2 * Copyright (c) 1982, 1986, 1988, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include "opt_param.h"
36 #include "opt_mbuf_stress_test.h"
37 #include "opt_mbuf_profiling.h"
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #include <sys/limits.h>
44 #include <sys/malloc.h>
46 #include <sys/sysctl.h>
47 #include <sys/domain.h>
48 #include <sys/protosw.h>
55 #ifdef MBUF_STRESS_TEST
60 int m_defragrandomfailures;
64 * sysctl(8) exported objects
66 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD,
67 &max_linkhdr, 0, "Size of largest link layer header");
68 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD,
69 &max_protohdr, 0, "Size of largest protocol layer header");
70 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD,
71 &max_hdr, 0, "Size of largest link plus protocol header");
72 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RD,
73 &max_datalen, 0, "Minimum space left in mbuf after max_hdr");
74 #ifdef MBUF_STRESS_TEST
75 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
76 &m_defragpackets, 0, "");
77 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
78 &m_defragbytes, 0, "");
79 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
80 &m_defraguseless, 0, "");
81 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
82 &m_defragfailure, 0, "");
83 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
84 &m_defragrandomfailures, 0, "");
88 * Ensure the correct size of various mbuf parameters. It could be off due
89 * to compiler-induced padding and alignment artifacts.
91 CTASSERT(sizeof(struct mbuf) == MSIZE);
92 CTASSERT(MSIZE - offsetof(struct mbuf, m_dat) == MLEN);
93 CTASSERT(MSIZE - offsetof(struct mbuf, m_pktdat) == MHLEN);
96 * m_get2() allocates minimum mbuf that would fit "size" argument.
99 m_get2(int size, int how, short type, int flags)
107 if (size <= MHLEN || (size <= MLEN && (flags & M_PKTHDR) == 0))
108 return (uma_zalloc_arg(zone_mbuf, &args, how));
109 if (size <= MCLBYTES)
110 return (uma_zalloc_arg(zone_pack, &args, how));
112 if (size > MJUMPAGESIZE)
115 m = uma_zalloc_arg(zone_mbuf, &args, how);
119 n = uma_zalloc_arg(zone_jumbop, m, how);
121 uma_zfree(zone_mbuf, m);
129 * m_getjcl() returns an mbuf with a cluster of the specified size attached.
130 * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES.
133 m_getjcl(int how, short type, int flags, int size)
139 if (size == MCLBYTES)
140 return m_getcl(how, type, flags);
145 m = uma_zalloc_arg(zone_mbuf, &args, how);
149 zone = m_getzone(size);
150 n = uma_zalloc_arg(zone, m, how);
152 uma_zfree(zone_mbuf, m);
159 * Allocate a given length worth of mbufs and/or clusters (whatever fits
160 * best) and return a pointer to the top of the allocated chain. If an
161 * existing mbuf chain is provided, then we will append the new chain
162 * to the existing one but still return the top of the newly allocated
166 m_getm2(struct mbuf *m, int len, int how, short type, int flags)
168 struct mbuf *mb, *nm = NULL, *mtail = NULL;
170 KASSERT(len >= 0, ("%s: len is < 0", __func__));
172 /* Validate flags. */
173 flags &= (M_PKTHDR | M_EOR);
175 /* Packet header mbuf must be first in chain. */
176 if ((flags & M_PKTHDR) && m != NULL)
179 /* Loop and append maximum sized mbufs to the chain tail. */
182 mb = m_getjcl(how, type, (flags & M_PKTHDR),
184 else if (len >= MINCLSIZE)
185 mb = m_getcl(how, type, (flags & M_PKTHDR));
186 else if (flags & M_PKTHDR)
187 mb = m_gethdr(how, type);
189 mb = m_get(how, type);
191 /* Fail the whole operation if one mbuf can't be allocated. */
199 len -= (mb->m_flags & M_EXT) ? mb->m_ext.ext_size :
200 ((mb->m_flags & M_PKTHDR) ? MHLEN : MLEN);
206 flags &= ~M_PKTHDR; /* Only valid on the first mbuf. */
209 mtail->m_flags |= M_EOR; /* Only valid on the last mbuf. */
211 /* If mbuf was supplied, append new chain to the end of it. */
213 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next)
216 mtail->m_flags &= ~M_EOR;
224 * Free an entire chain of mbufs and associated external buffers, if
228 m_freem(struct mbuf *mb)
236 * Configure a provided mbuf to refer to the provided external storage
237 * buffer and setup a reference count for said buffer. If the setting
238 * up of the reference count fails, the M_EXT bit will not be set. If
239 * successfull, the M_EXT bit is set in the mbuf's flags.
242 * mb The existing mbuf to which to attach the provided buffer.
243 * buf The address of the provided external storage buffer.
244 * size The size of the provided buffer.
245 * freef A pointer to a routine that is responsible for freeing the
246 * provided external storage buffer.
247 * args A pointer to an argument structure (of any type) to be passed
248 * to the provided freef routine (may be NULL).
249 * flags Any other flags to be passed to the provided mbuf.
