SYSCTLFN_ARGS)
352 {
353 	struct mowner *mo;
354 	size_t len = 0;
355 	int error = 0;
356 
357 	if (namelen != 0)
358 		return EINVAL;
359 	if (newp != NULL)
360 		return EPERM;
361 
362 	LIST_FOREACH(mo, &mowners, mo_link) {
363 		struct mowner_user mo_user;
364 
365 		mowner_convert_to_user(mo, &mo_user);
366 
367 		if (oldp != NULL) {
368 			if (*oldlenp - len < sizeof(mo_user)) {
369 				error = ENOMEM;
370 				break;
371 			}
372 			error = copyout(&mo_user, (char *)oldp + len,
373 			    sizeof(mo_user));
374 			if (error)
375 				break;
376 		}
377 		len += sizeof(mo_user);
378 	}
379 
380 	if (error == 0)
381 		*oldlenp = len;
382 
383 	return error;
384 }
385 #endif /* MBUFTRACE */
386 
387 void
388 mbstat_type_add(int type, int diff)
389 {
390 	struct mbstat_cpu *mb;
391 	int s;
392 
393 	s = splvm();
394 	mb = percpu_getref(mbstat_percpu);
395 	mb->m_mtypes[type] += diff;
396 	percpu_putref(mbstat_percpu);
397 	splx(s);
398 }
399 
400 static void
401 mbstat_convert_to_user_cb(void *v1, void *v2, struct cpu_info *ci)
402 {
403 	struct mbstat_cpu *mbsc = v1;
404 	struct mbstat *mbs = v2;
405 	int i;
406 
407 	for (i = 0; i < __arraycount(mbs->m_mtypes); i++) {
408 		mbs->m_mtypes[i] += mbsc->m_mtypes[i];
409 	}
410 }
411 
412 static void
413 mbstat_convert_to_user(struct mbstat *mbs)
414 {
415 
416 	memset(mbs, 0, sizeof(*mbs));
417 	mbs->m_drain = mbstat.m_drain;
418 	percpu_foreach(mbstat_percpu, mbstat_convert_to_user_cb, mbs);
419 }
420 
421 static int
422 sysctl_kern_mbuf_stats(SYSCTLFN_ARGS)
423 {
424 	struct sysctlnode node;
425 	struct mbstat mbs;
426 
427 	mbstat_convert_to_user(&mbs);
428 	node = *rnode;
429 	node.sysctl_data = &mbs;
430 	node.sysctl_size = sizeof(mbs);
431 	return sysctl_lookup(SYSCTLFN_CALL(&node));
432 }
433 
434 static void
435 sysctl_kern_mbuf_setup(void)
436 {
437 
438 	KASSERT(mbuf_sysctllog == NULL);
439 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
440 		       CTLFLAG_PERMANENT,
441 		       CTLTYPE_NODE, "mbuf",
442 		       SYSCTL_DESCR("mbuf control variables"),
443 		       NULL, 0, NULL, 0,
444 		       CTL_KERN, KERN_MBUF, CTL_EOL);
445 
446 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
447 		       CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
448 		       CTLTYPE_INT, "msize",
449 		       SYSCTL_DESCR("mbuf base size"),
450 		       NULL, msize, NULL, 0,
451 		       CTL_KERN, KERN_MBUF, MBUF_MSIZE, CTL_EOL);
452 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
453 		       CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
454 		       CTLTYPE_INT, "mclbytes",
455 		       SYSCTL_DESCR("mbuf cluster size"),
456 		       NULL, mclbytes, NULL, 0,
457 		       CTL_KERN, KERN_MBUF, MBUF_MCLBYTES, CTL_EOL);
458 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
459 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
460 		       CTLTYPE_INT, "nmbclusters",
461 		       SYSCTL_DESCR("Limit on the number of mbuf clusters"),
462 		       sysctl_kern_mbuf, 0, &nmbclusters, 0,
463 		       CTL_KERN, KERN_MBUF, MBUF_NMBCLUSTERS, CTL_EOL);
464 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
465 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
466 		       CTLTYPE_INT, "mblowat",
467 		       SYSCTL_DESCR("mbuf low water mark"),
468 		       sysctl_kern_mbuf, 0, &mblowat, 0,
469 		       CTL_KERN, KERN_MBUF, MBUF_MBLOWAT, CTL_EOL);
470 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
471 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
472 		       CTLTYPE_INT, "mcllowat",
473 		       SYSCTL_DESCR("mbuf cluster low water mark"),
474 		       sysctl_kern_mbuf, 0, &mcllowat, 0,
475 		       CTL_KERN, KERN_MBUF, MBUF_MCLLOWAT, CTL_EOL);
476 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
477 		       CTLFLAG_PERMANENT,
478 		       CTLTYPE_STRUCT, "stats",
479 		       SYSCTL_DESCR("mbuf allocation statistics"),
480 		       sysctl_kern_mbuf_stats, 0, NULL, 0,
481 		       CTL_KERN, KERN_MBUF, MBUF_STATS, CTL_EOL);
482 #ifdef MBUFTRACE
483 	sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL,
484 		       CTLFLAG_PERMANENT,
485 		       CTLTYPE_STRUCT, "mowners",
486 		       SYSCTL_DESCR("Information about mbuf owners"),
487 		       sysctl_kern_mbuf_mowners, 0, NULL, 0,
488 		       CTL_KERN, KERN_MBUF, MBUF_MOWNERS, CTL_EOL);
489 #endif
490 }
491 
492 static int
493 mb_ctor(void *arg, void *object, int flags)
494 {
495 	struct mbuf *m = object;
496 
497 #ifdef POOL_VTOPHYS
498 	m->m_paddr = POOL_VTOPHYS(m);
499 #else
500 	m->m_paddr = M_PADDR_INVALID;
501 #endif
502 	return 0;
503 }
504 
505 /*
506  * Add mbuf to the end of a chain
507  */
508 struct mbuf *
509 m_add(struct mbuf *c, struct mbuf *m)
510 {
511 	struct mbuf *n;
512 
513 	if (c == NULL)
514 		return m;
515 
516 	for (n = c; n->m_next != NULL; n = n->m_next)
517 		continue;
518 	n->m_next = m;
519 	return c;
520 }
521 
522 struct mbuf *
523 m_get(int how, int type)
524 {
525 	struct mbuf *m;
526 
527 	KASSERT(type != MT_FREE);
528 
529 	m = pool_cache_get(mb_cache,
530 	    how == M_WAIT ? PR_WAITOK|PR_LIMITFAIL : PR_NOWAIT);
531 	if (m == NULL)
532 		return NULL;
533 	KASSERT(((vaddr_t)m->m_dat & PAGE_MASK) + MLEN <= PAGE_SIZE);
534 
535 	mbstat_type_add(type, 1);
536 
537 	mowner_init(m, type);
538 	m->m_ext_ref = m; /* default */
539 	m->m_type = type;
540 	m->m_len = 0;
541 	m->m_next = NULL;
542 	m->m_nextpkt = NULL; /* default */
543 	m->m_data = m->m_dat;
544 	m->m_flags = 0; /* default */
545 
546 	return m;
547 }
548 
549 struct mbuf *
550 m_gethdr(int how, int type)
551 {
552 	struct mbuf *m;
553 
554 	m = m_get(how, type);
555 	if (m == NULL)
556 		return NULL;
557 
558 	m->m_data = m->m_pktdat;
559 	m->m_flags = M_PKTHDR;
560 
561 	m_reset_rcvif(m);
562 	m->m_pkthdr.len = 0;
563 	m->m_pkthdr.csum_flags = 0;
564 	m->m_pkthdr.csum_data = 0;
565 	m->m_pkthdr.segsz = 0;
566 	m->m_pkthdr.ether_vtag = 0;
567 	m->m_pkthdr.pkthdr_flags = 0;
568 	SLIST_INIT(&m->m_pkthdr.tags);
569 
570 	m->m_pkthdr.pattr_class = NULL;
571 	m->m_pkthdr.pattr_af = AF_UNSPEC;
572 	m->m_pkthdr.pattr_hdr = NULL;
573 
574 	return m;
575 }
576 
577 void
578 m_clget(struct mbuf *m, int how)
579 {
580 	m->m_ext_storage.ext_buf = (char *)pool_cache_get_paddr(mcl_cache,
581 	    how == M_WAIT ? (PR_WAITOK|PR_LIMITFAIL) : PR_NOWAIT,
582 	    &m->m_ext_storage.ext_paddr);
583 
584 	if (m->m_ext_storage.ext_buf == NULL)
585 		return;
586 
587 	KASSERT(((vaddr_t)m->m_ext_storage.ext_buf & PAGE_MASK) + mclbytes
588 	    <= PAGE_SIZE);
589 
590 	MCLINITREFERENCE(m);
591 	m->m_data = m->m_ext.ext_buf;
592 	m->m_flags = (m->m_flags & ~M_EXTCOPYFLAGS) |
593 	    M_EXT|M_EXT_CLUSTER|M_EXT_RW;
594 	m->m_ext.ext_size = MCLBYTES;
595 	m->m_ext.ext_free = NULL;
596 	m->m_ext.ext_arg = NULL;
597 	/* ext_paddr initialized above */
598 
599 	mowner_ref(m, M_EXT|M_EXT_CLUSTER);
600 }
601 
602 struct mbuf *
603 m_getcl(int how, int type, int flags)
604 {
605 	struct mbuf *mp;
606 
607 	if ((flags & M_PKTHDR) != 0)
608 		mp = m_gethdr(how, type);
609 	else
610 		mp = m_get(how, type);
611 
612 	if (mp == NULL)
613 		return NULL;
614 
615 	MCLGET(mp, how);
616 	if ((mp->m_flags & M_EXT) != 0)
617 		return mp;
618 
619 	m_free(mp);
620 	return NULL;
621 }
622 
623 /*
624  * Utility function for M_PREPEND. Do *NOT* use it directly.
