xref: /freebsd/sys/kern/uipc_mbuf.c (revision 9e5787d2284e187abb5b654d924394a65772e004)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)uipc_mbuf.c	8.2 (Berkeley) 1/4/94
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_param.h"
38 #include "opt_mbuf_stress_test.h"
39 #include "opt_mbuf_profiling.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/limits.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/sysctl.h>
49 #include <sys/domain.h>
50 #include <sys/protosw.h>
51 #include <sys/uio.h>
52 #include <sys/vmmeter.h>
53 #include <sys/sdt.h>
54 #include <vm/vm.h>
55 #include <vm/vm_pageout.h>
56 #include <vm/vm_page.h>
57 
58 SDT_PROBE_DEFINE5_XLATE(sdt, , , m__init,
59     "struct mbuf *", "mbufinfo_t *",
60     "uint32_t", "uint32_t",
61     "uint16_t", "uint16_t",
62     "uint32_t", "uint32_t",
63     "uint32_t", "uint32_t");
64 
65 SDT_PROBE_DEFINE3_XLATE(sdt, , , m__gethdr,
66     "uint32_t", "uint32_t",
67     "uint16_t", "uint16_t",
68     "struct mbuf *", "mbufinfo_t *");
69 
70 SDT_PROBE_DEFINE3_XLATE(sdt, , , m__get,
71     "uint32_t", "uint32_t",
72     "uint16_t", "uint16_t",
73     "struct mbuf *", "mbufinfo_t *");
74 
75 SDT_PROBE_DEFINE4_XLATE(sdt, , , m__getcl,
76     "uint32_t", "uint32_t",
77     "uint16_t", "uint16_t",
78     "uint32_t", "uint32_t",
79     "struct mbuf *", "mbufinfo_t *");
80 
81 SDT_PROBE_DEFINE5_XLATE(sdt, , , m__getjcl,
82     "uint32_t", "uint32_t",
83     "uint16_t", "uint16_t",
84     "uint32_t", "uint32_t",
85     "uint32_t", "uint32_t",
86     "struct mbuf *", "mbufinfo_t *");
87 
88 SDT_PROBE_DEFINE3_XLATE(sdt, , , m__clget,
89     "struct mbuf *", "mbufinfo_t *",
90     "uint32_t", "uint32_t",
91     "uint32_t", "uint32_t");
92 
93 SDT_PROBE_DEFINE4_XLATE(sdt, , , m__cljget,
94     "struct mbuf *", "mbufinfo_t *",
95     "uint32_t", "uint32_t",
96     "uint32_t", "uint32_t",
97     "void*", "void*");
98 
99 SDT_PROBE_DEFINE(sdt, , , m__cljset);
100 
101 SDT_PROBE_DEFINE1_XLATE(sdt, , , m__free,
102         "struct mbuf *", "mbufinfo_t *");
103 
104 SDT_PROBE_DEFINE1_XLATE(sdt, , , m__freem,
105     "struct mbuf *", "mbufinfo_t *");
106 
107 #include <security/mac/mac_framework.h>
108 
109 int	max_linkhdr;
110 int	max_protohdr;
111 int	max_hdr;
112 int	max_datalen;
113 #ifdef MBUF_STRESS_TEST
114 int	m_defragpackets;
115 int	m_defragbytes;
116 int	m_defraguseless;
117 int	m_defragfailure;
118 int	m_defragrandomfailures;
119 #endif
120 
121 /*
122  * sysctl(8) exported objects
123  */
124 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD,
125 	   &max_linkhdr, 0, "Size of largest link layer header");
126 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD,
127 	   &max_protohdr, 0, "Size of largest protocol layer header");
128 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD,
129 	   &max_hdr, 0, "Size of largest link plus protocol header");
130 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RD,
131 	   &max_datalen, 0, "Minimum space left in mbuf after max_hdr");
132 #ifdef MBUF_STRESS_TEST
133 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
134 	   &m_defragpackets, 0, "");
135 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
136 	   &m_defragbytes, 0, "");
137 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
138 	   &m_defraguseless, 0, "");
139 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
140 	   &m_defragfailure, 0, "");
141 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
142 	   &m_defragrandomfailures, 0, "");
143 #endif
144 
145 /*
146  * Ensure the correct size of various mbuf parameters.  It could be off due
147  * to compiler-induced padding and alignment artifacts.
148  */
149 CTASSERT(MSIZE - offsetof(struct mbuf, m_dat) == MLEN);
150 CTASSERT(MSIZE - offsetof(struct mbuf, m_pktdat) == MHLEN);
151 
152 /*
153  * mbuf data storage should be 64-bit aligned regardless of architectural
154  * pointer size; check this is the case with and without a packet header.
155  */
156 CTASSERT(offsetof(struct mbuf, m_dat) % 8 == 0);
157 CTASSERT(offsetof(struct mbuf, m_pktdat) % 8 == 0);
158 
159 /*
160  * While the specific values here don't matter too much (i.e., +/- a few
161  * words), we do want to ensure that changes to these values are carefully
162  * reasoned about and properly documented.  This is especially the case as
163  * network-protocol and device-driver modules encode these layouts, and must
164  * be recompiled if the structures change.  Check these values at compile time
165  * against the ones documented in comments in mbuf.h.
166  *
167  * NB: Possibly they should be documented there via #define's and not just
168  * comments.
169  */
170 #if defined(__LP64__)
171 CTASSERT(offsetof(struct mbuf, m_dat) == 32);
172 CTASSERT(sizeof(struct pkthdr) == 56);
173 CTASSERT(sizeof(struct m_ext) == 160);
174 #else
175 CTASSERT(offsetof(struct mbuf, m_dat) == 24);
176 CTASSERT(sizeof(struct pkthdr) == 48);
177 #if defined(__powerpc__) && defined(BOOKE)
178 /* PowerPC booke has 64-bit physical pointers. */
179 CTASSERT(sizeof(struct m_ext) == 184);
180 #else
181 CTASSERT(sizeof(struct m_ext) == 180);
182 #endif
183 #endif
184 
185 /*
186  * Assert that the queue(3) macros produce code of the same size as an old
187  * plain pointer does.
188  */
189 #ifdef INVARIANTS
190 static struct mbuf __used m_assertbuf;
191 CTASSERT(sizeof(m_assertbuf.m_slist) == sizeof(m_assertbuf.m_next));
192 CTASSERT(sizeof(m_assertbuf.m_stailq) == sizeof(m_assertbuf.m_next));
193 CTASSERT(sizeof(m_assertbuf.m_slistpkt) == sizeof(m_assertbuf.m_nextpkt));
194 CTASSERT(sizeof(m_assertbuf.m_stailqpkt) == sizeof(m_assertbuf.m_nextpkt));
195 #endif
196 
197 /*
198  * Attach the cluster from *m to *n, set up m_ext in *n
199  * and bump the refcount of the cluster.
200  */
201 void
202 mb_dupcl(struct mbuf *n, struct mbuf *m)
203 {
204 	volatile u_int *refcnt;
205 
206 	KASSERT(m->m_flags & (M_EXT|M_EXTPG),
207 	    ("%s: M_EXT|M_EXTPG not set on %p", __func__, m));
208 	KASSERT(!(n->m_flags & (M_EXT|M_EXTPG)),
209 	    ("%s: M_EXT|M_EXTPG set on %p", __func__, n));
210 
211 	/*
212 	 * Cache access optimization.
213 	 *
214 	 * o Regular M_EXT storage doesn't need full copy of m_ext, since
215 	 *   the holder of the 'ext_count' is responsible to carry the free
216 	 *   routine and its arguments.
217 	 * o M_EXTPG data is split between main part of mbuf and m_ext, the
218 	 *   main part is copied in full, the m_ext part is similar to M_EXT.
219 	 * o EXT_EXTREF, where 'ext_cnt' doesn't point into mbuf at all, is
220 	 *   special - it needs full copy of m_ext into each mbuf, since any
221 	 *   copy could end up as the last to free.
222 	 */
223 	if (m->m_flags & M_EXTPG) {
224 		bcopy(&m->m_epg_startcopy, &n->m_epg_startcopy,
225 		    __rangeof(struct mbuf, m_epg_startcopy, m_epg_endcopy));
226 		bcopy(&m->m_ext, &n->m_ext, m_epg_ext_copylen);
227 	} else if (m->m_ext.ext_type == EXT_EXTREF)
228 		bcopy(&m->m_ext, &n->m_ext, sizeof(struct m_ext));
229 	else
230 		bcopy(&m->m_ext, &n->m_ext, m_ext_copylen);
231 
232 	n->m_flags |= m->m_flags & (M_RDONLY | M_EXT | M_EXTPG);
233 
234 	/* See if this is the mbuf that holds the embedded refcount. */
235 	if (m->m_ext.ext_flags & EXT_FLAG_EMBREF) {
236 		refcnt = n->m_ext.ext_cnt = &m->m_ext.ext_count;
237 		n->m_ext.ext_flags &= ~EXT_FLAG_EMBREF;
238 	} else {
239 		KASSERT(m->m_ext.ext_cnt != NULL,
240 		    ("%s: no refcounting pointer on %p", __func__, m));
241 		refcnt = m->m_ext.ext_cnt;
242 	}
243 
244 	if (*refcnt == 1)
245 		*refcnt += 1;
246 	else
247 		atomic_add_int(refcnt, 1);
248 }
249 
250 void
251 m_demote_pkthdr(struct mbuf *m)
252 {
253 
254 	M_ASSERTPKTHDR(m);
255 
256 	m_tag_delete_chain(m, NULL);
257 	m->m_flags &= ~M_PKTHDR;
258 	bzero(&m->m_pkthdr, sizeof(struct pkthdr));
259 }
260 
261 /*
262  * Clean up mbuf (chain) from any tags and packet headers.