250 * type The type that the external storage buffer should be
257 m_extadd(struct mbuf *mb, caddr_t buf, u_int size,
258 int (*freef)(struct mbuf *, void *, void *), void *arg1, void *arg2,
259 int flags, int type, int wait)
261 KASSERT(type != EXT_CLUSTER, ("%s: EXT_CLUSTER not allowed", __func__));
263 if (type != EXT_EXTREF)
264 mb->m_ext.ref_cnt = uma_zalloc(zone_ext_refcnt, wait);
266 if (mb->m_ext.ref_cnt == NULL)
269 *(mb->m_ext.ref_cnt) = 1;
270 mb->m_flags |= (M_EXT | flags);
271 mb->m_ext.ext_buf = buf;
272 mb->m_data = mb->m_ext.ext_buf;
273 mb->m_ext.ext_size = size;
274 mb->m_ext.ext_free = freef;
275 mb->m_ext.ext_arg1 = arg1;
276 mb->m_ext.ext_arg2 = arg2;
277 mb->m_ext.ext_type = type;
278 mb->m_ext.ext_flags = 0;
284 * Non-directly-exported function to clean up after mbufs with M_EXT
285 * storage attached to them if the reference count hits 1.
288 mb_free_ext(struct mbuf *m)
292 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
293 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
296 * check if the header is embedded in the cluster
298 skipmbuf = (m->m_flags & M_NOFREE);
300 /* Free attached storage if this mbuf is the only reference to it. */
301 if (*(m->m_ext.ref_cnt) == 1 ||
302 atomic_fetchadd_int(m->m_ext.ref_cnt, -1) == 1) {
303 switch (m->m_ext.ext_type) {
304 case EXT_PACKET: /* The packet zone is special. */
305 if (*(m->m_ext.ref_cnt) == 0)
306 *(m->m_ext.ref_cnt) = 1;
307 uma_zfree(zone_pack, m);
308 return; /* Job done. */
310 uma_zfree(zone_clust, m->m_ext.ext_buf);
313 uma_zfree(zone_jumbop, m->m_ext.ext_buf);
316 uma_zfree(zone_jumbo9, m->m_ext.ext_buf);
319 uma_zfree(zone_jumbo16, m->m_ext.ext_buf);
325 *(m->m_ext.ref_cnt) = 0;
326 uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *,
330 KASSERT(m->m_ext.ext_free != NULL,
331 ("%s: ext_free not set", __func__));
332 (void)(*(m->m_ext.ext_free))(m, m->m_ext.ext_arg1,
336 KASSERT(m->m_ext.ext_type == 0,
337 ("%s: unknown ext_type", __func__));
344 * Free this mbuf back to the mbuf zone with all m_ext
345 * information purged.
347 m->m_ext.ext_buf = NULL;
348 m->m_ext.ext_free = NULL;
349 m->m_ext.ext_arg1 = NULL;
350 m->m_ext.ext_arg2 = NULL;
351 m->m_ext.ref_cnt = NULL;
352 m->m_ext.ext_size = 0;
353 m->m_ext.ext_type = 0;
354 m->m_ext.ext_flags = 0;
355 m->m_flags &= ~M_EXT;
356 uma_zfree(zone_mbuf, m);
360 * Attach the cluster from *m to *n, set up m_ext in *n
361 * and bump the refcount of the cluster.
364 mb_dupcl(struct mbuf *n, struct mbuf *m)
366 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
367 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
368 KASSERT((n->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
370 if (*(m->m_ext.ref_cnt) == 1)
371 *(m->m_ext.ref_cnt) += 1;
373 atomic_add_int(m->m_ext.ref_cnt, 1);
374 n->m_ext.ext_buf = m->m_ext.ext_buf;
375 n->m_ext.ext_free = m->m_ext.ext_free;
376 n->m_ext.ext_arg1 = m->m_ext.ext_arg1;
377 n->m_ext.ext_arg2 = m->m_ext.ext_arg2;
378 n->m_ext.ext_size = m->m_ext.ext_size;
379 n->m_ext.ref_cnt = m->m_ext.ref_cnt;
380 n->m_ext.ext_type = m->m_ext.ext_type;
381 n->m_ext.ext_flags = m->m_ext.ext_flags;
383 n->m_flags |= m->m_flags & M_RDONLY;
387 * Clean up mbuf (chain) from any tags and packet headers.
388 * If "all" is set then the first mbuf in the chain will be
392 m_demote(struct mbuf *m0, int all)
396 for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) {
397 if (m->m_flags & M_PKTHDR) {
398 m_tag_delete_chain(m, NULL);
399 m->m_flags &= ~M_PKTHDR;
400 bzero(&m->m_pkthdr, sizeof(struct pkthdr));
402 if (m != m0 && m->m_nextpkt != NULL) {
403 KASSERT(m->m_nextpkt == NULL,
404 ("%s: m_nextpkt not NULL", __func__));
405 m_freem(m->m_nextpkt);
408 m->m_flags = m->m_flags & (M_EXT|M_RDONLY|M_NOFREE);
413 * Sanity checks on mbuf (chain) for use in KASSERT() and general
415 * Returns 0 or panics when bad and 1 on all tests passed.