625  */
626 struct mbuf *
627 m_prepend(struct mbuf *m, int len, int how)
628 {
629 	struct mbuf *mn;
630 
631 	if (__predict_false(len > MHLEN)) {
632 		panic("%s: len > MHLEN", __func__);
633 	}
634 
635 	KASSERT(len != M_COPYALL);
636 	mn = m_get(how, m->m_type);
637 	if (mn == NULL) {
638 		m_freem(m);
639 		return NULL;
640 	}
641 
642 	if (m->m_flags & M_PKTHDR) {
643 		m_move_pkthdr(mn, m);
644 	} else {
645 		MCLAIM(mn, m->m_owner);
646 	}
647 	mn->m_next = m;
648 	m = mn;
649 
650 	if (m->m_flags & M_PKTHDR) {
651 		if (len < MHLEN)
652 			m_align(m, len);
653 	} else {
654 		if (len < MLEN)
655 			m_align(m, len);
656 	}
657 
658 	m->m_len = len;
659 	return m;
660 }
661 
662 struct mbuf *
663 m_copym(struct mbuf *m, int off, int len, int wait)
664 {
665 	/* Shallow copy on M_EXT. */
666 	return m_copy_internal(m, off, len, wait, false);
667 }
668 
669 struct mbuf *
670 m_dup(struct mbuf *m, int off, int len, int wait)
671 {
672 	/* Deep copy. */
673 	return m_copy_internal(m, off, len, wait, true);
674 }
675 
676 static inline int
677 m_copylen(int len, int copylen)
678 {
679 	return (len == M_COPYALL) ? copylen : uimin(len, copylen);
680 }
681 
682 static struct mbuf *
683 m_copy_internal(struct mbuf *m, int off0, int len, int wait, bool deep)
684 {
685 	struct mbuf *n, **np;
686 	int off = off0;
687 	struct mbuf *top;
688 	int copyhdr = 0;
689 
690 	if (off < 0 || (len != M_COPYALL && len < 0))
691 		panic("%s: off %d, len %d", __func__, off, len);
692 	if (off == 0 && m->m_flags & M_PKTHDR)
693 		copyhdr = 1;
694 	while (off > 0) {
695 		if (m == NULL)
696 			panic("%s: m == NULL, off %d", __func__, off);
697 		if (off < m->m_len)
698 			break;
699 		off -= m->m_len;
700 		m = m->m_next;
701 	}
702 
703 	np = &top;
704 	top = NULL;
705 	while (len == M_COPYALL || len > 0) {
706 		if (m == NULL) {
707 			if (len != M_COPYALL)
708 				panic("%s: m == NULL, len %d [!COPYALL]",
709 				    __func__, len);
710 			break;
711 		}
712 
713 		n = m_get(wait, m->m_type);
714 		*np = n;
715 		if (n == NULL)
716 			goto nospace;
717 		MCLAIM(n, m->m_owner);
718 
719 		if (copyhdr) {
720 			m_copy_pkthdr(n, m);
721 			if (len == M_COPYALL)
722 				n->m_pkthdr.len -= off0;
723 			else
724 				n->m_pkthdr.len = len;
725 			copyhdr = 0;
726 		}
727 		n->m_len = m_copylen(len, m->m_len - off);
728 
729 		if (m->m_flags & M_EXT) {
730 			if (!deep) {
731 				n->m_data = m->m_data + off;
732 				MCLADDREFERENCE(m, n);
733 			} else {
734 				/*
735 				 * We don't care if MCLGET fails. n->m_len is
736 				 * recomputed and handles that.
737 				 */
738 				MCLGET(n, wait);
739 				n->m_len = 0;
740 				n->m_len = M_TRAILINGSPACE(n);
741 				n->m_len = m_copylen(len, n->m_len);
742 				n->m_len = uimin(n->m_len, m->m_len - off);
743 				memcpy(mtod(n, void *), mtod(m, char *) + off,
744 				    (unsigned)n->m_len);
745 			}
746 		} else {
747 			memcpy(mtod(n, void *), mtod(m, char *) + off,
748 			    (unsigned)n->m_len);
749 		}
750 
751 		if (len != M_COPYALL)
752 			len -= n->m_len;
753 		off += n->m_len;
754 
755 		KASSERT(off <= m->m_len);
756 
757 		if (off == m->m_len) {
758 			m = m->m_next;
759 			off = 0;
760 		}
761 		np = &n->m_next;
762 	}
763 
764 	return top;
765 
766 nospace:
767 	m_freem(top);
768 	return NULL;
769 }
770 
771 /*
772  * Copy an entire packet, including header (which must be present).
773  * An optimization of the common case 'm_copym(m, 0, M_COPYALL, how)'.
774  */
775 struct mbuf *
776 m_copypacket(struct mbuf *m, int how)
777 {
778 	struct mbuf *top, *n, *o;
779 
780 	if (__predict_false((m->m_flags & M_PKTHDR) == 0)) {
781 		panic("%s: no header (m = %p)", __func__, m);
782 	}
783 
784 	n = m_get(how, m->m_type);
785 	top = n;
786 	if (!n)
787 		goto nospace;
788 
789 	MCLAIM(n, m->m_owner);
790 	m_copy_pkthdr(n, m);
791 	n->m_len = m->m_len;
792 	if (m->m_flags & M_EXT) {
793 		n->m_data = m->m_data;
794 		MCLADDREFERENCE(m, n);
795 	} else {
796 		memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
797 	}
798 
799 	m = m->m_next;
800 	while (m) {
801 		o = m_get(how, m->m_type);
802 		if (!o)
803 			goto nospace;
804 
805 		MCLAIM(o, m->m_owner);
806 		n->m_next = o;
807 		n = n->m_next;
808 
809 		n->m_len = m->m_len;
810 		if (m->m_flags & M_EXT) {
811 			n->m_data = m->m_data;
812 			MCLADDREFERENCE(m, n);
813 		} else {
814 			memcpy(mtod(n, char *), mtod(m, char *), n->m_len);
815 		}
816 
817 		m = m->m_next;
818 	}
819 	return top;
820 
821 nospace:
822 	m_freem(top);
823 	return NULL;
824 }
825 
826 void
827 m_copydata(struct mbuf *m, int off, int len, void *cp)
828 {
829 	unsigned int count;
830 	struct mbuf *m0 = m;
831 	int len0 = len;
832 	int off0 = off;
833 	void *cp0 = cp;
834 
835 	KASSERT(len != M_COPYALL);
836 	if (off < 0 || len < 0)
837 		panic("m_copydata: off %d, len %d", off, len);
838 	while (off > 0) {
839 		if (m == NULL)
840 			panic("m_copydata(%p,%d,%d,%p): m=NULL, off=%d (%d)",
841 			    m0, len0, off0, cp0, off, off0 - off);
842 		if (off < m->m_len)
843 			break;
844 		off -= m->m_len;
845 		m = m->m_next;
846 	}
847 	while (len > 0) {
848 		if (m == NULL)
849 			panic("m_copydata(%p,%d,%d,%p): "
850 			    "m=NULL, off=%d (%d), len=%d (%d)",
851 			    m0, len0, off0, cp0,
852 			    off, off0 - off, len, len0 - len);
853 		count = uimin(m->m_len - off, len);
854 		memcpy(cp, mtod(m, char *) + off, count);
855 		len -= count;
856 		cp = (char *)cp + count;
857 		off = 0;
858 		m = m->m_next;
859 	}
860 }
861 
862 /*
863  * Concatenate mbuf chain n to m.
864  * n might be copied into m (when n->m_len is small), therefore data portion of
865  * n could be copied into an mbuf of different mbuf type.
866  * Any m_pkthdr is not updated.
867  */
868 void
869 m_cat(struct mbuf *m, struct mbuf *n)
870 {
871 
872 	while (m->m_next)
873 		m = m->m_next;
874 	while (n) {
875 		if (M_READONLY(m) || n->m_len > M_TRAILINGSPACE(m)) {
876 			/* just join the two chains */
877 			m->m_next = n;
878 			return;
879 		}
880 		/* splat the data from one into the other */
881 		memcpy(mtod(m, char *) + m->m_len, mtod(n, void *),
882 		    (u_int)n->m_len);
883 		m->m_len += n->m_len;
884 		n = m_free(n);
885 	}
886 }
887 
888 void
889 m_adj(struct mbuf *mp, int req_len)
890 {
891 	int len = req_len;
892 	struct mbuf *m;
893 	int count;
894 
895 	if ((m = mp) == NULL)
896 		return;
897 	if (len >= 0) {
898 		/*
899 		 * Trim from head.