263  * If "all" is set then the first mbuf in the chain will be
264  * cleaned too.
265  */
266 void
267 m_demote(struct mbuf *m0, int all, int flags)
268 {
269 	struct mbuf *m;
270 
271 	for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) {
272 		KASSERT(m->m_nextpkt == NULL, ("%s: m_nextpkt in m %p, m0 %p",
273 		    __func__, m, m0));
274 		if (m->m_flags & M_PKTHDR)
275 			m_demote_pkthdr(m);
276 		m->m_flags = m->m_flags & (M_EXT | M_RDONLY | M_NOFREE |
277 		    M_EXTPG | flags);
278 	}
279 }
280 
281 /*
282  * Sanity checks on mbuf (chain) for use in KASSERT() and general
283  * debugging.
284  * Returns 0 or panics when bad and 1 on all tests passed.
285  * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they
286  * blow up later.
287  */
288 int
289 m_sanity(struct mbuf *m0, int sanitize)
290 {
291 	struct mbuf *m;
292 	caddr_t a, b;
293 	int pktlen = 0;
294 
295 #ifdef INVARIANTS
296 #define	M_SANITY_ACTION(s)	panic("mbuf %p: " s, m)
297 #else
298 #define	M_SANITY_ACTION(s)	printf("mbuf %p: " s, m)
299 #endif
300 
301 	for (m = m0; m != NULL; m = m->m_next) {
302 		/*
303 		 * Basic pointer checks.  If any of these fails then some
304 		 * unrelated kernel memory before or after us is trashed.
305 		 * No way to recover from that.
306 		 */
307 		a = M_START(m);
308 		b = a + M_SIZE(m);
309 		if ((caddr_t)m->m_data < a)
310 			M_SANITY_ACTION("m_data outside mbuf data range left");
311 		if ((caddr_t)m->m_data > b)
312 			M_SANITY_ACTION("m_data outside mbuf data range right");
313 		if ((caddr_t)m->m_data + m->m_len > b)
314 			M_SANITY_ACTION("m_data + m_len exeeds mbuf space");
315 
316 		/* m->m_nextpkt may only be set on first mbuf in chain. */
317 		if (m != m0 && m->m_nextpkt != NULL) {
318 			if (sanitize) {
319 				m_freem(m->m_nextpkt);
320 				m->m_nextpkt = (struct mbuf *)0xDEADC0DE;
321 			} else
322 				M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf");
323 		}
324 
325 		/* packet length (not mbuf length!) calculation */
326 		if (m0->m_flags & M_PKTHDR)
327 			pktlen += m->m_len;
328 
329 		/* m_tags may only be attached to first mbuf in chain. */
330 		if (m != m0 && m->m_flags & M_PKTHDR &&
331 		    !SLIST_EMPTY(&m->m_pkthdr.tags)) {
332 			if (sanitize) {
333 				m_tag_delete_chain(m, NULL);
334 				/* put in 0xDEADC0DE perhaps? */
335 			} else
336 				M_SANITY_ACTION("m_tags on in-chain mbuf");
337 		}
338 
339 		/* M_PKTHDR may only be set on first mbuf in chain */
340 		if (m != m0 && m->m_flags & M_PKTHDR) {
341 			if (sanitize) {
342 				bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
343 				m->m_flags &= ~M_PKTHDR;
344 				/* put in 0xDEADCODE and leave hdr flag in */
345 			} else
346 				M_SANITY_ACTION("M_PKTHDR on in-chain mbuf");
347 		}
348 	}
349 	m = m0;
350 	if (pktlen && pktlen != m->m_pkthdr.len) {
351 		if (sanitize)
352 			m->m_pkthdr.len = 0;
353 		else
354 			M_SANITY_ACTION("m_pkthdr.len != mbuf chain length");
355 	}
356 	return 1;
357 
358 #undef	M_SANITY_ACTION
359 }
360 
361 /*
362  * Non-inlined part of m_init().
363  */
364 int
365 m_pkthdr_init(struct mbuf *m, int how)
366 {
367 #ifdef MAC
368 	int error;
369 #endif
370 	m->m_data = m->m_pktdat;
371 	bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
372 #ifdef NUMA
373 	m->m_pkthdr.numa_domain = M_NODOM;
374 #endif
375 #ifdef MAC
376 	/* If the label init fails, fail the alloc */
377 	error = mac_mbuf_init(m, how);
378 	if (error)
379 		return (error);
380 #endif
381 
382 	return (0);
383 }
384 
385 /*
386  * "Move" mbuf pkthdr from "from" to "to".
387  * "from" must have M_PKTHDR set, and "to" must be empty.
388  */
389 void
390 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
391 {
392 
393 #if 0
394 	/* see below for why these are not enabled */
395 	M_ASSERTPKTHDR(to);
396 	/* Note: with MAC, this may not be a good assertion. */
397 	KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
398 	    ("m_move_pkthdr: to has tags"));
399 #endif
400 #ifdef MAC
401 	/*
402 	 * XXXMAC: It could be this should also occur for non-MAC?
403 	 */
404 	if (to->m_flags & M_PKTHDR)
405 		m_tag_delete_chain(to, NULL);
406 #endif
407 	to->m_flags = (from->m_flags & M_COPYFLAGS) |
408 	    (to->m_flags & (M_EXT | M_EXTPG));
409 	if ((to->m_flags & M_EXT) == 0)
410 		to->m_data = to->m_pktdat;
411 	to->m_pkthdr = from->m_pkthdr;		/* especially tags */
412 	SLIST_INIT(&from->m_pkthdr.tags);	/* purge tags from src */
413 	from->m_flags &= ~M_PKTHDR;
414 	if (from->m_pkthdr.csum_flags & CSUM_SND_TAG) {
415 		from->m_pkthdr.csum_flags &= ~CSUM_SND_TAG;
416 		from->m_pkthdr.snd_tag = NULL;
417 	}
418 }
419 
420 /*
421  * Duplicate "from"'s mbuf pkthdr in "to".
422  * "from" must have M_PKTHDR set, and "to" must be empty.
423  * In particular, this does a deep copy of the packet tags.
424  */
425 int
426 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how)
427 {
428 
429 #if 0
430 	/*
431 	 * The mbuf allocator only initializes the pkthdr
432 	 * when the mbuf is allocated with m_gethdr(). Many users
433 	 * (e.g. m_copy*, m_prepend) use m_get() and then
434 	 * smash the pkthdr as needed causing these
435 	 * assertions to trip.  For now just disable them.
436 	 */
437 	M_ASSERTPKTHDR(to);
438 	/* Note: with MAC, this may not be a good assertion. */
439 	KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
440 #endif
441 	MBUF_CHECKSLEEP(how);
442 #ifdef MAC
443 	if (to->m_flags & M_PKTHDR)
444 		m_tag_delete_chain(to, NULL);
445 #endif
446 	to->m_flags = (from->m_flags & M_COPYFLAGS) |
447 	    (to->m_flags & (M_EXT | M_EXTPG));
448 	if ((to->m_flags & M_EXT) == 0)
449 		to->m_data = to->m_pktdat;
450 	to->m_pkthdr = from->m_pkthdr;
451 	if (from->m_pkthdr.csum_flags & CSUM_SND_TAG)
452 		m_snd_tag_ref(from->m_pkthdr.snd_tag);
453 	SLIST_INIT(&to->m_pkthdr.tags);
454 	return (m_tag_copy_chain(to, from, how));
455 }
456 
457 /*
458  * Lesser-used path for M_PREPEND:
459  * allocate new mbuf to prepend to chain,
460  * copy junk along.
461  */
462 struct mbuf *
463 m_prepend(struct mbuf *m, int len, int how)
464 {
465 	struct mbuf *mn;
466 
467 	if (m->m_flags & M_PKTHDR)
468 		mn = m_gethdr(how, m->m_type);
469 	else
470 		mn = m_get(how, m->m_type);
471 	if (mn == NULL) {
472 		m_freem(m);
473 		return (NULL);
474 	}
475 	if (m->m_flags & M_PKTHDR)
476 		m_move_pkthdr(mn, m);
477 	mn->m_next = m;
478 	m = mn;
479 	if (len < M_SIZE(m))
480 		M_ALIGN(m, len);
481 	m->m_len = len;
482 	return (m);
483 }
484 
485 /*
486  * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
487  * continuing for "len" bytes.  If len is M_COPYALL, copy to end of mbuf.
488  * The wait parameter is a choice of M_WAITOK/M_NOWAIT from caller.
489  * Note that the copy is read-only, because clusters are not copied,
490  * only their reference counts are incremented.