416 * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they
420 m_sanity(struct mbuf *m0, int sanitize)
427 #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m)
429 #define M_SANITY_ACTION(s) printf("mbuf %p: " s, m)
432 for (m = m0; m != NULL; m = m->m_next) {
434 * Basic pointer checks. If any of these fails then some
435 * unrelated kernel memory before or after us is trashed.
436 * No way to recover from that.
438 a = ((m->m_flags & M_EXT) ? m->m_ext.ext_buf :
439 ((m->m_flags & M_PKTHDR) ? (caddr_t)(&m->m_pktdat) :
440 (caddr_t)(&m->m_dat)) );
441 b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size :
442 ((m->m_flags & M_PKTHDR) ? MHLEN : MLEN)));
443 if ((caddr_t)m->m_data < a)
444 M_SANITY_ACTION("m_data outside mbuf data range left");
445 if ((caddr_t)m->m_data > b)
446 M_SANITY_ACTION("m_data outside mbuf data range right");
447 if ((caddr_t)m->m_data + m->m_len > b)
448 M_SANITY_ACTION("m_data + m_len exeeds mbuf space");
450 /* m->m_nextpkt may only be set on first mbuf in chain. */
451 if (m != m0 && m->m_nextpkt != NULL) {
453 m_freem(m->m_nextpkt);
454 m->m_nextpkt = (struct mbuf *)0xDEADC0DE;
456 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf");
459 /* packet length (not mbuf length!) calculation */
460 if (m0->m_flags & M_PKTHDR)
463 /* m_tags may only be attached to first mbuf in chain. */
464 if (m != m0 && m->m_flags & M_PKTHDR &&
465 !SLIST_EMPTY(&m->m_pkthdr.tags)) {
467 m_tag_delete_chain(m, NULL);
468 /* put in 0xDEADC0DE perhaps? */
470 M_SANITY_ACTION("m_tags on in-chain mbuf");
473 /* M_PKTHDR may only be set on first mbuf in chain */
474 if (m != m0 && m->m_flags & M_PKTHDR) {
476 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
477 m->m_flags &= ~M_PKTHDR;
478 /* put in 0xDEADCODE and leave hdr flag in */
480 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf");
484 if (pktlen && pktlen != m->m_pkthdr.len) {
488 M_SANITY_ACTION("m_pkthdr.len != mbuf chain length");
492 #undef M_SANITY_ACTION
497 * "Move" mbuf pkthdr from "from" to "to".
498 * "from" must have M_PKTHDR set, and "to" must be empty.
501 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
505 /* see below for why these are not enabled */
507 /* Note: with MAC, this may not be a good assertion. */
508 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
509 ("m_move_pkthdr: to has tags"));
513 * XXXMAC: It could be this should also occur for non-MAC?
515 if (to->m_flags & M_PKTHDR)
516 m_tag_delete_chain(to, NULL);
518 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
519 if ((to->m_flags & M_EXT) == 0)
520 to->m_data = to->m_pktdat;
521 to->m_pkthdr = from->m_pkthdr; /* especially tags */
522 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
523 from->m_flags &= ~M_PKTHDR;
527 * Duplicate "from"'s mbuf pkthdr in "to".
528 * "from" must have M_PKTHDR set, and "to" must be empty.
529 * In particular, this does a deep copy of the packet tags.
532 m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how)
537 * The mbuf allocator only initializes the pkthdr
538 * when the mbuf is allocated with m_gethdr(). Many users
539 * (e.g. m_copy*, m_prepend) use m_get() and then
540 * smash the pkthdr as needed causing these
541 * assertions to trip. For now just disable them.
544 /* Note: with MAC, this may not be a good assertion. */
545 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
547 MBUF_CHECKSLEEP(how);
549 if (to->m_flags & M_PKTHDR)
550 m_tag_delete_chain(to, NULL);
552 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
553 if ((to->m_flags & M_EXT) == 0)
554 to->m_data = to->m_pktdat;
555 to->m_pkthdr = from->m_pkthdr;
556 SLIST_INIT(&to->m_pkthdr.tags);
557 return (m_tag_copy_chain(to, from, MBTOM(how)));
561 * Lesser-used path for M_PREPEND:
562 * allocate new mbuf to prepend to chain,
566 m_prepend(struct mbuf *m, int len, int how)
570 if (m->m_flags & M_PKTHDR)
571 mn = m_gethdr(how, m->m_type);
573 mn = m_get(how, m->m_type);
578 if (m->m_flags & M_PKTHDR)
579 m_move_pkthdr(mn, m);
582 if(m->m_flags & M_PKTHDR) {
594 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
595 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
596 * The wait parameter is a choice of M_WAITOK/M_NOWAIT from caller.
597 * Note that the copy is read-only, because clusters are not copied,
598 * only their reference counts are incremented.