900 		 */
901 		while (m != NULL && len > 0) {
902 			if (m->m_len <= len) {
903 				len -= m->m_len;
904 				m->m_len = 0;
905 				m = m->m_next;
906 			} else {
907 				m->m_len -= len;
908 				m->m_data += len;
909 				len = 0;
910 			}
911 		}
912 		if (mp->m_flags & M_PKTHDR)
913 			mp->m_pkthdr.len -= (req_len - len);
914 	} else {
915 		/*
916 		 * Trim from tail.  Scan the mbuf chain,
917 		 * calculating its length and finding the last mbuf.
918 		 * If the adjustment only affects this mbuf, then just
919 		 * adjust and return.  Otherwise, rescan and truncate
920 		 * after the remaining size.
921 		 */
922 		len = -len;
923 		count = 0;
924 		for (;;) {
925 			count += m->m_len;
926 			if (m->m_next == NULL)
927 				break;
928 			m = m->m_next;
929 		}
930 		if (m->m_len >= len) {
931 			m->m_len -= len;
932 			if (mp->m_flags & M_PKTHDR)
933 				mp->m_pkthdr.len -= len;
934 			return;
935 		}
936 
937 		count -= len;
938 		if (count < 0)
939 			count = 0;
940 
941 		/*
942 		 * Correct length for chain is "count".
943 		 * Find the mbuf with last data, adjust its length,
944 		 * and toss data from remaining mbufs on chain.
945 		 */
946 		m = mp;
947 		if (m->m_flags & M_PKTHDR)
948 			m->m_pkthdr.len = count;
949 		for (; m; m = m->m_next) {
950 			if (m->m_len >= count) {
951 				m->m_len = count;
952 				break;
953 			}
954 			count -= m->m_len;
955 		}
956 		if (m) {
957 			while (m->m_next)
958 				(m = m->m_next)->m_len = 0;
959 		}
960 	}
961 }
962 
963 /*
964  * m_ensure_contig: rearrange an mbuf chain that given length of bytes
965  * would be contiguous and in the data area of an mbuf (therefore, mtod()
966  * would work for a structure of given length).
967  *
968  * => On success, returns true and the resulting mbuf chain; false otherwise.
969  * => The mbuf chain may change, but is always preserved valid.
970  */
971 bool
972 m_ensure_contig(struct mbuf **m0, int len)
973 {
974 	struct mbuf *n = *m0, *m;
975 	size_t count, space;
976 
977 	KASSERT(len != M_COPYALL);
978 	/*
979 	 * If first mbuf has no cluster, and has room for len bytes
980 	 * without shifting current data, pullup into it,
981 	 * otherwise allocate a new mbuf to prepend to the chain.
982 	 */
983 	if ((n->m_flags & M_EXT) == 0 &&
984 	    n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
985 		if (n->m_len >= len) {
986 			return true;
987 		}
988 		m = n;
989 		n = n->m_next;
990 		len -= m->m_len;
991 	} else {
992 		if (len > MHLEN) {
993 			return false;
994 		}
995 		m = m_get(M_DONTWAIT, n->m_type);
996 		if (m == NULL) {
997 			return false;
998 		}
999 		MCLAIM(m, n->m_owner);
1000 		if (n->m_flags & M_PKTHDR) {
1001 			m_move_pkthdr(m, n);
1002 		}
1003 	}
1004 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1005 	do {
1006 		count = MIN(MIN(MAX(len, max_protohdr), space), n->m_len);
1007 		memcpy(mtod(m, char *) + m->m_len, mtod(n, void *),
1008 		  (unsigned)count);
1009 		len -= count;
1010 		m->m_len += count;
1011 		n->m_len -= count;
1012 		space -= count;
1013 		if (n->m_len)
1014 			n->m_data += count;
1015 		else
1016 			n = m_free(n);
1017 	} while (len > 0 && n);
1018 
1019 	m->m_next = n;
1020 	*m0 = m;
1021 
1022 	return len <= 0;
1023 }
1024 
1025 /*
1026  * m_pullup: same as m_ensure_contig(), but destroys mbuf chain on error.
1027  */
1028 struct mbuf *
1029 m_pullup(struct mbuf *n, int len)
1030 {
1031 	struct mbuf *m = n;
1032 
1033 	KASSERT(len != M_COPYALL);
1034 	if (!m_ensure_contig(&m, len)) {
1035 		KASSERT(m != NULL);
1036 		m_freem(m);
1037 		m = NULL;
1038 	}
1039 	return m;
1040 }
1041 
1042 /*
1043  * ensure that [off, off + len) is contiguous on the mbuf chain "m".
1044  * packet chain before "off" is kept untouched.
1045  * if offp == NULL, the target will start at <retval, 0> on resulting chain.
1046  * if offp != NULL, the target will start at <retval, *offp> on resulting chain.
1047  *
1048  * on error return (NULL return value), original "m" will be freed.
1049  *
1050  * XXX M_TRAILINGSPACE/M_LEADINGSPACE on shared cluster (sharedcluster)
1051  */
1052 struct mbuf *
1053 m_pulldown(struct mbuf *m, int off, int len, int *offp)
1054 {
1055 	struct mbuf *n, *o;
1056 	int hlen, tlen, olen;
1057 	int sharedcluster;
1058 
1059 	/* Check invalid arguments. */
1060 	if (m == NULL)
1061 		panic("%s: m == NULL", __func__);
1062 	if (len > MCLBYTES) {
1063 		m_freem(m);
1064 		return NULL;
1065 	}
1066 
1067 	n = m;
1068 	while (n != NULL && off > 0) {
1069 		if (n->m_len > off)
1070 			break;
1071 		off -= n->m_len;
1072 		n = n->m_next;
1073 	}
1074 	/* Be sure to point non-empty mbuf. */
1075 	while (n != NULL && n->m_len == 0)
1076 		n = n->m_next;
1077 	if (!n) {
1078 		m_freem(m);
1079 		return NULL;	/* mbuf chain too short */
1080 	}
1081 
1082 	sharedcluster = M_READONLY(n);
1083 
1084 	/*
1085 	 * The target data is on <n, off>. If we got enough data on the mbuf
1086 	 * "n", we're done.
1087 	 */
1088 #ifdef __NO_STRICT_ALIGNMENT
1089 	if ((off == 0 || offp) && len <= n->m_len - off && !sharedcluster)
1090 #else
1091 	if ((off == 0 || offp) && len <= n->m_len - off && !sharedcluster &&
1092 	    ALIGNED_POINTER((mtod(n, char *) + off), uint32_t))
1093 #endif
1094 		goto ok;
1095 
1096 	/*
1097 	 * When (len <= n->m_len - off) and (off != 0), it is a special case.
1098 	 * Len bytes from <n, off> sit in single mbuf, but the caller does
1099 	 * not like the starting position (off).
1100 	 *
1101 	 * Chop the current mbuf into two pieces, set off to 0.
1102 	 */
1103 	if (len <= n->m_len - off) {
1104 		struct mbuf *mlast;
1105 
1106 		o = m_dup(n, off, n->m_len - off, M_DONTWAIT);
1107 		if (o == NULL) {
1108 			m_freem(m);
1109 			return NULL;	/* ENOBUFS */
1110 		}
1111 		KASSERT(o->m_len >= len);
1112 		for (mlast = o; mlast->m_next != NULL; mlast = mlast->m_next)
1113 			;
1114 		n->m_len = off;
1115 		mlast->m_next = n->m_next;
1116 		n->m_next = o;
1117 		n = o;
1118 		off = 0;
1119 		goto ok;
1120 	}
1121 
1122 	/*
1123 	 * We need to take hlen from <n, off> and tlen from <n->m_next, 0>,
1124 	 * and construct contiguous mbuf with m_len == len.
1125 	 *
1126 	 * Note that hlen + tlen == len, and tlen > 0.
1127 	 */
1128 	hlen = n->m_len - off;
1129 	tlen = len - hlen;
1130 
1131 	/*
1132 	 * Ensure that we have enough trailing data on mbuf chain. If not,
1133 	 * we can do nothing about the chain.
1134 	 */
1135 	olen = 0;
1136 	for (o = n->m_next; o != NULL; o = o->m_next)
1137 		olen += o->m_len;
1138 	if (hlen + olen < len) {
1139 		m_freem(m);
1140 		return NULL;	/* mbuf chain too short */
1141 	}
1142 
1143 	/*
1144 	 * Easy cases first. We need to use m_copydata() to get data from
1145 	 * <n->m_next, 0>.