491  */
492 struct mbuf *
493 m_copym(struct mbuf *m, int off0, int len, int wait)
494 {
495 	struct mbuf *n, **np;
496 	int off = off0;
497 	struct mbuf *top;
498 	int copyhdr = 0;
499 
500 	KASSERT(off >= 0, ("m_copym, negative off %d", off));
501 	KASSERT(len >= 0, ("m_copym, negative len %d", len));
502 	MBUF_CHECKSLEEP(wait);
503 	if (off == 0 && m->m_flags & M_PKTHDR)
504 		copyhdr = 1;
505 	while (off > 0) {
506 		KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
507 		if (off < m->m_len)
508 			break;
509 		off -= m->m_len;
510 		m = m->m_next;
511 	}
512 	np = &top;
513 	top = NULL;
514 	while (len > 0) {
515 		if (m == NULL) {
516 			KASSERT(len == M_COPYALL,
517 			    ("m_copym, length > size of mbuf chain"));
518 			break;
519 		}
520 		if (copyhdr)
521 			n = m_gethdr(wait, m->m_type);
522 		else
523 			n = m_get(wait, m->m_type);
524 		*np = n;
525 		if (n == NULL)
526 			goto nospace;
527 		if (copyhdr) {
528 			if (!m_dup_pkthdr(n, m, wait))
529 				goto nospace;
530 			if (len == M_COPYALL)
531 				n->m_pkthdr.len -= off0;
532 			else
533 				n->m_pkthdr.len = len;
534 			copyhdr = 0;
535 		}
536 		n->m_len = min(len, m->m_len - off);
537 		if (m->m_flags & (M_EXT|M_EXTPG)) {
538 			n->m_data = m->m_data + off;
539 			mb_dupcl(n, m);
540 		} else
541 			bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
542 			    (u_int)n->m_len);
543 		if (len != M_COPYALL)
544 			len -= n->m_len;
545 		off = 0;
546 		m = m->m_next;
547 		np = &n->m_next;
548 	}
549 
550 	return (top);
551 nospace:
552 	m_freem(top);
553 	return (NULL);
554 }
555 
556 /*
557  * Copy an entire packet, including header (which must be present).
558  * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
559  * Note that the copy is read-only, because clusters are not copied,
560  * only their reference counts are incremented.
561  * Preserve alignment of the first mbuf so if the creator has left
562  * some room at the beginning (e.g. for inserting protocol headers)
563  * the copies still have the room available.
564  */
565 struct mbuf *
566 m_copypacket(struct mbuf *m, int how)
567 {
568 	struct mbuf *top, *n, *o;
569 
570 	MBUF_CHECKSLEEP(how);
571 	n = m_get(how, m->m_type);
572 	top = n;
573 	if (n == NULL)
574 		goto nospace;
575 
576 	if (!m_dup_pkthdr(n, m, how))
577 		goto nospace;
578 	n->m_len = m->m_len;
579 	if (m->m_flags & (M_EXT|M_EXTPG)) {
580 		n->m_data = m->m_data;
581 		mb_dupcl(n, m);
582 	} else {
583 		n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
584 		bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
585 	}
586 
587 	m = m->m_next;
588 	while (m) {
589 		o = m_get(how, m->m_type);
590 		if (o == NULL)
591 			goto nospace;
592 
593 		n->m_next = o;
594 		n = n->m_next;
595 
596 		n->m_len = m->m_len;
597 		if (m->m_flags & (M_EXT|M_EXTPG)) {
598 			n->m_data = m->m_data;
599 			mb_dupcl(n, m);
600 		} else {
601 			bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
602 		}
603 
604 		m = m->m_next;
605 	}
606 	return top;
607 nospace:
608 	m_freem(top);
609 	return (NULL);
610 }
611 
612 static void
613 m_copyfromunmapped(const struct mbuf *m, int off, int len, caddr_t cp)
614 {
615 	struct iovec iov;
616 	struct uio uio;
617 	int error;
618 
619 	KASSERT(off >= 0, ("m_copyfromunmapped: negative off %d", off));
620 	KASSERT(len >= 0, ("m_copyfromunmapped: negative len %d", len));
621 	KASSERT(off < m->m_len,
622 	    ("m_copyfromunmapped: len exceeds mbuf length"));
623 	iov.iov_base = cp;
624 	iov.iov_len = len;
625 	uio.uio_resid = len;
626 	uio.uio_iov = &iov;
627 	uio.uio_segflg = UIO_SYSSPACE;
628 	uio.uio_iovcnt = 1;
629 	uio.uio_offset = 0;
630 	uio.uio_rw = UIO_READ;
631 	error = m_unmappedtouio(m, off, &uio, len);
632 	KASSERT(error == 0, ("m_unmappedtouio failed: off %d, len %d", off,
633 	   len));
634 }
635 
636 /*
637  * Copy data from an mbuf chain starting "off" bytes from the beginning,
638  * continuing for "len" bytes, into the indicated buffer.
639  */
640 void
641 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
642 {
643 	u_int count;
644 
645 	KASSERT(off >= 0, ("m_copydata, negative off %d", off));
646 	KASSERT(len >= 0, ("m_copydata, negative len %d", len));
647 	while (off > 0) {
648 		KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
649 		if (off < m->m_len)
650 			break;
651 		off -= m->m_len;
652 		m = m->m_next;
653 	}
654 	while (len > 0) {
655 		KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
656 		count = min(m->m_len - off, len);
657 		if ((m->m_flags & M_EXTPG) != 0)
658 			m_copyfromunmapped(m, off, count, cp);
659 		else
660 			bcopy(mtod(m, caddr_t) + off, cp, count);
661 		len -= count;
662 		cp += count;
663 		off = 0;
664 		m = m->m_next;
665 	}
666 }
667 
668 /*
669  * Copy a packet header mbuf chain into a completely new chain, including
670  * copying any mbuf clusters.  Use this instead of m_copypacket() when
671  * you need a writable copy of an mbuf chain.
672  */
673 struct mbuf *
674 m_dup(const struct mbuf *m, int how)
675 {
676 	struct mbuf **p, *top = NULL;
677 	int remain, moff, nsize;
678 
679 	MBUF_CHECKSLEEP(how);
680 	/* Sanity check */
681 	if (m == NULL)
682 		return (NULL);
683 	M_ASSERTPKTHDR(m);
684 
685 	/* While there's more data, get a new mbuf, tack it on, and fill it */
686 	remain = m->m_pkthdr.len;
687 	moff = 0;
688 	p = &top;
689 	while (remain > 0 || top == NULL) {	/* allow m->m_pkthdr.len == 0 */
690 		struct mbuf *n;
691 
692 		/* Get the next new mbuf */
693 		if (remain >= MINCLSIZE) {
694 			n = m_getcl(how, m->m_type, 0);
695 			nsize = MCLBYTES;
696 		} else {
697 			n = m_get(how, m->m_type);
698 			nsize = MLEN;
699 		}
700 		if (n == NULL)
701 			goto nospace;
702 
703 		if (top == NULL) {		/* First one, must be PKTHDR */
704 			if (!m_dup_pkthdr(n, m, how)) {
705 				m_free(n);
706 				goto nospace;
707 			}
708 			if ((n->m_flags & M_EXT) == 0)
709 				nsize = MHLEN;
710 			n->m_flags &= ~M_RDONLY;
711 		}
712 		n->m_len = 0;
713 
714 		/* Link it into the new chain */
715 		*p = n;
716 		p = &n->m_next;
717 
718 		/* Copy data from original mbuf(s) into new mbuf */
719 		while (n->m_len < nsize && m != NULL) {
720 			int chunk = min(nsize - n->m_len, m->m_len - moff);
721 
722 			bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
723 			moff += chunk;
724 			n->m_len += chunk;
725 			remain -= chunk;
726 			if (moff == m->m_len) {
727 				m = m->m_next;
728 				moff = 0;
729 			}
730 		}
731 
732 		/* Check correct total mbuf length */
733 		KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
734 		    	("%s: bogus m_pkthdr.len", __func__));
735 	}
736 	return (top);
737 
738 nospace:
739 	m_freem(top);
740 	return (NULL);
741 }
742 
743 /*
744  * Concatenate mbuf chain n to m.
745  * Both chains must be of the same type (e.g. MT_DATA).
746  * Any m_pkthdr is not updated.
747  */
748 void
749 m_cat(struct mbuf *m, struct mbuf *n)
750 {
751 	while (m->m_next)
752 		m = m->m_next;
753 	while (n) {
754 		if (!M_WRITABLE(m) ||
755 		    (n->m_flags & M_EXTPG) != 0 ||
756 		    M_TRAILINGSPACE(m) < n->m_len) {
757 			/* just join the two chains */
758 			m->m_next = n;
759 			return;
760 		}
761 		/* splat the data from one into the other */
762 		bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
763 		    (u_int)n->m_len);
764 		m->m_len += n->m_len;
765 		n = m_free(n);
766 	}
767 }
768 
769 /*
770  * Concatenate two pkthdr mbuf chains.
771  */
772 void
773 m_catpkt(struct mbuf *m, struct mbuf *n)
774 {
775 
776 	M_ASSERTPKTHDR(m);
777 	M_ASSERTPKTHDR(n);
778 
779 	m->m_pkthdr.len += n->m_pkthdr.len;
780 	m_demote(n, 1, 0);
781 
782 	m_cat(m, n);
783 }
784 
785 void
786 m_adj(struct mbuf *mp, int req_len)
787 {
788 	int len = req_len;
789 	struct mbuf *m;
790 	int count;
791 
792 	if ((m = mp) == NULL)
793 		return;
794 	if (len >= 0) {
795 		/*
796 		 * Trim from head.
797 		 */
798 		while (m != NULL && len > 0) {
799 			if (m->m_len <= len) {
800 				len -= m->m_len;
801 				m->m_len = 0;
802 				m = m->m_next;
803 			} else {
804 				m->m_len -= len;
805 				m->m_data += len;
806 				len = 0;
807 			}
808 		}
809 		if (mp->m_flags & M_PKTHDR)
810 			mp->m_pkthdr.len -= (req_len - len);
811 	} else {
812 		/*
813 		 * Trim from tail.  Scan the mbuf chain,
814 		 * calculating its length and finding the last mbuf.