601 m_copym(struct mbuf *m, int off0, int len, int wait)
603 struct mbuf *n, **np;
608 KASSERT(off >= 0, ("m_copym, negative off %d", off));
609 KASSERT(len >= 0, ("m_copym, negative len %d", len));
610 MBUF_CHECKSLEEP(wait);
611 if (off == 0 && m->m_flags & M_PKTHDR)
614 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
624 KASSERT(len == M_COPYALL,
625 ("m_copym, length > size of mbuf chain"));
629 n = m_gethdr(wait, m->m_type);
631 n = m_get(wait, m->m_type);
636 if (!m_dup_pkthdr(n, m, wait))
638 if (len == M_COPYALL)
639 n->m_pkthdr.len -= off0;
641 n->m_pkthdr.len = len;
644 n->m_len = min(len, m->m_len - off);
645 if (m->m_flags & M_EXT) {
646 n->m_data = m->m_data + off;
649 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
651 if (len != M_COPYALL)
665 * Returns mbuf chain with new head for the prepending case.
666 * Copies from mbuf (chain) n from off for len to mbuf (chain) m
667 * either prepending or appending the data.
668 * The resulting mbuf (chain) m is fully writeable.
669 * m is destination (is made writeable)
670 * n is source, off is offset in source, len is len from offset
671 * dir, 0 append, 1 prepend
672 * how, wait or nowait
676 m_bcopyxxx(void *s, void *t, u_int len)
678 bcopy(s, t, (size_t)len);
683 m_copymdata(struct mbuf *m, struct mbuf *n, int off, int len,
686 struct mbuf *mm, *x, *z, *prev = NULL;
691 KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source"));
692 KASSERT(off >= 0, ("m_copymdata, negative off %d", off));
693 KASSERT(len >= 0, ("m_copymdata, negative len %d", len));
694 KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep));
703 for (z = n; z != NULL; z = z->m_next)
705 if (len == M_COPYALL)
707 if (off + len > nlen || len < 1)
710 if (!M_WRITABLE(mm)) {
711 /* XXX: Use proper m_xxx function instead. */
712 x = m_getcl(how, MT_DATA, mm->m_flags);
715 bcopy(mm->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size);
716 p = x->m_ext.ext_buf + (mm->m_data - mm->m_ext.ext_buf);
726 * Append/prepend the data. Allocating mbufs as necessary.
728 /* Shortcut if enough free space in first/last mbuf. */
729 if (!prep && M_TRAILINGSPACE(mm) >= len) {
730 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) +
733 mm->m_pkthdr.len += len;
736 if (prep && M_LEADINGSPACE(mm) >= len) {
737 mm->m_data = mtod(mm, caddr_t) - len;
738 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t));
740 mm->m_pkthdr.len += len;
744 /* Expand first/last mbuf to cluster if possible. */
745 if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) {
746 bcopy(mm->m_data, &buf, mm->m_len);
748 if (!(mm->m_flags & M_EXT))
750 bcopy(&buf, mm->m_ext.ext_buf, mm->m_len);
751 mm->m_data = mm->m_ext.ext_buf;
753 if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) {
754 bcopy(mm->m_data, &buf, mm->m_len);
756 if (!(mm->m_flags & M_EXT))
758 bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf +
759 mm->m_ext.ext_size - mm->m_len, mm->m_len);
760 mm->m_data = (caddr_t)mm->m_ext.ext_buf +
761 mm->m_ext.ext_size - mm->m_len;
764 /* Append/prepend as many mbuf (clusters) as necessary to fit len. */
765 if (!prep && len > M_TRAILINGSPACE(mm)) {
766 if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA))
769 if (prep && len > M_LEADINGSPACE(mm)) {
770 if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA)))
773 for (x = z; x != NULL; x = x->m_next) {
774 i += x->m_flags & M_EXT ? x->m_ext.ext_size :
775 (x->m_flags & M_PKTHDR ? MHLEN : MLEN);
779 z->m_data += i - len;
780 m_move_pkthdr(mm, z);
785 /* Seek to start position in source mbuf. Optimization for long chains. */
793 /* Copy data into target mbuf. */
796 KASSERT(z != NULL, ("m_copymdata, falling off target edge"));
797 i = M_TRAILINGSPACE(z);
798 m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len);
800 /* fixup pkthdr.len if necessary */
801 if ((prep ? mm : m)->m_flags & M_PKTHDR)
802 (prep ? mm : m)->m_pkthdr.len += i;
807 return (prep ? mm : m);
811 * Copy an entire packet, including header (which must be present).
812 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
813 * Note that the copy is read-only, because clusters are not copied,
814 * only their reference counts are incremented.
815 * Preserve alignment of the first mbuf so if the creator has left
816 * some room at the beginning (e.g. for inserting protocol headers)
817 * the copies still have the room available.