1146 	 */
1147 	if ((off == 0 || offp) && M_TRAILINGSPACE(n) >= tlen &&
1148 	    !sharedcluster) {
1149 		m_copydata(n->m_next, 0, tlen, mtod(n, char *) + n->m_len);
1150 		n->m_len += tlen;
1151 		m_adj(n->m_next, tlen);
1152 		goto ok;
1153 	}
1154 	if ((off == 0 || offp) && M_LEADINGSPACE(n->m_next) >= hlen &&
1155 #ifndef __NO_STRICT_ALIGNMENT
1156 	    ALIGNED_POINTER((n->m_next->m_data - hlen), uint32_t) &&
1157 #endif
1158 	    !sharedcluster && n->m_next->m_len >= tlen) {
1159 		n->m_next->m_data -= hlen;
1160 		n->m_next->m_len += hlen;
1161 		memcpy(mtod(n->m_next, void *), mtod(n, char *) + off, hlen);
1162 		n->m_len -= hlen;
1163 		n = n->m_next;
1164 		off = 0;
1165 		goto ok;
1166 	}
1167 
1168 	/*
1169 	 * Now, we need to do the hard way. Don't copy as there's no room
1170 	 * on both ends.
1171 	 */
1172 	o = m_get(M_DONTWAIT, m->m_type);
1173 	if (o && len > MLEN) {
1174 		MCLGET(o, M_DONTWAIT);
1175 		if ((o->m_flags & M_EXT) == 0) {
1176 			m_free(o);
1177 			o = NULL;
1178 		}
1179 	}
1180 	if (!o) {
1181 		m_freem(m);
1182 		return NULL;	/* ENOBUFS */
1183 	}
1184 	/* get hlen from <n, off> into <o, 0> */
1185 	o->m_len = hlen;
1186 	memcpy(mtod(o, void *), mtod(n, char *) + off, hlen);
1187 	n->m_len -= hlen;
1188 	/* get tlen from <n->m_next, 0> into <o, hlen> */
1189 	m_copydata(n->m_next, 0, tlen, mtod(o, char *) + o->m_len);
1190 	o->m_len += tlen;
1191 	m_adj(n->m_next, tlen);
1192 	o->m_next = n->m_next;
1193 	n->m_next = o;
1194 	n = o;
1195 	off = 0;
1196 
1197 ok:
1198 	if (offp)
1199 		*offp = off;
1200 	return n;
1201 }
1202 
1203 /*
1204  * Like m_pullup(), except a new mbuf is always allocated, and we allow
1205  * the amount of empty space before the data in the new mbuf to be specified
1206  * (in the event that the caller expects to prepend later).
1207  */
1208 struct mbuf *
1209 m_copyup(struct mbuf *n, int len, int dstoff)
1210 {
1211 	struct mbuf *m;
1212 	int count, space;
1213 
1214 	KASSERT(len != M_COPYALL);
1215 	if (len > ((int)MHLEN - dstoff))
1216 		goto bad;
1217 	m = m_get(M_DONTWAIT, n->m_type);
1218 	if (m == NULL)
1219 		goto bad;
1220 	MCLAIM(m, n->m_owner);
1221 	if (n->m_flags & M_PKTHDR) {
1222 		m_move_pkthdr(m, n);
1223 	}
1224 	m->m_data += dstoff;
1225 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1226 	do {
1227 		count = uimin(uimin(uimax(len, max_protohdr), space), n->m_len);
1228 		memcpy(mtod(m, char *) + m->m_len, mtod(n, void *),
1229 		    (unsigned)count);
1230 		len -= count;
1231 		m->m_len += count;
1232 		n->m_len -= count;
1233 		space -= count;
1234 		if (n->m_len)
1235 			n->m_data += count;
1236 		else
1237 			n = m_free(n);
1238 	} while (len > 0 && n);
1239 	if (len > 0) {
1240 		(void) m_free(m);
1241 		goto bad;
1242 	}
1243 	m->m_next = n;
1244 	return m;
1245  bad:
1246 	m_freem(n);
1247 	return NULL;
1248 }
1249 
1250 struct mbuf *
1251 m_split(struct mbuf *m0, int len, int wait)
1252 {
1253 	return m_split_internal(m0, len, wait, true);
1254 }
1255 
1256 static struct mbuf *
1257 m_split_internal(struct mbuf *m0, int len0, int wait, bool copyhdr)
1258 {
1259 	struct mbuf *m, *n;
1260 	unsigned len = len0, remain, len_save;
1261 
1262 	KASSERT(len0 != M_COPYALL);
1263 	for (m = m0; m && len > m->m_len; m = m->m_next)
1264 		len -= m->m_len;
1265 	if (m == NULL)
1266 		return NULL;
1267 
1268 	remain = m->m_len - len;
1269 	if (copyhdr && (m0->m_flags & M_PKTHDR)) {
1270 		n = m_gethdr(wait, m0->m_type);
1271 		if (n == NULL)
1272 			return NULL;
1273 
1274 		MCLAIM(n, m0->m_owner);
1275 		m_copy_rcvif(n, m0);
1276 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1277 		len_save = m0->m_pkthdr.len;
1278 		m0->m_pkthdr.len = len0;
1279 
1280 		if (m->m_flags & M_EXT)
1281 			goto extpacket;
1282 
1283 		if (remain > MHLEN) {
1284 			/* m can't be the lead packet */
1285 			m_align(n, 0);
1286 			n->m_len = 0;
1287 			n->m_next = m_split(m, len, wait);
1288 			if (n->m_next == NULL) {
1289 				(void)m_free(n);
1290 				m0->m_pkthdr.len = len_save;
1291 				return NULL;
1292 			}
1293 			return n;
1294 		} else {
1295 			m_align(n, remain);
1296 		}
1297 	} else if (remain == 0) {
1298 		n = m->m_next;
1299 		m->m_next = NULL;
1300 		return n;
1301 	} else {
1302 		n = m_get(wait, m->m_type);
1303 		if (n == NULL)
1304 			return NULL;
1305 		MCLAIM(n, m->m_owner);
1306 		m_align(n, remain);
1307 	}
1308 
1309 extpacket:
1310 	if (m->m_flags & M_EXT) {
1311 		n->m_data = m->m_data + len;
1312 		MCLADDREFERENCE(m, n);
1313 	} else {
1314 		memcpy(mtod(n, void *), mtod(m, char *) + len, remain);
1315 	}
1316 
1317 	n->m_len = remain;
1318 	m->m_len = len;
1319 	n->m_next = m->m_next;
1320 	m->m_next = NULL;
1321 	return n;
1322 }
1323 
1324 /*
1325  * Routine to copy from device local memory into mbufs.
1326  */
1327 struct mbuf *
1328 m_devget(char *buf, int totlen, int off, struct ifnet *ifp)
1329 {
1330 	struct mbuf *m;
1331 	struct mbuf *top = NULL, **mp = &top;
1332 	char *cp, *epkt;
1333 	int len;
1334 
1335 	cp = buf;
1336 	epkt = cp + totlen;
1337 	if (off) {
1338 		/*
1339 		 * If 'off' is non-zero, packet is trailer-encapsulated,
1340 		 * so we have to skip the type and length fields.
1341 		 */
1342 		cp += off + 2 * sizeof(uint16_t);
1343 		totlen -= 2 * sizeof(uint16_t);
1344 	}
1345 
1346 	m = m_gethdr(M_DONTWAIT, MT_DATA);
1347 	if (m == NULL)
1348 		return NULL;
1349 	m_set_rcvif(m, ifp);
1350 	m->m_pkthdr.len = totlen;
1351 	m->m_len = MHLEN;
1352 
1353 	while (totlen > 0) {
1354 		if (top) {
1355 			m = m_get(M_DONTWAIT, MT_DATA);
1356 			if (m == NULL) {
1357 				m_freem(top);
1358 				return NULL;
1359 			}
1360 			m->m_len = MLEN;
1361 		}
1362 
1363 		len = uimin(totlen, epkt - cp);
1364 
1365 		if (len >= MINCLSIZE) {
1366 			MCLGET(m, M_DONTWAIT);
1367 			if ((m->m_flags & M_EXT) == 0) {
1368 				m_free(m);
1369 				m_freem(top);
1370 				return NULL;
1371 			}
1372 			m->m_len = len = uimin(len, MCLBYTES);
1373 		} else {
1374 			/*
1375 			 * Place initial small packet/header at end of mbuf.
1376 			 */
1377 			if (len < m->m_len) {
1378 				if (top == 0 && len + max_linkhdr <= m->m_len)
1379 					m->m_data += max_linkhdr;
1380 				m->m_len = len;
1381 			} else
1382 				len = m->m_len;
1383 		}
1384 
1385 		memcpy(mtod(m, void *), cp, (size_t)len);
1386 
1387 		cp += len;
1388 		*mp = m;
1389 		mp = &m->m_next;
1390 		totlen -= len;
1391 		if (cp == epkt)
1392 			cp = buf;
1393 	}
1394 
1395 	return top;
1396 }
1397 
1398 /*
1399  * Copy data from a buffer back into the indicated mbuf chain,
1400  * starting "off" bytes from the beginning, extending the mbuf
1401  * chain if necessary.