815 		 * If the adjustment only affects this mbuf, then just
816 		 * adjust and return.  Otherwise, rescan and truncate
817 		 * after the remaining size.
818 		 */
819 		len = -len;
820 		count = 0;
821 		for (;;) {
822 			count += m->m_len;
823 			if (m->m_next == (struct mbuf *)0)
824 				break;
825 			m = m->m_next;
826 		}
827 		if (m->m_len >= len) {
828 			m->m_len -= len;
829 			if (mp->m_flags & M_PKTHDR)
830 				mp->m_pkthdr.len -= len;
831 			return;
832 		}
833 		count -= len;
834 		if (count < 0)
835 			count = 0;
836 		/*
837 		 * Correct length for chain is "count".
838 		 * Find the mbuf with last data, adjust its length,
839 		 * and toss data from remaining mbufs on chain.
840 		 */
841 		m = mp;
842 		if (m->m_flags & M_PKTHDR)
843 			m->m_pkthdr.len = count;
844 		for (; m; m = m->m_next) {
845 			if (m->m_len >= count) {
846 				m->m_len = count;
847 				if (m->m_next != NULL) {
848 					m_freem(m->m_next);
849 					m->m_next = NULL;
850 				}
851 				break;
852 			}
853 			count -= m->m_len;
854 		}
855 	}
856 }
857 
858 /*
859  * Rearange an mbuf chain so that len bytes are contiguous
860  * and in the data area of an mbuf (so that mtod will work
861  * for a structure of size len).  Returns the resulting
862  * mbuf chain on success, frees it and returns null on failure.
863  * If there is room, it will add up to max_protohdr-len extra bytes to the
864  * contiguous region in an attempt to avoid being called next time.
865  */
866 struct mbuf *
867 m_pullup(struct mbuf *n, int len)
868 {
869 	struct mbuf *m;
870 	int count;
871 	int space;
872 
873 	KASSERT((n->m_flags & M_EXTPG) == 0,
874 	    ("%s: unmapped mbuf %p", __func__, n));
875 
876 	/*
877 	 * If first mbuf has no cluster, and has room for len bytes
878 	 * without shifting current data, pullup into it,
879 	 * otherwise allocate a new mbuf to prepend to the chain.
880 	 */
881 	if ((n->m_flags & M_EXT) == 0 &&
882 	    n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
883 		if (n->m_len >= len)
884 			return (n);
885 		m = n;
886 		n = n->m_next;
887 		len -= m->m_len;
888 	} else {
889 		if (len > MHLEN)
890 			goto bad;
891 		m = m_get(M_NOWAIT, n->m_type);
892 		if (m == NULL)
893 			goto bad;
894 		if (n->m_flags & M_PKTHDR)
895 			m_move_pkthdr(m, n);
896 	}
897 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
898 	do {
899 		count = min(min(max(len, max_protohdr), space), n->m_len);
900 		bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
901 		  (u_int)count);
902 		len -= count;
903 		m->m_len += count;
904 		n->m_len -= count;
905 		space -= count;
906 		if (n->m_len)
907 			n->m_data += count;
908 		else
909 			n = m_free(n);
910 	} while (len > 0 && n);
911 	if (len > 0) {
912 		(void) m_free(m);
913 		goto bad;
914 	}
915 	m->m_next = n;
916 	return (m);
917 bad:
918 	m_freem(n);
919 	return (NULL);
920 }
921 
922 /*
923  * Like m_pullup(), except a new mbuf is always allocated, and we allow
924  * the amount of empty space before the data in the new mbuf to be specified
925  * (in the event that the caller expects to prepend later).
926  */
927 struct mbuf *
928 m_copyup(struct mbuf *n, int len, int dstoff)
929 {
930 	struct mbuf *m;
931 	int count, space;
932 
933 	if (len > (MHLEN - dstoff))
934 		goto bad;
935 	m = m_get(M_NOWAIT, n->m_type);
936 	if (m == NULL)
937 		goto bad;
938 	if (n->m_flags & M_PKTHDR)
939 		m_move_pkthdr(m, n);
940 	m->m_data += dstoff;
941 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
942 	do {
943 		count = min(min(max(len, max_protohdr), space), n->m_len);
944 		memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
945 		    (unsigned)count);
946 		len -= count;
947 		m->m_len += count;
948 		n->m_len -= count;
949 		space -= count;
950 		if (n->m_len)
951 			n->m_data += count;
952 		else
953 			n = m_free(n);
954 	} while (len > 0 && n);
955 	if (len > 0) {
956 		(void) m_free(m);
957 		goto bad;
958 	}
959 	m->m_next = n;
960 	return (m);
961  bad:
962 	m_freem(n);
963 	return (NULL);
964 }
965 
966 /*
967  * Partition an mbuf chain in two pieces, returning the tail --
968  * all but the first len0 bytes.  In case of failure, it returns NULL and
969  * attempts to restore the chain to its original state.
970  *
971  * Note that the resulting mbufs might be read-only, because the new
972  * mbuf can end up sharing an mbuf cluster with the original mbuf if
973  * the "breaking point" happens to lie within a cluster mbuf. Use the
974  * M_WRITABLE() macro to check for this case.
975  */
976 struct mbuf *
977 m_split(struct mbuf *m0, int len0, int wait)
978 {
979 	struct mbuf *m, *n;
980 	u_int len = len0, remain;
981 
982 	MBUF_CHECKSLEEP(wait);
983 	for (m = m0; m && len > m->m_len; m = m->m_next)
984 		len -= m->m_len;
985 	if (m == NULL)
986 		return (NULL);
987 	remain = m->m_len - len;
988 	if (m0->m_flags & M_PKTHDR && remain == 0) {
989 		n = m_gethdr(wait, m0->m_type);
990 		if (n == NULL)
991 			return (NULL);
992 		n->m_next = m->m_next;
993 		m->m_next = NULL;
994 		if (m0->m_pkthdr.csum_flags & CSUM_SND_TAG) {
995 			n->m_pkthdr.snd_tag =
996 			    m_snd_tag_ref(m0->m_pkthdr.snd_tag);
997 			n->m_pkthdr.csum_flags |= CSUM_SND_TAG;
998 		} else
999 			n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1000 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1001 		m0->m_pkthdr.len = len0;
1002 		return (n);
1003 	} else if (m0->m_flags & M_PKTHDR) {
1004 		n = m_gethdr(wait, m0->m_type);
1005 		if (n == NULL)
1006 			return (NULL);
1007 		if (m0->m_pkthdr.csum_flags & CSUM_SND_TAG) {
1008 			n->m_pkthdr.snd_tag =
1009 			    m_snd_tag_ref(m0->m_pkthdr.snd_tag);
1010 			n->m_pkthdr.csum_flags |= CSUM_SND_TAG;
1011 		} else
1012 			n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1013 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1014 		m0->m_pkthdr.len = len0;
1015 		if (m->m_flags & (M_EXT|M_EXTPG))
1016 			goto extpacket;
1017 		if (remain > MHLEN) {
1018 			/* m can't be the lead packet */
1019 			M_ALIGN(n, 0);
1020 			n->m_next = m_split(m, len, wait);
1021 			if (n->m_next == NULL) {
1022 				(void) m_free(n);
1023 				return (NULL);
1024 			} else {
1025 				n->m_len = 0;
1026 				return (n);
1027 			}
1028 		} else
1029 			M_ALIGN(n, remain);
1030 	} else if (remain == 0) {
1031 		n = m->m_next;
1032 		m->m_next = NULL;
1033 		return (n);
1034 	} else {
1035 		n = m_get(wait, m->m_type);
1036 		if (n == NULL)
1037 			return (NULL);
1038 		M_ALIGN(n, remain);
1039 	}
1040 extpacket:
1041 	if (m->m_flags & (M_EXT|M_EXTPG)) {
1042 		n->m_data = m->m_data + len;
1043 		mb_dupcl(n, m);
1044 	} else {
1045 		bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1046 	}
1047 	n->m_len = remain;
1048 	m->m_len = len;
1049 	n->m_next = m->m_next;
1050 	m->m_next = NULL;
1051 	return (n);
1052 }
1053 /*
1054  * Routine to copy from device local memory into mbufs.
1055  * Note that `off' argument is offset into first mbuf of target chain from
1056  * which to begin copying the data to.
1057  */
1058 struct mbuf *
1059 m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
1060     void (*copy)(char *from, caddr_t to, u_int len))
1061 {
1062 	struct mbuf *m;
1063 	struct mbuf *top = NULL, **mp = &top;
1064 	int len;
1065 
1066 	if (off < 0 || off > MHLEN)
1067 		return (NULL);
1068 
1069 	while (totlen > 0) {
1070 		if (top == NULL) {	/* First one, must be PKTHDR */
1071 			if (totlen + off >= MINCLSIZE) {
1072 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1073 				len = MCLBYTES;
1074 			} else {
1075 				m = m_gethdr(M_NOWAIT, MT_DATA);
1076 				len = MHLEN;
1077 
1078 				/* Place initial small packet/header at end of mbuf */
1079 				if (m && totlen + off + max_linkhdr <= MHLEN) {
1080 					m->m_data += max_linkhdr;
1081 					len -= max_linkhdr;
1082 				}
1083 			}
1084 			if (m == NULL)
1085 				return NULL;
1086 			m->m_pkthdr.rcvif = ifp;
1087 			m->m_pkthdr.len = totlen;
1088 		} else {
1089 			if (totlen + off >= MINCLSIZE) {
1090 				m = m_getcl(M_NOWAIT, MT_DATA, 0);
1091 				len = MCLBYTES;
1092 			} else {
1093 				m = m_get(M_NOWAIT, MT_DATA);
1094 				len = MLEN;
1095 			}
1096 			if (m == NULL) {
1097 				m_freem(top);
1098 				return NULL;
1099 			}
1100 		}
1101 		if (off) {
1102 			m->m_data += off;
1103 			len -= off;
1104 			off = 0;
1105 		}
1106 		m->m_len = len = min(totlen, len);
1107 		if (copy)
1108 			copy(buf, mtod(m, caddr_t), (u_int)len);
1109 		else
1110 			bcopy(buf, mtod(m, caddr_t), (u_int)len);
1111 		buf += len;
1112 		*mp = m;
1113 		mp = &m->m_next;
1114 		totlen -= len;
1115 	}
1116 	return (top);
1117 }
1118 
1119 /*
1120  * Copy data from a buffer back into the indicated mbuf chain,
1121  * starting "off" bytes from the beginning, extending the mbuf
1122  * chain if necessary.