820 m_copypacket(struct mbuf *m, int how)
822 struct mbuf *top, *n, *o;
824 MBUF_CHECKSLEEP(how);
825 n = m_get(how, m->m_type);
830 if (!m_dup_pkthdr(n, m, how))
833 if (m->m_flags & M_EXT) {
834 n->m_data = m->m_data;
837 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
838 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
843 o = m_get(how, m->m_type);
851 if (m->m_flags & M_EXT) {
852 n->m_data = m->m_data;
855 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
867 * Copy data from an mbuf chain starting "off" bytes from the beginning,
868 * continuing for "len" bytes, into the indicated buffer.
871 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
875 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
876 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
878 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
885 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
886 count = min(m->m_len - off, len);
887 bcopy(mtod(m, caddr_t) + off, cp, count);
896 * Copy a packet header mbuf chain into a completely new chain, including
897 * copying any mbuf clusters. Use this instead of m_copypacket() when
898 * you need a writable copy of an mbuf chain.
901 m_dup(struct mbuf *m, int how)
903 struct mbuf **p, *top = NULL;
904 int remain, moff, nsize;
906 MBUF_CHECKSLEEP(how);
912 /* While there's more data, get a new mbuf, tack it on, and fill it */
913 remain = m->m_pkthdr.len;
916 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
919 /* Get the next new mbuf */
920 if (remain >= MINCLSIZE) {
921 n = m_getcl(how, m->m_type, 0);
924 n = m_get(how, m->m_type);
930 if (top == NULL) { /* First one, must be PKTHDR */
931 if (!m_dup_pkthdr(n, m, how)) {
935 if ((n->m_flags & M_EXT) == 0)
937 n->m_flags &= ~M_RDONLY;
941 /* Link it into the new chain */
945 /* Copy data from original mbuf(s) into new mbuf */
946 while (n->m_len < nsize && m != NULL) {
947 int chunk = min(nsize - n->m_len, m->m_len - moff);
949 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
953 if (moff == m->m_len) {
959 /* Check correct total mbuf length */
960 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
961 ("%s: bogus m_pkthdr.len", __func__));
971 * Concatenate mbuf chain n to m.
972 * Both chains must be of the same type (e.g. MT_DATA).
973 * Any m_pkthdr is not updated.
976 m_cat(struct mbuf *m, struct mbuf *n)
981 if (!M_WRITABLE(m) ||
982 M_TRAILINGSPACE(m) < n->m_len) {
983 /* just join the two chains */
987 /* splat the data from one into the other */
988 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
990 m->m_len += n->m_len;
996 m_adj(struct mbuf *mp, int req_len)
1002 if ((m = mp) == NULL)
1008 while (m != NULL && len > 0) {
1009 if (m->m_len <= len) {
1019 if (mp->m_flags & M_PKTHDR)
1020 mp->m_pkthdr.len -= (req_len - len);
1023 * Trim from tail. Scan the mbuf chain,
1024 * calculating its length and finding the last mbuf.
1025 * If the adjustment only affects this mbuf, then just
1026 * adjust and return. Otherwise, rescan and truncate
1027 * after the remaining size.
1033 if (m->m_next == (struct mbuf *)0)
1037 if (m->m_len >= len) {
1039 if (mp->m_flags & M_PKTHDR)
1040 mp->m_pkthdr.len -= len;
1047 * Correct length for chain is "count".
1048 * Find the mbuf with last data, adjust its length,
1049 * and toss data from remaining mbufs on chain.
1052 if (m->m_flags & M_PKTHDR)
1053 m->m_pkthdr.len = count;
1054 for (; m; m = m->m_next) {
1055 if (m->m_len >= count) {
1057 if (m->m_next != NULL) {
1069 * Rearange an mbuf chain so that len bytes are contiguous
1070 * and in the data area of an mbuf (so that mtod will work
1071 * for a structure of size len). Returns the resulting
1072 * mbuf chain on success, frees it and returns null on failure.
1073 * If there is room, it will add up to max_protohdr-len extra bytes to the
1074 * contiguous region in an attempt to avoid being called next time.
1077 m_pullup(struct mbuf *n, int len)
1084 * If first mbuf has no cluster, and has room for len bytes
1085 * without shifting current data, pullup into it,
1086 * otherwise allocate a new mbuf to prepend to the chain.
1088 if ((n->m_flags & M_EXT) == 0 &&
1089 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
1090 if (n->m_len >= len)
1098 m = m_get(M_NOWAIT, n->m_type);
1101 if (n->m_flags & M_PKTHDR)
1102 m_move_pkthdr(m, n);
1104 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1106 count = min(min(max(len, max_protohdr), space), n->m_len);
1107 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1117 } while (len > 0 && n);
1130 * Like m_pullup(), except a new mbuf is always allocated, and we allow
1131 * the amount of empty space before the data in the new mbuf to be specified
1132 * (in the event that the caller expects to prepend later).
1137 m_copyup(struct mbuf *n, int len, int dstoff)
1142 if (len > (MHLEN - dstoff))
1144 m = m_get(M_NOWAIT, n->m_type);
1147 if (n->m_flags & M_PKTHDR)
1148 m_move_pkthdr(m, n);
1149 m->m_data += dstoff;
1150 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1152 count = min(min(max(len, max_protohdr), space), n->m_len);
1153 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
1163 } while (len > 0 && n);
1177 * Partition an mbuf chain in two pieces, returning the tail --
1178 * all but the first len0 bytes. In case of failure, it returns NULL and
1179 * attempts to restore the chain to its original state.