1402  */
1403 void
1404 m_copyback(struct mbuf *m0, int off, int len, const void *cp)
1405 {
1406 #if defined(DEBUG)
1407 	struct mbuf *origm = m0;
1408 	int error;
1409 #endif
1410 
1411 	if (m0 == NULL)
1412 		return;
1413 
1414 #if defined(DEBUG)
1415 	error =
1416 #endif
1417 	m_copyback_internal(&m0, off, len, cp, CB_COPYBACK|CB_EXTEND,
1418 	    M_DONTWAIT);
1419 
1420 #if defined(DEBUG)
1421 	if (error != 0 || (m0 != NULL && origm != m0))
1422 		panic("m_copyback");
1423 #endif
1424 }
1425 
1426 struct mbuf *
1427 m_copyback_cow(struct mbuf *m0, int off, int len, const void *cp, int how)
1428 {
1429 	int error;
1430 
1431 	/* don't support chain expansion */
1432 	KASSERT(len != M_COPYALL);
1433 	KDASSERT(off + len <= m_length(m0));
1434 
1435 	error = m_copyback_internal(&m0, off, len, cp, CB_COPYBACK|CB_COW,
1436 	    how);
1437 	if (error) {
1438 		/*
1439 		 * no way to recover from partial success.
1440 		 * just free the chain.
1441 		 */
1442 		m_freem(m0);
1443 		return NULL;
1444 	}
1445 	return m0;
1446 }
1447 
1448 int
1449 m_makewritable(struct mbuf **mp, int off, int len, int how)
1450 {
1451 	int error;
1452 #if defined(DEBUG)
1453 	int origlen = m_length(*mp);
1454 #endif
1455 
1456 	error = m_copyback_internal(mp, off, len, NULL, CB_PRESERVE|CB_COW,
1457 	    how);
1458 	if (error)
1459 		return error;
1460 
1461 #if defined(DEBUG)
1462 	int reslen = 0;
1463 	for (struct mbuf *n = *mp; n; n = n->m_next)
1464 		reslen += n->m_len;
1465 	if (origlen != reslen)
1466 		panic("m_makewritable: length changed");
1467 	if (((*mp)->m_flags & M_PKTHDR) != 0 && reslen != (*mp)->m_pkthdr.len)
1468 		panic("m_makewritable: inconsist");
1469 #endif
1470 
1471 	return 0;
1472 }
1473 
1474 static int
1475 m_copyback_internal(struct mbuf **mp0, int off, int len, const void *vp,
1476     int flags, int how)
1477 {
1478 	int mlen;
1479 	struct mbuf *m, *n;
1480 	struct mbuf **mp;
1481 	int totlen = 0;
1482 	const char *cp = vp;
1483 
1484 	KASSERT(mp0 != NULL);
1485 	KASSERT(*mp0 != NULL);
1486 	KASSERT((flags & CB_PRESERVE) == 0 || cp == NULL);
1487 	KASSERT((flags & CB_COPYBACK) == 0 || cp != NULL);
1488 
1489 	if (len == M_COPYALL)
1490 		len = m_length(*mp0) - off;
1491 
1492 	/*
1493 	 * we don't bother to update "totlen" in the case of CB_COW,
1494 	 * assuming that CB_EXTEND and CB_COW are exclusive.
1495 	 */
1496 
1497 	KASSERT((~flags & (CB_EXTEND|CB_COW)) != 0);
1498 
1499 	mp = mp0;
1500 	m = *mp;
1501 	while (off > (mlen = m->m_len)) {
1502 		off -= mlen;
1503 		totlen += mlen;
1504 		if (m->m_next == NULL) {
1505 			int tspace;
1506 extend:
1507 			if ((flags & CB_EXTEND) == 0)
1508 				goto out;
1509 
1510 			/*
1511 			 * try to make some space at the end of "m".
1512 			 */
1513 
1514 			mlen = m->m_len;
1515 			if (off + len >= MINCLSIZE &&
1516 			    (m->m_flags & M_EXT) == 0 && m->m_len == 0) {
1517 				MCLGET(m, how);
1518 			}
1519 			tspace = M_TRAILINGSPACE(m);
1520 			if (tspace > 0) {
1521 				tspace = uimin(tspace, off + len);
1522 				KASSERT(tspace > 0);
1523 				memset(mtod(m, char *) + m->m_len, 0,
1524 				    uimin(off, tspace));
1525 				m->m_len += tspace;
1526 				off += mlen;
1527 				totlen -= mlen;
1528 				continue;
1529 			}
1530 
1531 			/*
1532 			 * need to allocate an mbuf.
1533 			 */
1534 
1535 			if (off + len >= MINCLSIZE) {
1536 				n = m_getcl(how, m->m_type, 0);
1537 			} else {
1538 				n = m_get(how, m->m_type);
1539 			}
1540 			if (n == NULL) {
1541 				goto out;
1542 			}
1543 			n->m_len = uimin(M_TRAILINGSPACE(n), off + len);
1544 			memset(mtod(n, char *), 0, uimin(n->m_len, off));
1545 			m->m_next = n;
1546 		}
1547 		mp = &m->m_next;
1548 		m = m->m_next;
1549 	}
1550 	while (len > 0) {
1551 		mlen = m->m_len - off;
1552 		if (mlen != 0 && M_READONLY(m)) {
1553 			/*
1554 			 * This mbuf is read-only. Allocate a new writable
1555 			 * mbuf and try again.
1556 			 */
1557 			char *datap;
1558 			int eatlen;
1559 
1560 			KASSERT((flags & CB_COW) != 0);
1561 
1562 			/*
1563 			 * if we're going to write into the middle of
1564 			 * a mbuf, split it first.
1565 			 */
1566 			if (off > 0) {
1567 				n = m_split_internal(m, off, how, false);
1568 				if (n == NULL)
1569 					goto enobufs;
1570 				m->m_next = n;
1571 				mp = &m->m_next;
1572 				m = n;
1573 				off = 0;
1574 				continue;
1575 			}
1576 
1577 			/*
1578 			 * XXX TODO coalesce into the trailingspace of
1579 			 * the previous mbuf when possible.
1580 			 */
1581 
1582 			/*
1583 			 * allocate a new mbuf.  copy packet header if needed.
1584 			 */
1585 			n = m_get(how, m->m_type);
1586 			if (n == NULL)
1587 				goto enobufs;
1588 			MCLAIM(n, m->m_owner);
1589 			if (off == 0 && (m->m_flags & M_PKTHDR) != 0) {
1590 				m_move_pkthdr(n, m);
1591 				n->m_len = MHLEN;
1592 			} else {
1593 				if (len >= MINCLSIZE)
1594 					MCLGET(n, M_DONTWAIT);
1595 				n->m_len =
1596 				    (n->m_flags & M_EXT) ? MCLBYTES : MLEN;
1597 			}
1598 			if (n->m_len > len)
1599 				n->m_len = len;
1600 
1601 			/*
1602 			 * free the region which has been overwritten.
1603 			 * copying data from old mbufs if requested.
1604 			 */
1605 			if (flags & CB_PRESERVE)
1606 				datap = mtod(n, char *);
1607 			else
1608 				datap = NULL;
1609 			eatlen = n->m_len;
1610 			while (m != NULL && M_READONLY(m) &&
1611 			    n->m_type == m->m_type && eatlen > 0) {
1612 				mlen = uimin(eatlen, m->m_len);
1613 				if (datap) {
1614 					m_copydata(m, 0, mlen, datap);
1615 					datap += mlen;
1616 				}
1617 				m->m_data += mlen;
1618 				m->m_len -= mlen;
1619 				eatlen -= mlen;
1620 				if (m->m_len == 0)
1621 					*mp = m = m_free(m);
1622 			}
1623 			if (eatlen > 0)
1624 				n->m_len -= eatlen;
1625 			n->m_next = m;
1626 			*mp = m = n;
1627 			continue;
1628 		}
1629 		mlen = uimin(mlen, len);
1630 		if (flags & CB_COPYBACK) {
1631 			memcpy(mtod(m, char *) + off, cp, (unsigned)mlen);
1632 			cp += mlen;
1633 		}
1634 		len -= mlen;
1635 		mlen += off;
1636 		off = 0;
1637 		totlen += mlen;
1638 		if (len == 0)
1639 			break;
1640 		if (m->m_next == NULL) {
1641 			goto extend;
1642 		}
1643 		mp = &m->m_next;
1644 		m = m->m_next;
1645 	}
1646 
1647 out:
1648 	if (((m = *mp0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) {
1649 		KASSERT((flags & CB_EXTEND) != 0);
1650 		m->m_pkthdr.len = totlen;
1651 	}
1652 
1653 	return 0;
1654 
1655 enobufs:
1656 	return ENOBUFS;
1657 }
1658 
1659 /*
1660  * Compress the mbuf chain. Return the new mbuf chain on success, NULL on
1661  * failure. The first mbuf is preserved, and on success the pointer returned
1662  * is the same as the one passed.