1123  */
1124 void
1125 m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
1126 {
1127 	int mlen;
1128 	struct mbuf *m = m0, *n;
1129 	int totlen = 0;
1130 
1131 	if (m0 == NULL)
1132 		return;
1133 	while (off > (mlen = m->m_len)) {
1134 		off -= mlen;
1135 		totlen += mlen;
1136 		if (m->m_next == NULL) {
1137 			n = m_get(M_NOWAIT, m->m_type);
1138 			if (n == NULL)
1139 				goto out;
1140 			bzero(mtod(n, caddr_t), MLEN);
1141 			n->m_len = min(MLEN, len + off);
1142 			m->m_next = n;
1143 		}
1144 		m = m->m_next;
1145 	}
1146 	while (len > 0) {
1147 		if (m->m_next == NULL && (len > m->m_len - off)) {
1148 			m->m_len += min(len - (m->m_len - off),
1149 			    M_TRAILINGSPACE(m));
1150 		}
1151 		mlen = min (m->m_len - off, len);
1152 		bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
1153 		cp += mlen;
1154 		len -= mlen;
1155 		mlen += off;
1156 		off = 0;
1157 		totlen += mlen;
1158 		if (len == 0)
1159 			break;
1160 		if (m->m_next == NULL) {
1161 			n = m_get(M_NOWAIT, m->m_type);
1162 			if (n == NULL)
1163 				break;
1164 			n->m_len = min(MLEN, len);
1165 			m->m_next = n;
1166 		}
1167 		m = m->m_next;
1168 	}
1169 out:	if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1170 		m->m_pkthdr.len = totlen;
1171 }
1172 
1173 /*
1174  * Append the specified data to the indicated mbuf chain,
1175  * Extend the mbuf chain if the new data does not fit in
1176  * existing space.
1177  *
1178  * Return 1 if able to complete the job; otherwise 0.
1179  */
1180 int
1181 m_append(struct mbuf *m0, int len, c_caddr_t cp)
1182 {
1183 	struct mbuf *m, *n;
1184 	int remainder, space;
1185 
1186 	for (m = m0; m->m_next != NULL; m = m->m_next)
1187 		;
1188 	remainder = len;
1189 	space = M_TRAILINGSPACE(m);
1190 	if (space > 0) {
1191 		/*
1192 		 * Copy into available space.
1193 		 */
1194 		if (space > remainder)
1195 			space = remainder;
1196 		bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1197 		m->m_len += space;
1198 		cp += space, remainder -= space;
1199 	}
1200 	while (remainder > 0) {
1201 		/*
1202 		 * Allocate a new mbuf; could check space
1203 		 * and allocate a cluster instead.
1204 		 */
1205 		n = m_get(M_NOWAIT, m->m_type);
1206 		if (n == NULL)
1207 			break;
1208 		n->m_len = min(MLEN, remainder);
1209 		bcopy(cp, mtod(n, caddr_t), n->m_len);
1210 		cp += n->m_len, remainder -= n->m_len;
1211 		m->m_next = n;
1212 		m = n;
1213 	}
1214 	if (m0->m_flags & M_PKTHDR)
1215 		m0->m_pkthdr.len += len - remainder;
1216 	return (remainder == 0);
1217 }
1218 
1219 /*
1220  * Apply function f to the data in an mbuf chain starting "off" bytes from
1221  * the beginning, continuing for "len" bytes.
1222  */
1223 int
1224 m_apply(struct mbuf *m, int off, int len,
1225     int (*f)(void *, void *, u_int), void *arg)
1226 {
1227 	u_int count;
1228 	int rval;
1229 
1230 	KASSERT(off >= 0, ("m_apply, negative off %d", off));
1231 	KASSERT(len >= 0, ("m_apply, negative len %d", len));
1232 	while (off > 0) {
1233 		KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1234 		if (off < m->m_len)
1235 			break;
1236 		off -= m->m_len;
1237 		m = m->m_next;
1238 	}
1239 	while (len > 0) {
1240 		KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1241 		count = min(m->m_len - off, len);
1242 		rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1243 		if (rval)
1244 			return (rval);
1245 		len -= count;
1246 		off = 0;
1247 		m = m->m_next;
1248 	}
1249 	return (0);
1250 }
1251 
1252 /*
1253  * Return a pointer to mbuf/offset of location in mbuf chain.
1254  */
1255 struct mbuf *
1256 m_getptr(struct mbuf *m, int loc, int *off)
1257 {
1258 
1259 	while (loc >= 0) {
1260 		/* Normal end of search. */
1261 		if (m->m_len > loc) {
1262 			*off = loc;
1263 			return (m);
1264 		} else {
1265 			loc -= m->m_len;
1266 			if (m->m_next == NULL) {
1267 				if (loc == 0) {
1268 					/* Point at the end of valid data. */
1269 					*off = m->m_len;
1270 					return (m);
1271 				}
1272 				return (NULL);
1273 			}
1274 			m = m->m_next;
1275 		}
1276 	}
1277 	return (NULL);
1278 }
1279 
1280 void
1281 m_print(const struct mbuf *m, int maxlen)
1282 {
1283 	int len;
1284 	int pdata;
1285 	const struct mbuf *m2;
1286 
1287 	if (m == NULL) {
1288 		printf("mbuf: %p\n", m);
1289 		return;
1290 	}
1291 
1292 	if (m->m_flags & M_PKTHDR)
1293 		len = m->m_pkthdr.len;
1294 	else
1295 		len = -1;
1296 	m2 = m;
1297 	while (m2 != NULL && (len == -1 || len)) {
1298 		pdata = m2->m_len;
1299 		if (maxlen != -1 && pdata > maxlen)
1300 			pdata = maxlen;
1301 		printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
1302 		    m2->m_next, m2->m_flags, "\20\20freelist\17skipfw"
1303 		    "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
1304 		    "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
1305 		if (pdata)
1306 			printf(", %*D\n", pdata, (u_char *)m2->m_data, "-");
1307 		if (len != -1)
1308 			len -= m2->m_len;
1309 		m2 = m2->m_next;
1310 	}
1311 	if (len > 0)
1312 		printf("%d bytes unaccounted for.\n", len);
1313 	return;
1314 }
1315 
1316 u_int
1317 m_fixhdr(struct mbuf *m0)
1318 {
1319 	u_int len;
1320 
1321 	len = m_length(m0, NULL);
1322 	m0->m_pkthdr.len = len;
1323 	return (len);
1324 }
1325 
1326 u_int
1327 m_length(struct mbuf *m0, struct mbuf **last)
1328 {
1329 	struct mbuf *m;
1330 	u_int len;
1331 
1332 	len = 0;
1333 	for (m = m0; m != NULL; m = m->m_next) {
1334 		len += m->m_len;
1335 		if (m->m_next == NULL)
1336 			break;
1337 	}
1338 	if (last != NULL)
1339 		*last = m;
1340 	return (len);
1341 }
1342 
1343 /*
1344  * Defragment a mbuf chain, returning the shortest possible
1345  * chain of mbufs and clusters.  If allocation fails and
1346  * this cannot be completed, NULL will be returned, but
1347  * the passed in chain will be unchanged.  Upon success,
1348  * the original chain will be freed, and the new chain
1349  * will be returned.
1350  *
1351  * If a non-packet header is passed in, the original
1352  * mbuf (chain?) will be returned unharmed.