1181 * Note that the resulting mbufs might be read-only, because the new
1182 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1183 * the "breaking point" happens to lie within a cluster mbuf. Use the
1184 * M_WRITABLE() macro to check for this case.
1187 m_split(struct mbuf *m0, int len0, int wait)
1190 u_int len = len0, remain;
1192 MBUF_CHECKSLEEP(wait);
1193 for (m = m0; m && len > m->m_len; m = m->m_next)
1197 remain = m->m_len - len;
1198 if (m0->m_flags & M_PKTHDR && remain == 0) {
1199 n = m_gethdr(wait, m0->m_type);
1202 n->m_next = m->m_next;
1204 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1205 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1206 m0->m_pkthdr.len = len0;
1208 } else if (m0->m_flags & M_PKTHDR) {
1209 n = m_gethdr(wait, m0->m_type);
1212 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1213 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1214 m0->m_pkthdr.len = len0;
1215 if (m->m_flags & M_EXT)
1217 if (remain > MHLEN) {
1218 /* m can't be the lead packet */
1220 n->m_next = m_split(m, len, wait);
1221 if (n->m_next == NULL) {
1229 MH_ALIGN(n, remain);
1230 } else if (remain == 0) {
1235 n = m_get(wait, m->m_type);
1241 if (m->m_flags & M_EXT) {
1242 n->m_data = m->m_data + len;
1245 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1249 n->m_next = m->m_next;
1254 * Routine to copy from device local memory into mbufs.
1255 * Note that `off' argument is offset into first mbuf of target chain from
1256 * which to begin copying the data to.
1259 m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
1260 void (*copy)(char *from, caddr_t to, u_int len))
1263 struct mbuf *top = NULL, **mp = ⊤
1266 if (off < 0 || off > MHLEN)
1269 while (totlen > 0) {
1270 if (top == NULL) { /* First one, must be PKTHDR */
1271 if (totlen + off >= MINCLSIZE) {
1272 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1275 m = m_gethdr(M_NOWAIT, MT_DATA);
1278 /* Place initial small packet/header at end of mbuf */
1279 if (m && totlen + off + max_linkhdr <= MLEN) {
1280 m->m_data += max_linkhdr;
1286 m->m_pkthdr.rcvif = ifp;
1287 m->m_pkthdr.len = totlen;
1289 if (totlen + off >= MINCLSIZE) {
1290 m = m_getcl(M_NOWAIT, MT_DATA, 0);
1293 m = m_get(M_NOWAIT, MT_DATA);
1306 m->m_len = len = min(totlen, len);
1308 copy(buf, mtod(m, caddr_t), (u_int)len);
1310 bcopy(buf, mtod(m, caddr_t), (u_int)len);
1320 * Copy data from a buffer back into the indicated mbuf chain,
1321 * starting "off" bytes from the beginning, extending the mbuf
1322 * chain if necessary.
1325 m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
1328 struct mbuf *m = m0, *n;
1333 while (off > (mlen = m->m_len)) {
1336 if (m->m_next == NULL) {
1337 n = m_get(M_NOWAIT, m->m_type);
1340 bzero(mtod(n, caddr_t), MLEN);
1341 n->m_len = min(MLEN, len + off);
1347 if (m->m_next == NULL && (len > m->m_len - off)) {
1348 m->m_len += min(len - (m->m_len - off),
1349 M_TRAILINGSPACE(m));
1351 mlen = min (m->m_len - off, len);
1352 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
1360 if (m->m_next == NULL) {
1361 n = m_get(M_NOWAIT, m->m_type);
1364 n->m_len = min(MLEN, len);
1369 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1370 m->m_pkthdr.len = totlen;
1374 * Append the specified data to the indicated mbuf chain,
1375 * Extend the mbuf chain if the new data does not fit in
1378 * Return 1 if able to complete the job; otherwise 0.
1381 m_append(struct mbuf *m0, int len, c_caddr_t cp)
1384 int remainder, space;
1386 for (m = m0; m->m_next != NULL; m = m->m_next)
1389 space = M_TRAILINGSPACE(m);
1392 * Copy into available space.
1394 if (space > remainder)
1396 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1398 cp += space, remainder -= space;
1400 while (remainder > 0) {
1402 * Allocate a new mbuf; could check space
1403 * and allocate a cluster instead.
1405 n = m_get(M_NOWAIT, m->m_type);
1408 n->m_len = min(MLEN, remainder);
1409 bcopy(cp, mtod(n, caddr_t), n->m_len);
1410 cp += n->m_len, remainder -= n->m_len;
1414 if (m0->m_flags & M_PKTHDR)
1415 m0->m_pkthdr.len += len - remainder;
1416 return (remainder == 0);
1420 * Apply function f to the data in an mbuf chain starting "off" bytes from
1421 * the beginning, continuing for "len" bytes.