1663  */
1664 struct mbuf *
1665 m_defrag(struct mbuf *m, int how)
1666 {
1667 	struct mbuf *m0, *mn, *n;
1668 	int sz;
1669 
1670 	KASSERT((m->m_flags & M_PKTHDR) != 0);
1671 
1672 	if (m->m_next == NULL)
1673 		return m;
1674 
1675 	/* Defrag to single mbuf if at all possible */
1676 	if ((m->m_flags & M_EXT) == 0 && m->m_pkthdr.len <= MCLBYTES) {
1677 		if (m->m_pkthdr.len <= MHLEN) {
1678 			if (M_TRAILINGSPACE(m) < (m->m_pkthdr.len - m->m_len)) {
1679 				KASSERTMSG(M_LEADINGSPACE(m) +
1680 				    M_TRAILINGSPACE(m) >=
1681 				    (m->m_pkthdr.len - m->m_len),
1682 				    "too small leading %d trailing %d ro? %d"
1683 				    " pkthdr.len %d mlen %d",
1684 				    (int)M_LEADINGSPACE(m),
1685 				    (int)M_TRAILINGSPACE(m),
1686 				    M_READONLY(m),
1687 				    m->m_pkthdr.len, m->m_len);
1688 
1689 				memmove(m->m_pktdat, m->m_data, m->m_len);
1690 				m->m_data = m->m_pktdat;
1691 
1692 				KASSERT(M_TRAILINGSPACE(m) >=
1693 				    (m->m_pkthdr.len - m->m_len));
1694 			}
1695 		} else {
1696 			/* Must copy data before adding cluster */
1697 			m0 = m_get(how, MT_DATA);
1698 			if (m0 == NULL)
1699 				return NULL;
1700 			KASSERT(m->m_len <= MHLEN);
1701 			m_copydata(m, 0, m->m_len, mtod(m0, void *));
1702 
1703 			MCLGET(m, how);
1704 			if ((m->m_flags & M_EXT) == 0) {
1705 				m_free(m0);
1706 				return NULL;
1707 			}
1708 			memcpy(m->m_data, mtod(m0, void *), m->m_len);
1709 			m_free(m0);
1710 		}
1711 		KASSERT(M_TRAILINGSPACE(m) >= (m->m_pkthdr.len - m->m_len));
1712 		m_copydata(m->m_next, 0, m->m_pkthdr.len - m->m_len,
1713 			    mtod(m, char *) + m->m_len);
1714 		m->m_len = m->m_pkthdr.len;
1715 		m_freem(m->m_next);
1716 		m->m_next = NULL;
1717 		return m;
1718 	}
1719 
1720 	m0 = m_get(how, MT_DATA);
1721 	if (m0 == NULL)
1722 		return NULL;
1723 	mn = m0;
1724 
1725 	sz = m->m_pkthdr.len - m->m_len;
1726 	KASSERT(sz >= 0);
1727 
1728 	do {
1729 		if (sz > MLEN) {
1730 			MCLGET(mn, how);
1731 			if ((mn->m_flags & M_EXT) == 0) {
1732 				m_freem(m0);
1733 				return NULL;
1734 			}
1735 		}
1736 
1737 		mn->m_len = MIN(sz, MCLBYTES);
1738 
1739 		m_copydata(m, m->m_pkthdr.len - sz, mn->m_len,
1740 		     mtod(mn, void *));
1741 
1742 		sz -= mn->m_len;
1743 
1744 		if (sz > 0) {
1745 			/* need more mbufs */
1746 			n = m_get(how, MT_DATA);
1747 			if (n == NULL) {
1748 				m_freem(m0);
1749 				return NULL;
1750 			}
1751 
1752 			mn->m_next = n;
1753 			mn = n;
1754 		}
1755 	} while (sz > 0);
1756 
1757 	m_freem(m->m_next);
1758 	m->m_next = m0;
1759 
1760 	return m;
1761 }
1762 
1763 void
1764 m_remove_pkthdr(struct mbuf *m)
1765 {
1766 	KASSERT(m->m_flags & M_PKTHDR);
1767 
1768 	m_tag_delete_chain(m);
1769 	m->m_flags &= ~M_PKTHDR;
1770 	memset(&m->m_pkthdr, 0, sizeof(m->m_pkthdr));
1771 }
1772 
1773 void
1774 m_copy_pkthdr(struct mbuf *to, struct mbuf *from)
1775 {
1776 	KASSERT((to->m_flags & M_EXT) == 0);
1777 	KASSERT((to->m_flags & M_PKTHDR) == 0 ||
1778 	    SLIST_FIRST(&to->m_pkthdr.tags) == NULL);
1779 	KASSERT((from->m_flags & M_PKTHDR) != 0);
1780 
1781 	to->m_pkthdr = from->m_pkthdr;
1782 	to->m_flags = from->m_flags & M_COPYFLAGS;
1783 	to->m_data = to->m_pktdat;
1784 
1785 	SLIST_INIT(&to->m_pkthdr.tags);
1786 	m_tag_copy_chain(to, from);
1787 }
1788 
1789 void
1790 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
1791 {
1792 	KASSERT((to->m_flags & M_EXT) == 0);
1793 	KASSERT((to->m_flags & M_PKTHDR) == 0 ||
1794 	    SLIST_FIRST(&to->m_pkthdr.tags) == NULL);
1795 	KASSERT((from->m_flags & M_PKTHDR) != 0);
1796 
1797 	to->m_pkthdr = from->m_pkthdr;
1798 	to->m_flags = from->m_flags & M_COPYFLAGS;
1799 	to->m_data = to->m_pktdat;
1800 
1801 	from->m_flags &= ~M_PKTHDR;
1802 }
1803 
1804 /*
1805  * Set the m_data pointer of a newly-allocated mbuf to place an object of the
1806  * specified size at the end of the mbuf, longword aligned.
1807  */
1808 void
1809 m_align(struct mbuf *m, int len)
1810 {
1811 	int buflen, adjust;
1812 
1813 	KASSERT(len != M_COPYALL);
1814 	KASSERT(M_LEADINGSPACE(m) == 0);
1815 
1816 	buflen = M_BUFSIZE(m);
1817 
1818 	KASSERT(len <= buflen);
1819 	adjust = buflen - len;
1820 	m->m_data += adjust &~ (sizeof(long)-1);
1821 }
1822 
1823 /*
1824  * Apply function f to the data in an mbuf chain starting "off" bytes from the
1825  * beginning, continuing for "len" bytes.
1826  */
1827 int
1828 m_apply(struct mbuf *m, int off, int len,
1829     int (*f)(void *, void *, unsigned int), void *arg)
1830 {
1831 	unsigned int count;
1832 	int rval;
1833 
1834 	KASSERT(len != M_COPYALL);
1835 	KASSERT(len >= 0);
1836 	KASSERT(off >= 0);
1837 
1838 	while (off > 0) {
1839 		KASSERT(m != NULL);
1840 		if (off < m->m_len)
1841 			break;
1842 		off -= m->m_len;
1843 		m = m->m_next;
1844 	}
1845 	while (len > 0) {
1846 		KASSERT(m != NULL);
1847 		count = uimin(m->m_len - off, len);
1848 
1849 		rval = (*f)(arg, mtod(m, char *) + off, count);
1850 		if (rval)
1851 			return rval;
1852 
1853 		len -= count;
1854 		off = 0;
1855 		m = m->m_next;
1856 	}
1857 
1858 	return 0;
1859 }
1860 
1861 /*
1862  * Return a pointer to mbuf/offset of location in mbuf chain.
1863  */
1864 struct mbuf *
1865 m_getptr(struct mbuf *m, int loc, int *off)
1866 {
1867 
1868 	while (loc >= 0) {
1869 		/* Normal end of search */
1870 		if (m->m_len > loc) {
1871 			*off = loc;
1872 			return m;
1873 		}
1874 
1875 		loc -= m->m_len;
1876 
1877 		if (m->m_next == NULL) {
1878 			if (loc == 0) {
1879 				/* Point at the end of valid data */
1880 				*off = m->m_len;
1881 				return m;
1882 			}
1883 			return NULL;
1884 		} else {
1885 			m = m->m_next;
1886 		}
1887 	}
1888 
1889 	return NULL;
1890 }
1891 
1892 /*
1893  * Release a reference to the mbuf external storage.
1894  *
1895  * => free the mbuf m itself as well.
1896  */
1897 static void
1898 m_ext_free(struct mbuf *m)
1899 {
1900 	const bool embedded = MEXT_ISEMBEDDED(m);
1901 	bool dofree = true;
1902 	u_int refcnt;
1903 
1904 	KASSERT((m->m_flags & M_EXT) != 0);
1905 	KASSERT(MEXT_ISEMBEDDED(m->m_ext_ref));
1906 	KASSERT((m->m_ext_ref->m_flags & M_EXT) != 0);
1907 	KASSERT((m->m_flags & M_EXT_CLUSTER) ==
1908 	    (m->m_ext_ref->m_flags & M_EXT_CLUSTER));
1909 
1910 	if (__predict_false(m->m_type == MT_FREE)) {
1911 		panic("mbuf %p already freed", m);
1912 	}
1913 
1914 	if (__predict_true(m->m_ext.ext_refcnt == 1)) {
1915 		refcnt = m->m_ext.ext_refcnt = 0;
1916 	} else {
1917 #ifndef __HAVE_ATOMIC_AS_MEMBAR
1918 		membar_release();
1919 #endif
1920 		refcnt = atomic_dec_uint_nv(&m->m_ext.ext_refcnt);
1921 	}
1922 
1923 	if (refcnt > 0) {
1924 		if (embedded) {
1925 			/*
1926 			 * other mbuf's m_ext_ref still points to us.