1353  */
1354 struct mbuf *
1355 m_defrag(struct mbuf *m0, int how)
1356 {
1357 	struct mbuf *m_new = NULL, *m_final = NULL;
1358 	int progress = 0, length;
1359 
1360 	MBUF_CHECKSLEEP(how);
1361 	if (!(m0->m_flags & M_PKTHDR))
1362 		return (m0);
1363 
1364 	m_fixhdr(m0); /* Needed sanity check */
1365 
1366 #ifdef MBUF_STRESS_TEST
1367 	if (m_defragrandomfailures) {
1368 		int temp = arc4random() & 0xff;
1369 		if (temp == 0xba)
1370 			goto nospace;
1371 	}
1372 #endif
1373 
1374 	if (m0->m_pkthdr.len > MHLEN)
1375 		m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1376 	else
1377 		m_final = m_gethdr(how, MT_DATA);
1378 
1379 	if (m_final == NULL)
1380 		goto nospace;
1381 
1382 	if (m_dup_pkthdr(m_final, m0, how) == 0)
1383 		goto nospace;
1384 
1385 	m_new = m_final;
1386 
1387 	while (progress < m0->m_pkthdr.len) {
1388 		length = m0->m_pkthdr.len - progress;
1389 		if (length > MCLBYTES)
1390 			length = MCLBYTES;
1391 
1392 		if (m_new == NULL) {
1393 			if (length > MLEN)
1394 				m_new = m_getcl(how, MT_DATA, 0);
1395 			else
1396 				m_new = m_get(how, MT_DATA);
1397 			if (m_new == NULL)
1398 				goto nospace;
1399 		}
1400 
1401 		m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1402 		progress += length;
1403 		m_new->m_len = length;
1404 		if (m_new != m_final)
1405 			m_cat(m_final, m_new);
1406 		m_new = NULL;
1407 	}
1408 #ifdef MBUF_STRESS_TEST
1409 	if (m0->m_next == NULL)
1410 		m_defraguseless++;
1411 #endif
1412 	m_freem(m0);
1413 	m0 = m_final;
1414 #ifdef MBUF_STRESS_TEST
1415 	m_defragpackets++;
1416 	m_defragbytes += m0->m_pkthdr.len;
1417 #endif
1418 	return (m0);
1419 nospace:
1420 #ifdef MBUF_STRESS_TEST
1421 	m_defragfailure++;
1422 #endif
1423 	if (m_final)
1424 		m_freem(m_final);
1425 	return (NULL);
1426 }
1427 
1428 /*
1429  * Return the number of fragments an mbuf will use.  This is usually
1430  * used as a proxy for the number of scatter/gather elements needed by
1431  * a DMA engine to access an mbuf.  In general mapped mbufs are
1432  * assumed to be backed by physically contiguous buffers that only
1433  * need a single fragment.  Unmapped mbufs, on the other hand, can
1434  * span disjoint physical pages.
1435  */
1436 static int
1437 frags_per_mbuf(struct mbuf *m)
1438 {
1439 	int frags;
1440 
1441 	if ((m->m_flags & M_EXTPG) == 0)
1442 		return (1);
1443 
1444 	/*
1445 	 * The header and trailer are counted as a single fragment
1446 	 * each when present.
1447 	 *
1448 	 * XXX: This overestimates the number of fragments by assuming
1449 	 * all the backing physical pages are disjoint.
1450 	 */
1451 	frags = 0;
1452 	if (m->m_epg_hdrlen != 0)
1453 		frags++;
1454 	frags += m->m_epg_npgs;
1455 	if (m->m_epg_trllen != 0)
1456 		frags++;
1457 
1458 	return (frags);
1459 }
1460 
1461 /*
1462  * Defragment an mbuf chain, returning at most maxfrags separate
1463  * mbufs+clusters.  If this is not possible NULL is returned and
1464  * the original mbuf chain is left in its present (potentially
1465  * modified) state.  We use two techniques: collapsing consecutive
1466  * mbufs and replacing consecutive mbufs by a cluster.
1467  *
1468  * NB: this should really be named m_defrag but that name is taken
1469  */
1470 struct mbuf *
1471 m_collapse(struct mbuf *m0, int how, int maxfrags)
1472 {
1473 	struct mbuf *m, *n, *n2, **prev;
1474 	u_int curfrags;
1475 
1476 	/*
1477 	 * Calculate the current number of frags.
1478 	 */
1479 	curfrags = 0;
1480 	for (m = m0; m != NULL; m = m->m_next)
1481 		curfrags += frags_per_mbuf(m);
1482 	/*
1483 	 * First, try to collapse mbufs.  Note that we always collapse
1484 	 * towards the front so we don't need to deal with moving the
1485 	 * pkthdr.  This may be suboptimal if the first mbuf has much
1486 	 * less data than the following.
1487 	 */
1488 	m = m0;
1489 again:
1490 	for (;;) {
1491 		n = m->m_next;
1492 		if (n == NULL)
1493 			break;
1494 		if (M_WRITABLE(m) &&
1495 		    n->m_len < M_TRAILINGSPACE(m)) {
1496 			m_copydata(n, 0, n->m_len,
1497 			    mtod(m, char *) + m->m_len);
1498 			m->m_len += n->m_len;
1499 			m->m_next = n->m_next;
1500 			curfrags -= frags_per_mbuf(n);
1501 			m_free(n);
1502 			if (curfrags <= maxfrags)
1503 				return m0;
1504 		} else
1505 			m = n;
1506 	}
1507 	KASSERT(maxfrags > 1,
1508 		("maxfrags %u, but normal collapse failed", maxfrags));
1509 	/*
1510 	 * Collapse consecutive mbufs to a cluster.
1511 	 */
1512 	prev = &m0->m_next;		/* NB: not the first mbuf */
1513 	while ((n = *prev) != NULL) {
1514 		if ((n2 = n->m_next) != NULL &&
1515 		    n->m_len + n2->m_len < MCLBYTES) {
1516 			m = m_getcl(how, MT_DATA, 0);
1517 			if (m == NULL)
1518 				goto bad;
1519 			m_copydata(n, 0,  n->m_len, mtod(m, char *));
1520 			m_copydata(n2, 0,  n2->m_len,
1521 			    mtod(m, char *) + n->m_len);
1522 			m->m_len = n->m_len + n2->m_len;
1523 			m->m_next = n2->m_next;
1524 			*prev = m;
1525 			curfrags += 1;  /* For the new cluster */
1526 			curfrags -= frags_per_mbuf(n);
1527 			curfrags -= frags_per_mbuf(n2);
1528 			m_free(n);
1529 			m_free(n2);
1530 			if (curfrags <= maxfrags)
1531 				return m0;
1532 			/*
1533 			 * Still not there, try the normal collapse
1534 			 * again before we allocate another cluster.
1535 			 */
1536 			goto again;
1537 		}
1538 		prev = &n->m_next;
1539 	}
1540 	/*
1541 	 * No place where we can collapse to a cluster; punt.
1542 	 * This can occur if, for example, you request 2 frags
1543 	 * but the packet requires that both be clusters (we
1544 	 * never reallocate the first mbuf to avoid moving the
1545 	 * packet header).
1546 	 */
1547 bad:
1548 	return NULL;
1549 }
1550 
1551 #ifdef MBUF_STRESS_TEST
1552 
1553 /*
1554  * Fragment an mbuf chain.  There's no reason you'd ever want to do
1555  * this in normal usage, but it's great for stress testing various
1556  * mbuf consumers.
1557  *
1558  * If fragmentation is not possible, the original chain will be
1559  * returned.
1560  *
1561  * Possible length values:
1562  * 0	 no fragmentation will occur
1563  * > 0	each fragment will be of the specified length
1564  * -1	each fragment will be the same random value in length
1565  * -2	each fragment's length will be entirely random
1566  * (Random values range from 1 to 256)
1567  */
1568 struct mbuf *
1569 m_fragment(struct mbuf *m0, int how, int length)
1570 {
1571 	struct mbuf *m_first, *m_last;
1572 	int divisor = 255, progress = 0, fraglen;
1573 
1574 	if (!(m0->m_flags & M_PKTHDR))
1575 		return (m0);
1576 
1577 	if (length == 0 || length < -2)
1578 		return (m0);
1579 	if (length > MCLBYTES)
1580 		length = MCLBYTES;
1581 	if (length < 0 && divisor > MCLBYTES)
1582 		divisor = MCLBYTES;
1583 	if (length == -1)
1584 		length = 1 + (arc4random() % divisor);
1585 	if (length > 0)
1586 		fraglen = length;
1587 
1588 	m_fixhdr(m0); /* Needed sanity check */
1589 
1590 	m_first = m_getcl(how, MT_DATA, M_PKTHDR);
1591 	if (m_first == NULL)
1592 		goto nospace;
1593 
1594 	if (m_dup_pkthdr(m_first, m0, how) == 0)
1595 		goto nospace;
1596 
1597 	m_last = m_first;
1598 
1599 	while (progress < m0->m_pkthdr.len) {
1600 		if (length == -2)
1601 			fraglen = 1 + (arc4random() % divisor);
1602 		if (fraglen > m0->m_pkthdr.len - progress)
1603 			fraglen = m0->m_pkthdr.len - progress;
1604 
1605 		if (progress != 0) {
1606 			struct mbuf *m_new = m_getcl(how, MT_DATA, 0);
1607 			if (m_new == NULL)
1608 				goto nospace;
1609 
1610 			m_last->m_next = m_new;
1611 			m_last = m_new;
1612 		}
1613 
1614 		m_copydata(m0, progress, fraglen, mtod(m_last, caddr_t));
1615 		progress += fraglen;
1616 		m_last->m_len = fraglen;
1617 	}
1618 	m_freem(m0);
1619 	m0 = m_first;
1620 	return (m0);
1621 nospace:
1622 	if (m_first)
1623 		m_freem(m_first);
1624 	/* Return the original chain on failure */
1625 	return (m0);
1626 }
1627 
1628 #endif
1629 
1630 /*
1631  * Free pages from mbuf_ext_pgs, assuming they were allocated via
1632  * vm_page_alloc() and aren't associated with any object.  Complement
1633  * to allocator from m_uiotombuf_nomap().