1424 m_apply(struct mbuf *m, int off, int len,
1425 int (*f)(void *, void *, u_int), void *arg)
1430 KASSERT(off >= 0, ("m_apply, negative off %d", off));
1431 KASSERT(len >= 0, ("m_apply, negative len %d", len));
1433 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1440 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1441 count = min(m->m_len - off, len);
1442 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1453 * Return a pointer to mbuf/offset of location in mbuf chain.
1456 m_getptr(struct mbuf *m, int loc, int *off)
1460 /* Normal end of search. */
1461 if (m->m_len > loc) {
1466 if (m->m_next == NULL) {
1468 /* Point at the end of valid data. */
1481 m_print(const struct mbuf *m, int maxlen)
1485 const struct mbuf *m2;
1488 printf("mbuf: %p\n", m);
1492 if (m->m_flags & M_PKTHDR)
1493 len = m->m_pkthdr.len;
1497 while (m2 != NULL && (len == -1 || len)) {
1499 if (maxlen != -1 && pdata > maxlen)
1501 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
1502 m2->m_next, m2->m_flags, "\20\20freelist\17skipfw"
1503 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
1504 "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
1506 printf(", %*D\n", pdata, (u_char *)m2->m_data, "-");
1512 printf("%d bytes unaccounted for.\n", len);
1517 m_fixhdr(struct mbuf *m0)
1521 len = m_length(m0, NULL);
1522 m0->m_pkthdr.len = len;
1527 m_length(struct mbuf *m0, struct mbuf **last)
1533 for (m = m0; m != NULL; m = m->m_next) {
1535 if (m->m_next == NULL)
1544 * Defragment a mbuf chain, returning the shortest possible
1545 * chain of mbufs and clusters. If allocation fails and
1546 * this cannot be completed, NULL will be returned, but
1547 * the passed in chain will be unchanged. Upon success,
1548 * the original chain will be freed, and the new chain
1551 * If a non-packet header is passed in, the original
1552 * mbuf (chain?) will be returned unharmed.
1555 m_defrag(struct mbuf *m0, int how)
1557 struct mbuf *m_new = NULL, *m_final = NULL;
1558 int progress = 0, length;
1560 MBUF_CHECKSLEEP(how);
1561 if (!(m0->m_flags & M_PKTHDR))
1564 m_fixhdr(m0); /* Needed sanity check */
1566 #ifdef MBUF_STRESS_TEST
1567 if (m_defragrandomfailures) {
1568 int temp = arc4random() & 0xff;
1574 if (m0->m_pkthdr.len > MHLEN)
1575 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1577 m_final = m_gethdr(how, MT_DATA);
1579 if (m_final == NULL)
1582 if (m_dup_pkthdr(m_final, m0, how) == 0)
1587 while (progress < m0->m_pkthdr.len) {
1588 length = m0->m_pkthdr.len - progress;
1589 if (length > MCLBYTES)
1592 if (m_new == NULL) {
1594 m_new = m_getcl(how, MT_DATA, 0);
1596 m_new = m_get(how, MT_DATA);
1601 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1603 m_new->m_len = length;
1604 if (m_new != m_final)
1605 m_cat(m_final, m_new);
1608 #ifdef MBUF_STRESS_TEST
1609 if (m0->m_next == NULL)
1614 #ifdef MBUF_STRESS_TEST
1616 m_defragbytes += m0->m_pkthdr.len;
1620 #ifdef MBUF_STRESS_TEST
1629 * Defragment an mbuf chain, returning at most maxfrags separate
1630 * mbufs+clusters. If this is not possible NULL is returned and
1631 * the original mbuf chain is left in it's present (potentially
1632 * modified) state. We use two techniques: collapsing consecutive
1633 * mbufs and replacing consecutive mbufs by a cluster.
1635 * NB: this should really be named m_defrag but that name is taken
1638 m_collapse(struct mbuf *m0, int how, int maxfrags)
1640 struct mbuf *m, *n, *n2, **prev;
1644 * Calculate the current number of frags.
1647 for (m = m0; m != NULL; m = m->m_next)
1650 * First, try to collapse mbufs. Note that we always collapse
1651 * towards the front so we don't need to deal with moving the
1652 * pkthdr. This may be suboptimal if the first mbuf has much
1653 * less data than the following.
1661 if (M_WRITABLE(m) &&
1662 n->m_len < M_TRAILINGSPACE(m)) {
1663 bcopy(mtod(n, void *), mtod(m, char *) + m->m_len,
1665 m->m_len += n->m_len;
1666 m->m_next = n->m_next;
1668 if (--curfrags <= maxfrags)
1673 KASSERT(maxfrags > 1,
1674 ("maxfrags %u, but normal collapse failed", maxfrags));
1676 * Collapse consecutive mbufs to a cluster.