1927 			 */
1928 			dofree = false;
1929 		} else {
1930 			m->m_ext_ref = m;
1931 		}
1932 	} else {
1933 		/*
1934 		 * dropping the last reference
1935 		 */
1936 #ifndef __HAVE_ATOMIC_AS_MEMBAR
1937 		membar_acquire();
1938 #endif
1939 		if (!embedded) {
1940 			m->m_ext.ext_refcnt++; /* XXX */
1941 			m_ext_free(m->m_ext_ref);
1942 			m->m_ext_ref = m;
1943 		} else if ((m->m_flags & M_EXT_CLUSTER) != 0) {
1944 			pool_cache_put_paddr(mcl_cache,
1945 			    m->m_ext.ext_buf, m->m_ext.ext_paddr);
1946 		} else if (m->m_ext.ext_free) {
1947 			(*m->m_ext.ext_free)(m,
1948 			    m->m_ext.ext_buf, m->m_ext.ext_size,
1949 			    m->m_ext.ext_arg);
1950 			/*
1951 			 * 'm' is already freed by the ext_free callback.
1952 			 */
1953 			dofree = false;
1954 		} else {
1955 			free(m->m_ext.ext_buf, 0);
1956 		}
1957 	}
1958 
1959 	if (dofree) {
1960 		m->m_type = MT_FREE;
1961 		m->m_data = NULL;
1962 		pool_cache_put(mb_cache, m);
1963 	}
1964 }
1965 
1966 /*
1967  * Free a single mbuf and associated external storage. Return the
1968  * successor, if any.
1969  */
1970 struct mbuf *
1971 m_free(struct mbuf *m)
1972 {
1973 	struct mbuf *n;
1974 
1975 	mowner_revoke(m, 1, m->m_flags);
1976 	mbstat_type_add(m->m_type, -1);
1977 
1978 	if (m->m_flags & M_PKTHDR)
1979 		m_tag_delete_chain(m);
1980 
1981 	n = m->m_next;
1982 
1983 	if (m->m_flags & M_EXT) {
1984 		m_ext_free(m);
1985 	} else {
1986 		if (__predict_false(m->m_type == MT_FREE)) {
1987 			panic("mbuf %p already freed", m);
1988 		}
1989 		m->m_type = MT_FREE;
1990 		m->m_data = NULL;
1991 		pool_cache_put(mb_cache, m);
1992 	}
1993 
1994 	return n;
1995 }
1996 
1997 void
1998 m_freem(struct mbuf *m)
1999 {
2000 	if (m == NULL)
2001 		return;
2002 	do {
2003 		m = m_free(m);
2004 	} while (m);
2005 }
2006 
2007 #if defined(DDB)
2008 void
2009 m_print(const struct mbuf *m, const char *modif, void (*pr)(const char *, ...))
2010 {
2011 	char ch;
2012 	bool opt_c = false;
2013 	bool opt_d = false;
2014 #if NETHER > 0
2015 	bool opt_v = false;
2016 	const struct mbuf *m0 = NULL;
2017 #endif
2018 	int no = 0;
2019 	char buf[512];
2020 
2021 	while ((ch = *(modif++)) != '\0') {
2022 		switch (ch) {
2023 		case 'c':
2024 			opt_c = true;
2025 			break;
2026 		case 'd':
2027 			opt_d = true;
2028 			break;
2029 #if NETHER > 0
2030 		case 'v':
2031 			opt_v = true;
2032 			m0 = m;
2033 			break;
2034 #endif
2035 		default:
2036 			break;
2037 		}
2038 	}
2039 
2040 nextchain:
2041 	(*pr)("MBUF(%d) %p\n", no, m);
2042 	snprintb(buf, sizeof(buf), M_FLAGS_BITS, (u_int)m->m_flags);
2043 	(*pr)("  data=%p, len=%d, type=%d, flags=%s\n",
2044 	    m->m_data, m->m_len, m->m_type, buf);
2045 	if (opt_d) {
2046 		int i;
2047 		unsigned char *p = m->m_data;
2048 
2049 		(*pr)("  data:");
2050 
2051 		for (i = 0; i < m->m_len; i++) {
2052 			if (i % 16 == 0)
2053 				(*pr)("\n");
2054 			(*pr)(" %02x", p[i]);
2055 		}
2056 
2057 		(*pr)("\n");
2058 	}
2059 	(*pr)("  owner=%p, next=%p, nextpkt=%p\n", m->m_owner, m->m_next,
2060 	    m->m_nextpkt);
2061 	(*pr)("  leadingspace=%u, trailingspace=%u, readonly=%u\n",
2062 	    (int)M_LEADINGSPACE(m), (int)M_TRAILINGSPACE(m),
2063 	    (int)M_READONLY(m));
2064 	if ((m->m_flags & M_PKTHDR) != 0) {
2065 		snprintb(buf, sizeof(buf), M_CSUM_BITS, m->m_pkthdr.csum_flags);
2066 		(*pr)("  pktlen=%d, rcvif=%p, csum_flags=%s, csum_data=0x%"
2067 		    PRIx32 ", segsz=%u\n",
2068 		    m->m_pkthdr.len, m_get_rcvif_NOMPSAFE(m),
2069 		    buf, m->m_pkthdr.csum_data, m->m_pkthdr.segsz);
2070 	}
2071 	if ((m->m_flags & M_EXT)) {
2072 		(*pr)("  ext_refcnt=%u, ext_buf=%p, ext_size=%zd, "
2073 		    "ext_free=%p, ext_arg=%p\n",
2074 		    m->m_ext.ext_refcnt,
2075 		    m->m_ext.ext_buf, m->m_ext.ext_size,
2076 		    m->m_ext.ext_free, m->m_ext.ext_arg);
2077 	}
2078 	if ((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0) {
2079 		vaddr_t sva = (vaddr_t)m->m_ext.ext_buf;
2080 		vaddr_t eva = sva + m->m_ext.ext_size;
2081 		int n = (round_page(eva) - trunc_page(sva)) >> PAGE_SHIFT;
2082 		int i;
2083 
2084 		(*pr)("  pages:");
2085 		for (i = 0; i < n; i ++) {
2086 			(*pr)(" %p", m->m_ext.ext_pgs[i]);
2087 		}
2088 		(*pr)("\n");
2089 	}
2090 
2091 	if (opt_c) {
2092 		m = m->m_next;
2093 		if (m != NULL) {
2094 			no++;
2095 			goto nextchain;
2096 		}
2097 	}
2098 
2099 #if NETHER > 0
2100 	if (opt_v && m0)
2101 		m_examine(m0, AF_ETHER, modif, pr);
2102 #endif
2103 }
2104 #endif /* defined(DDB) */
2105 
2106 #if defined(MBUFTRACE)
2107 void
2108 mowner_init_owner(struct mowner *mo, const char *name, const char *descr)
2109 {
2110 	memset(mo, 0, sizeof(*mo));
2111 	strlcpy(mo->mo_name, name, sizeof(mo->mo_name));
2112 	strlcpy(mo->mo_descr, descr, sizeof(mo->mo_descr));
2113 }
2114 
2115 void
2116 mowner_attach(struct mowner *mo)
2117 {
2118 
2119 	KASSERT(mo->mo_counters == NULL);
2120 	mo->mo_counters = percpu_alloc(sizeof(struct mowner_counter));
2121 
2122 	/* XXX lock */
2123 	LIST_INSERT_HEAD(&mowners, mo, mo_link);
2124 }
2125 
2126 void
2127 mowner_detach(struct mowner *mo)
2128 {
2129 
2130 	KASSERT(mo->mo_counters != NULL);
2131 
2132 	/* XXX lock */
2133 	LIST_REMOVE(mo, mo_link);
2134 
2135 	percpu_free(mo->mo_counters, sizeof(struct mowner_counter));
2136 	mo->mo_counters = NULL;
2137 }
2138 
2139 void
2140 mowner_init(struct mbuf *m, int type)
2141 {
2142 	struct mowner_counter *mc;
2143 	struct mowner *mo;
2144 	int s;
2145 
2146 	m->m_owner = mo = &unknown_mowners[type];
2147 	s = splvm();
2148 	mc = percpu_getref(mo->mo_counters);
2149 	mc->mc_counter[MOWNER_COUNTER_CLAIMS]++;
2150 	percpu_putref(mo->mo_counters);
2151 	splx(s);
2152 }
2153 
2154 void
2155 mowner_ref(struct mbuf *m, int flags)
2156 {
2157 	struct mowner *mo = m->m_owner;
2158 	struct mowner_counter *mc;
2159 	int s;
2160 
2161 	s = splvm();
2162 	mc = percpu_getref(mo->mo_counters);
2163 	if ((flags & M_EXT) != 0)
2164 		mc->mc_counter[MOWNER_COUNTER_EXT_CLAIMS]++;
2165 	if ((flags & M_EXT_CLUSTER) != 0)
2166 		mc->mc_counter[MOWNER_COUNTER_CLUSTER_CLAIMS]++;