1634  */
1635 void
1636 mb_free_mext_pgs(struct mbuf *m)
1637 {
1638 	vm_page_t pg;
1639 
1640 	M_ASSERTEXTPG(m);
1641 	for (int i = 0; i < m->m_epg_npgs; i++) {
1642 		pg = PHYS_TO_VM_PAGE(m->m_epg_pa[i]);
1643 		vm_page_unwire_noq(pg);
1644 		vm_page_free(pg);
1645 	}
1646 }
1647 
1648 static struct mbuf *
1649 m_uiotombuf_nomap(struct uio *uio, int how, int len, int maxseg, int flags)
1650 {
1651 	struct mbuf *m, *mb, *prev;
1652 	vm_page_t pg_array[MBUF_PEXT_MAX_PGS];
1653 	int error, length, i, needed;
1654 	ssize_t total;
1655 	int pflags = malloc2vm_flags(how) | VM_ALLOC_NOOBJ | VM_ALLOC_NODUMP |
1656 	    VM_ALLOC_WIRED;
1657 
1658 	/*
1659 	 * len can be zero or an arbitrary large value bound by
1660 	 * the total data supplied by the uio.
1661 	 */
1662 	if (len > 0)
1663 		total = MIN(uio->uio_resid, len);
1664 	else
1665 		total = uio->uio_resid;
1666 
1667 	if (maxseg == 0)
1668 		maxseg = MBUF_PEXT_MAX_PGS * PAGE_SIZE;
1669 
1670 	/*
1671 	 * Allocate the pages
1672 	 */
1673 	m = NULL;
1674 	MPASS((flags & M_PKTHDR) == 0);
1675 	while (total > 0) {
1676 		mb = mb_alloc_ext_pgs(how, mb_free_mext_pgs);
1677 		if (mb == NULL)
1678 			goto failed;
1679 		if (m == NULL)
1680 			m = mb;
1681 		else
1682 			prev->m_next = mb;
1683 		prev = mb;
1684 		mb->m_epg_flags = EPG_FLAG_ANON;
1685 		needed = length = MIN(maxseg, total);
1686 		for (i = 0; needed > 0; i++, needed -= PAGE_SIZE) {
1687 retry_page:
1688 			pg_array[i] = vm_page_alloc(NULL, 0, pflags);
1689 			if (pg_array[i] == NULL) {
1690 				if (how & M_NOWAIT) {
1691 					goto failed;
1692 				} else {
1693 					vm_wait(NULL);
1694 					goto retry_page;
1695 				}
1696 			}
1697 			pg_array[i]->flags &= ~PG_ZERO;
1698 			mb->m_epg_pa[i] = VM_PAGE_TO_PHYS(pg_array[i]);
1699 			mb->m_epg_npgs++;
1700 		}
1701 		mb->m_epg_last_len = length - PAGE_SIZE * (mb->m_epg_npgs - 1);
1702 		MBUF_EXT_PGS_ASSERT_SANITY(mb);
1703 		total -= length;
1704 		error = uiomove_fromphys(pg_array, 0, length, uio);
1705 		if (error != 0)
1706 			goto failed;
1707 		mb->m_len = length;
1708 		mb->m_ext.ext_size += PAGE_SIZE * mb->m_epg_npgs;
1709 		if (flags & M_PKTHDR)
1710 			m->m_pkthdr.len += length;
1711 	}
1712 	return (m);
1713 
1714 failed:
1715 	m_freem(m);
1716 	return (NULL);
1717 }
1718 
1719 /*
1720  * Copy the contents of uio into a properly sized mbuf chain.
1721  */
1722 struct mbuf *
1723 m_uiotombuf(struct uio *uio, int how, int len, int align, int flags)
1724 {
1725 	struct mbuf *m, *mb;
1726 	int error, length;
1727 	ssize_t total;
1728 	int progress = 0;
1729 
1730 	if (flags & M_EXTPG)
1731 		return (m_uiotombuf_nomap(uio, how, len, align, flags));
1732 
1733 	/*
1734 	 * len can be zero or an arbitrary large value bound by
1735 	 * the total data supplied by the uio.
1736 	 */
1737 	if (len > 0)
1738 		total = (uio->uio_resid < len) ? uio->uio_resid : len;
1739 	else
1740 		total = uio->uio_resid;
1741 
1742 	/*
1743 	 * The smallest unit returned by m_getm2() is a single mbuf
1744 	 * with pkthdr.  We can't align past it.
1745 	 */
1746 	if (align >= MHLEN)
1747 		return (NULL);
1748 
1749 	/*
1750 	 * Give us the full allocation or nothing.
1751 	 * If len is zero return the smallest empty mbuf.
1752 	 */
1753 	m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags);
1754 	if (m == NULL)
1755 		return (NULL);
1756 	m->m_data += align;
1757 
1758 	/* Fill all mbufs with uio data and update header information. */
1759 	for (mb = m; mb != NULL; mb = mb->m_next) {
1760 		length = min(M_TRAILINGSPACE(mb), total - progress);
1761 
1762 		error = uiomove(mtod(mb, void *), length, uio);
1763 		if (error) {
1764 			m_freem(m);
1765 			return (NULL);
1766 		}
1767 
1768 		mb->m_len = length;
1769 		progress += length;
1770 		if (flags & M_PKTHDR)
1771 			m->m_pkthdr.len += length;
1772 	}
1773 	KASSERT(progress == total, ("%s: progress != total", __func__));
1774 
1775 	return (m);
1776 }
1777 
1778 /*
1779  * Copy data from an unmapped mbuf into a uio limited by len if set.
1780  */
1781 int
1782 m_unmappedtouio(const struct mbuf *m, int m_off, struct uio *uio, int len)
1783 {
1784 	vm_page_t pg;
1785 	int error, i, off, pglen, pgoff, seglen, segoff;
1786 
1787 	M_ASSERTEXTPG(m);
1788 	error = 0;
1789 
1790 	/* Skip over any data removed from the front. */
1791 	off = mtod(m, vm_offset_t);
1792 
1793 	off += m_off;
1794 	if (m->m_epg_hdrlen != 0) {
1795 		if (off >= m->m_epg_hdrlen) {
1796 			off -= m->m_epg_hdrlen;
1797 		} else {
1798 			seglen = m->m_epg_hdrlen - off;
1799 			segoff = off;
1800 			seglen = min(seglen, len);
1801 			off = 0;
1802 			len -= seglen;
1803 			error = uiomove(__DECONST(void *,
1804 			    &m->m_epg_hdr[segoff]), seglen, uio);
1805 		}
1806 	}
1807 	pgoff = m->m_epg_1st_off;
1808 	for (i = 0; i < m->m_epg_npgs && error == 0 && len > 0; i++) {
1809 		pglen = m_epg_pagelen(m, i, pgoff);
1810 		if (off >= pglen) {
1811 			off -= pglen;
1812 			pgoff = 0;
1813 			continue;
1814 		}
1815 		seglen = pglen - off;
1816 		segoff = pgoff + off;
1817 		off = 0;
1818 		seglen = min(seglen, len);
1819 		len -= seglen;
1820 		pg = PHYS_TO_VM_PAGE(m->m_epg_pa[i]);
1821 		error = uiomove_fromphys(&pg, segoff, seglen, uio);
1822 		pgoff = 0;
1823 	};
1824 	if (len != 0 && error == 0) {
1825 		KASSERT((off + len) <= m->m_epg_trllen,
1826 		    ("off + len > trail (%d + %d > %d, m_off = %d)", off, len,
1827 		    m->m_epg_trllen, m_off));
1828 		error = uiomove(__DECONST(void *, &m->m_epg_trail[off]),
1829 		    len, uio);
1830 	}
1831 	return (error);
1832 }
1833 
1834 /*
1835  * Copy an mbuf chain into a uio limited by len if set.
1836  */
1837 int
1838 m_mbuftouio(struct uio *uio, const struct mbuf *m, int len)
1839 {
1840 	int error, length, total;
1841 	int progress = 0;
1842 
1843 	if (len > 0)
1844 		total = min(uio->uio_resid, len);
1845 	else
1846 		total = uio->uio_resid;
1847 
1848 	/* Fill the uio with data from the mbufs. */
1849 	for (; m != NULL; m = m->m_next) {
1850 		length = min(m->m_len, total - progress);
1851 
1852 		if ((m->m_flags & M_EXTPG) != 0)
1853 			error = m_unmappedtouio(m, 0, uio, length);
1854 		else
1855 			error = uiomove(mtod(m, void *), length, uio);
1856 		if (error)
1857 			return (error);
1858 
1859 		progress += length;
1860 	}
1861 
1862 	return (0);
1863 }
1864 
1865 /*
1866  * Create a writable copy of the mbuf chain.  While doing this
1867  * we compact the chain with a goal of producing a chain with
1868  * at most two mbufs.  The second mbuf in this chain is likely
1869  * to be a cluster.  The primary purpose of this work is to create
1870  * a writable packet for encryption, compression, etc.  The
1871  * secondary goal is to linearize the data so the data can be
1872  * passed to crypto hardware in the most efficient manner possible.
1873  */
1874 struct mbuf *
1875 m_unshare(struct mbuf *m0, int how)
1876 {
1877 	struct mbuf *m, *mprev;
1878 	struct mbuf *n, *mfirst, *mlast;
1879 	int len, off;
1880 
1881 	mprev = NULL;
1882 	for (m = m0; m != NULL; m = mprev->m_next) {
1883 		/*
1884 		 * Regular mbufs are ignored unless there's a cluster
1885 		 * in front of it that we can use to coalesce.  We do
1886 		 * the latter mainly so later clusters can be coalesced
1887 		 * also w/o having to handle them specially (i.e. convert
1888 		 * mbuf+cluster -> cluster).  This optimization is heavily
1889 		 * influenced by the assumption that we're running over
1890 		 * Ethernet where MCLBYTES is large enough that the max
1891 		 * packet size will permit lots of coalescing into a
1892 		 * single cluster.  This in turn permits efficient
1893 		 * crypto operations, especially when using hardware.