2167 	percpu_putref(mo->mo_counters);
2168 	splx(s);
2169 }
2170 
2171 void
2172 mowner_revoke(struct mbuf *m, bool all, int flags)
2173 {
2174 	struct mowner *mo = m->m_owner;
2175 	struct mowner_counter *mc;
2176 	int s;
2177 
2178 	s = splvm();
2179 	mc = percpu_getref(mo->mo_counters);
2180 	if ((flags & M_EXT) != 0)
2181 		mc->mc_counter[MOWNER_COUNTER_EXT_RELEASES]++;
2182 	if ((flags & M_EXT_CLUSTER) != 0)
2183 		mc->mc_counter[MOWNER_COUNTER_CLUSTER_RELEASES]++;
2184 	if (all)
2185 		mc->mc_counter[MOWNER_COUNTER_RELEASES]++;
2186 	percpu_putref(mo->mo_counters);
2187 	splx(s);
2188 	if (all)
2189 		m->m_owner = &revoked_mowner;
2190 }
2191 
2192 static void
2193 mowner_claim(struct mbuf *m, struct mowner *mo)
2194 {
2195 	struct mowner_counter *mc;
2196 	int flags = m->m_flags;
2197 	int s;
2198 
2199 	s = splvm();
2200 	mc = percpu_getref(mo->mo_counters);
2201 	mc->mc_counter[MOWNER_COUNTER_CLAIMS]++;
2202 	if ((flags & M_EXT) != 0)
2203 		mc->mc_counter[MOWNER_COUNTER_EXT_CLAIMS]++;
2204 	if ((flags & M_EXT_CLUSTER) != 0)
2205 		mc->mc_counter[MOWNER_COUNTER_CLUSTER_CLAIMS]++;
2206 	percpu_putref(mo->mo_counters);
2207 	splx(s);
2208 	m->m_owner = mo;
2209 }
2210 
2211 void
2212 m_claim(struct mbuf *m, struct mowner *mo)
2213 {
2214 
2215 	if (m->m_owner == mo || mo == NULL)
2216 		return;
2217 
2218 	mowner_revoke(m, true, m->m_flags);
2219 	mowner_claim(m, mo);
2220 }
2221 
2222 void
2223 m_claimm(struct mbuf *m, struct mowner *mo)
2224 {
2225 
2226 	for (; m != NULL; m = m->m_next)
2227 		m_claim(m, mo);
2228 }
2229 #endif /* defined(MBUFTRACE) */
2230 
2231 #ifdef DIAGNOSTIC
2232 /*
2233  * Verify that the mbuf chain is not malformed. Used only for diagnostic.
2234  * Panics on error.
2235  */
2236 void
2237 m_verify_packet(struct mbuf *m)
2238 {
2239 	struct mbuf *n = m;
2240 	char *low, *high, *dat;
2241 	int totlen = 0, len;
2242 
2243 	if (__predict_false((m->m_flags & M_PKTHDR) == 0)) {
2244 		panic("%s: mbuf doesn't have M_PKTHDR", __func__);
2245 	}
2246 
2247 	while (n != NULL) {
2248 		if (__predict_false(n->m_type == MT_FREE)) {
2249 			panic("%s: mbuf already freed (n = %p)", __func__, n);
2250 		}
2251 #if 0
2252 		/*
2253 		 * This ought to be a rule of the mbuf API. Unfortunately,
2254 		 * many places don't respect that rule.
2255 		 */
2256 		if (__predict_false((n != m) && (n->m_flags & M_PKTHDR) != 0)) {
2257 			panic("%s: M_PKTHDR set on secondary mbuf", __func__);
2258 		}
2259 #endif
2260 		if (__predict_false(n->m_nextpkt != NULL)) {
2261 			panic("%s: m_nextpkt not null (m_nextpkt = %p)",
2262 			    __func__, n->m_nextpkt);
2263 		}
2264 
2265 		dat = n->m_data;
2266 		len = n->m_len;
2267 		if (__predict_false(len < 0)) {
2268 			panic("%s: incorrect length (len = %d)", __func__, len);
2269 		}
2270 
2271 		low = M_BUFADDR(n);
2272 		high = low + M_BUFSIZE(n);
2273 		if (__predict_false((dat < low) || (dat + len > high))) {
2274 			panic("%s: m_data not in packet"
2275 			    "(dat = %p, len = %d, low = %p, high = %p)",
2276 			    __func__, dat, len, low, high);
2277 		}
2278 
2279 		totlen += len;
2280 		n = n->m_next;
2281 	}
2282 
2283 	if (__predict_false(totlen != m->m_pkthdr.len)) {
2284 		panic("%s: inconsistent mbuf length (%d != %d)", __func__,
2285 		    totlen, m->m_pkthdr.len);
2286 	}
2287 }
2288 #endif
2289 
2290 struct m_tag *
2291 m_tag_get(int type, int len, int wait)
2292 {
2293 	struct m_tag *t;
2294 
2295 	if (len < 0)
2296 		return NULL;
2297 	t = malloc(len + sizeof(struct m_tag), M_PACKET_TAGS, wait);
2298 	if (t == NULL)
2299 		return NULL;
2300 	t->m_tag_id = type;
2301 	t->m_tag_len = len;
2302 	return t;
2303 }
2304 
2305 void
2306 m_tag_free(struct m_tag *t)
2307 {
2308 	free(t, M_PACKET_TAGS);
2309 }
2310 
2311 void
2312 m_tag_prepend(struct mbuf *m, struct m_tag *t)
2313 {
2314 	KASSERT((m->m_flags & M_PKTHDR) != 0);
2315 	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
2316 }
2317 
2318 void
2319 m_tag_unlink(struct mbuf *m, struct m_tag *t)
2320 {
2321 	KASSERT((m->m_flags & M_PKTHDR) != 0);
2322 	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
2323 }
2324 
2325 void
2326 m_tag_delete(struct mbuf *m, struct m_tag *t)
2327 {
2328 	m_tag_unlink(m, t);
2329 	m_tag_free(t);
2330 }
2331 
2332 void
2333 m_tag_delete_chain(struct mbuf *m)
2334 {
2335 	struct m_tag *p, *q;
2336 
2337 	KASSERT((m->m_flags & M_PKTHDR) != 0);
2338 
2339 	p = SLIST_FIRST(&m->m_pkthdr.tags);
2340 	if (p == NULL)
2341 		return;
2342 	while ((q = SLIST_NEXT(p, m_tag_link)) != NULL)
2343 		m_tag_delete(m, q);
2344 	m_tag_delete(m, p);
2345 }
2346 
2347 struct m_tag *
2348 m_tag_find(const struct mbuf *m, int type)
2349 {
2350 	struct m_tag *p;
2351 
2352 	KASSERT((m->m_flags & M_PKTHDR) != 0);
2353 
2354 	p = SLIST_FIRST(&m->m_pkthdr.tags);
2355 	while (p != NULL) {
2356 		if (p->m_tag_id == type)
2357 			return p;
2358 		p = SLIST_NEXT(p, m_tag_link);
2359 	}
2360 	return NULL;
2361 }
2362 
2363 struct m_tag *
2364 m_tag_copy(struct m_tag *t)
2365 {
2366 	struct m_tag *p;
2367 
2368 	p = m_tag_get(t->m_tag_id, t->m_tag_len, M_NOWAIT);
2369 	if (p == NULL)
2370 		return NULL;
2371 	memcpy(p + 1, t + 1, t->m_tag_len);
2372 	return p;
2373 }
2374 
2375 /*
2376  * Copy two tag chains. The destination mbuf (to) loses any attached
2377  * tags even if the operation fails. This should not be a problem, as
2378  * m_tag_copy_chain() is typically called with a newly-allocated
2379  * destination mbuf.
2380  */
2381 int
2382 m_tag_copy_chain(struct mbuf *to, struct mbuf *from)
2383 {
2384 	struct m_tag *p, *t, *tprev = NULL;
2385 
2386 	KASSERT((from->m_flags & M_PKTHDR) != 0);
2387 
2388 	m_tag_delete_chain(to);
2389 	SLIST_FOREACH(p, &from->m_pkthdr.tags, m_tag_link) {
2390 		t = m_tag_copy(p);
2391 		if (t == NULL) {
2392 			m_tag_delete_chain(to);
2393 			return 0;
2394 		}
2395 		if (tprev == NULL)
2396 			SLIST_INSERT_HEAD(&to->m_pkthdr.tags, t, m_tag_link);
2397 		else
2398 			SLIST_INSERT_AFTER(tprev, t, m_tag_link);
2399 		tprev = t;
2400 	}
2401 	return 1;
2402 }
2403