1894 		 */
1895 		if ((m->m_flags & M_EXT) == 0) {
1896 			if (mprev && (mprev->m_flags & M_EXT) &&
1897 			    m->m_len <= M_TRAILINGSPACE(mprev)) {
1898 				/* XXX: this ignores mbuf types */
1899 				memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1900 				    mtod(m, caddr_t), m->m_len);
1901 				mprev->m_len += m->m_len;
1902 				mprev->m_next = m->m_next;	/* unlink from chain */
1903 				m_free(m);			/* reclaim mbuf */
1904 			} else {
1905 				mprev = m;
1906 			}
1907 			continue;
1908 		}
1909 		/*
1910 		 * Writable mbufs are left alone (for now).
1911 		 */
1912 		if (M_WRITABLE(m)) {
1913 			mprev = m;
1914 			continue;
1915 		}
1916 
1917 		/*
1918 		 * Not writable, replace with a copy or coalesce with
1919 		 * the previous mbuf if possible (since we have to copy
1920 		 * it anyway, we try to reduce the number of mbufs and
1921 		 * clusters so that future work is easier).
1922 		 */
1923 		KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags));
1924 		/* NB: we only coalesce into a cluster or larger */
1925 		if (mprev != NULL && (mprev->m_flags & M_EXT) &&
1926 		    m->m_len <= M_TRAILINGSPACE(mprev)) {
1927 			/* XXX: this ignores mbuf types */
1928 			memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1929 			    mtod(m, caddr_t), m->m_len);
1930 			mprev->m_len += m->m_len;
1931 			mprev->m_next = m->m_next;	/* unlink from chain */
1932 			m_free(m);			/* reclaim mbuf */
1933 			continue;
1934 		}
1935 
1936 		/*
1937 		 * Allocate new space to hold the copy and copy the data.
1938 		 * We deal with jumbo mbufs (i.e. m_len > MCLBYTES) by
1939 		 * splitting them into clusters.  We could just malloc a
1940 		 * buffer and make it external but too many device drivers
1941 		 * don't know how to break up the non-contiguous memory when
1942 		 * doing DMA.
1943 		 */
1944 		n = m_getcl(how, m->m_type, m->m_flags & M_COPYFLAGS);
1945 		if (n == NULL) {
1946 			m_freem(m0);
1947 			return (NULL);
1948 		}
1949 		if (m->m_flags & M_PKTHDR) {
1950 			KASSERT(mprev == NULL, ("%s: m0 %p, m %p has M_PKTHDR",
1951 			    __func__, m0, m));
1952 			m_move_pkthdr(n, m);
1953 		}
1954 		len = m->m_len;
1955 		off = 0;
1956 		mfirst = n;
1957 		mlast = NULL;
1958 		for (;;) {
1959 			int cc = min(len, MCLBYTES);
1960 			memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc);
1961 			n->m_len = cc;
1962 			if (mlast != NULL)
1963 				mlast->m_next = n;
1964 			mlast = n;
1965 #if 0
1966 			newipsecstat.ips_clcopied++;
1967 #endif
1968 
1969 			len -= cc;
1970 			if (len <= 0)
1971 				break;
1972 			off += cc;
1973 
1974 			n = m_getcl(how, m->m_type, m->m_flags & M_COPYFLAGS);
1975 			if (n == NULL) {
1976 				m_freem(mfirst);
1977 				m_freem(m0);
1978 				return (NULL);
1979 			}
1980 		}
1981 		n->m_next = m->m_next;
1982 		if (mprev == NULL)
1983 			m0 = mfirst;		/* new head of chain */
1984 		else
1985 			mprev->m_next = mfirst;	/* replace old mbuf */
1986 		m_free(m);			/* release old mbuf */
1987 		mprev = mfirst;
1988 	}
1989 	return (m0);
1990 }
1991 
1992 #ifdef MBUF_PROFILING
1993 
1994 #define MP_BUCKETS 32 /* don't just change this as things may overflow.*/
1995 struct mbufprofile {
1996 	uintmax_t wasted[MP_BUCKETS];
1997 	uintmax_t used[MP_BUCKETS];
1998 	uintmax_t segments[MP_BUCKETS];
1999 } mbprof;
2000 
2001 #define MP_MAXDIGITS 21	/* strlen("16,000,000,000,000,000,000") == 21 */
2002 #define MP_NUMLINES 6
2003 #define MP_NUMSPERLINE 16
2004 #define MP_EXTRABYTES 64	/* > strlen("used:\nwasted:\nsegments:\n") */
2005 /* work out max space needed and add a bit of spare space too */
2006 #define MP_MAXLINE ((MP_MAXDIGITS+1) * MP_NUMSPERLINE)
2007 #define MP_BUFSIZE ((MP_MAXLINE * MP_NUMLINES) + 1 + MP_EXTRABYTES)
2008 
2009 char mbprofbuf[MP_BUFSIZE];
2010 
2011 void
2012 m_profile(struct mbuf *m)
2013 {
2014 	int segments = 0;
2015 	int used = 0;
2016 	int wasted = 0;
2017 
2018 	while (m) {
2019 		segments++;
2020 		used += m->m_len;
2021 		if (m->m_flags & M_EXT) {
2022 			wasted += MHLEN - sizeof(m->m_ext) +
2023 			    m->m_ext.ext_size - m->m_len;
2024 		} else {
2025 			if (m->m_flags & M_PKTHDR)
2026 				wasted += MHLEN - m->m_len;
2027 			else
2028 				wasted += MLEN - m->m_len;
2029 		}
2030 		m = m->m_next;
2031 	}
2032 	/* be paranoid.. it helps */
2033 	if (segments > MP_BUCKETS - 1)
2034 		segments = MP_BUCKETS - 1;
2035 	if (used > 100000)
2036 		used = 100000;
2037 	if (wasted > 100000)
2038 		wasted = 100000;
2039 	/* store in the appropriate bucket */
2040 	/* don't bother locking. if it's slightly off, so what? */
2041 	mbprof.segments[segments]++;
2042 	mbprof.used[fls(used)]++;
2043 	mbprof.wasted[fls(wasted)]++;
2044 }
2045 
2046 static void
2047 mbprof_textify(void)
2048 {
2049 	int offset;
2050 	char *c;
2051 	uint64_t *p;
2052 
2053 	p = &mbprof.wasted[0];
2054 	c = mbprofbuf;
2055 	offset = snprintf(c, MP_MAXLINE + 10,
2056 	    "wasted:\n"
2057 	    "%ju %ju %ju %ju %ju %ju %ju %ju "
2058 	    "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2059 	    p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2060 	    p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2061 #ifdef BIG_ARRAY
2062 	p = &mbprof.wasted[16];
2063 	c += offset;
2064 	offset = snprintf(c, MP_MAXLINE,
2065 	    "%ju %ju %ju %ju %ju %ju %ju %ju "
2066 	    "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2067 	    p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2068 	    p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2069 #endif
2070 	p = &mbprof.used[0];
2071 	c += offset;
2072 	offset = snprintf(c, MP_MAXLINE + 10,
2073 	    "used:\n"
2074 	    "%ju %ju %ju %ju %ju %ju %ju %ju "
2075 	    "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2076 	    p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2077 	    p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2078 #ifdef BIG_ARRAY
2079 	p = &mbprof.used[16];
2080 	c += offset;
2081 	offset = snprintf(c, MP_MAXLINE,
2082 	    "%ju %ju %ju %ju %ju %ju %ju %ju "
2083 	    "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2084 	    p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2085 	    p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2086 #endif
2087 	p = &mbprof.segments[0];
2088 	c += offset;
2089 	offset = snprintf(c, MP_MAXLINE + 10,
2090 	    "segments:\n"
2091 	    "%ju %ju %ju %ju %ju %ju %ju %ju "
2092 	    "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2093 	    p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2094 	    p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2095 #ifdef BIG_ARRAY
2096 	p = &mbprof.segments[16];
2097 	c += offset;
2098 	offset = snprintf(c, MP_MAXLINE,
2099 	    "%ju %ju %ju %ju %ju %ju %ju %ju "
2100 	    "%ju %ju %ju %ju %ju %ju %ju %jju",
2101 	    p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2102 	    p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2103 #endif
2104 }
2105 
2106 static int
2107 mbprof_handler(SYSCTL_HANDLER_ARGS)
2108 {
2109 	int error;
2110 
2111 	mbprof_textify();
2112 	error = SYSCTL_OUT(req, mbprofbuf, strlen(mbprofbuf) + 1);
2113 	return (error);
2114 }
2115 
2116 static int
2117 mbprof_clr_handler(SYSCTL_HANDLER_ARGS)
2118 {
2119 	int clear, error;
2120 
2121 	clear = 0;
2122 	error = sysctl_handle_int(oidp, &clear, 0, req);
2123 	if (error || !req->newptr)
2124 		return (error);
2125 
2126 	if (clear) {
2127 		bzero(&mbprof, sizeof(mbprof));
2128 	}
2129 
2130 	return (error);
2131 }
2132 
2133 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofile,
2134     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, NULL, 0,
2135     mbprof_handler, "A",
2136     "mbuf profiling statistics");
2137 
2138 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofileclr,
2139     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, NULL, 0,
2140     mbprof_clr_handler, "I",
2141     "clear mbuf profiling statistics");
2142 #endif
2143 
2144