1678 prev = &m0->m_next; /* NB: not the first mbuf */
1679 while ((n = *prev) != NULL) {
1680 if ((n2 = n->m_next) != NULL &&
1681 n->m_len + n2->m_len < MCLBYTES) {
1682 m = m_getcl(how, MT_DATA, 0);
1685 bcopy(mtod(n, void *), mtod(m, void *), n->m_len);
1686 bcopy(mtod(n2, void *), mtod(m, char *) + n->m_len,
1688 m->m_len = n->m_len + n2->m_len;
1689 m->m_next = n2->m_next;
1693 if (--curfrags <= maxfrags) /* +1 cl -2 mbufs */
1696 * Still not there, try the normal collapse
1697 * again before we allocate another cluster.
1704 * No place where we can collapse to a cluster; punt.
1705 * This can occur if, for example, you request 2 frags
1706 * but the packet requires that both be clusters (we
1707 * never reallocate the first mbuf to avoid moving the
1714 #ifdef MBUF_STRESS_TEST
1717 * Fragment an mbuf chain. There's no reason you'd ever want to do
1718 * this in normal usage, but it's great for stress testing various
1721 * If fragmentation is not possible, the original chain will be
1724 * Possible length values:
1725 * 0 no fragmentation will occur
1726 * > 0 each fragment will be of the specified length
1727 * -1 each fragment will be the same random value in length
1728 * -2 each fragment's length will be entirely random
1729 * (Random values range from 1 to 256)
1732 m_fragment(struct mbuf *m0, int how, int length)
1734 struct mbuf *m_new = NULL, *m_final = NULL;
1737 if (!(m0->m_flags & M_PKTHDR))
1740 if ((length == 0) || (length < -2))
1743 m_fixhdr(m0); /* Needed sanity check */
1745 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1747 if (m_final == NULL)
1750 if (m_dup_pkthdr(m_final, m0, how) == 0)
1756 length = 1 + (arc4random() & 255);
1758 while (progress < m0->m_pkthdr.len) {
1764 fraglen = 1 + (arc4random() & 255);
1765 if (fraglen > m0->m_pkthdr.len - progress)
1766 fraglen = m0->m_pkthdr.len - progress;
1768 if (fraglen > MCLBYTES)
1771 if (m_new == NULL) {
1772 m_new = m_getcl(how, MT_DATA, 0);
1777 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t));
1778 progress += fraglen;
1779 m_new->m_len = fraglen;
1780 if (m_new != m_final)
1781 m_cat(m_final, m_new);
1790 /* Return the original chain on failure */
1797 * Copy the contents of uio into a properly sized mbuf chain.
1800 m_uiotombuf(struct uio *uio, int how, int len, int align, int flags)
1802 struct mbuf *m, *mb;
1808 * len can be zero or an arbitrary large value bound by
1809 * the total data supplied by the uio.
1812 total = min(uio->uio_resid, len);
1814 total = uio->uio_resid;
1817 * The smallest unit returned by m_getm2() is a single mbuf
1818 * with pkthdr. We can't align past it.
1824 * Give us the full allocation or nothing.
1825 * If len is zero return the smallest empty mbuf.
1827 m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags);
1832 /* Fill all mbufs with uio data and update header information. */
1833 for (mb = m; mb != NULL; mb = mb->m_next) {
1834 length = min(M_TRAILINGSPACE(mb), total - progress);
1836 error = uiomove(mtod(mb, void *), length, uio);
1844 if (flags & M_PKTHDR)
1845 m->m_pkthdr.len += length;
1847 KASSERT(progress == total, ("%s: progress != total", __func__));
1853 * Copy an mbuf chain into a uio limited by len if set.
1856 m_mbuftouio(struct uio *uio, struct mbuf *m, int len)
1858 int error, length, total;
1862 total = min(uio->uio_resid, len);
1864 total = uio->uio_resid;
1866 /* Fill the uio with data from the mbufs. */
1867 for (; m != NULL; m = m->m_next) {
1868 length = min(m->m_len, total - progress);
1870 error = uiomove(mtod(m, void *), length, uio);
1881 * Set the m_data pointer of a newly-allocated mbuf
1882 * to place an object of the specified size at the
1883 * end of the mbuf, longword aligned.
1886 m_align(struct mbuf *m, int len)
1889 const char *msg = "%s: not a virgin mbuf";
1893 if (m->m_flags & M_EXT) {
1894 KASSERT(m->m_data == m->m_ext.ext_buf, (msg, __func__));
1895 adjust = m->m_ext.ext_size - len;
1896 } else if (m->m_flags & M_PKTHDR) {
1897 KASSERT(m->m_data == m->m_pktdat, (msg, __func__));
1898 adjust = MHLEN - len;
1900 KASSERT(m->m_data == m->m_dat, (msg, __func__));
1901 adjust = MLEN - len;
1904 m->m_data += adjust &~ (sizeof(long)-1);
1908 * Create a writable copy of the mbuf chain. While doing this
1909 * we compact the chain with a goal of producing a chain with
1910 * at most two mbufs. The second mbuf in this chain is likely
1911 * to be a cluster. The primary purpose of this work is to create
1912 * a writable packet for encryption, compression, etc. The
1913 * secondary goal is to linearize the data so the data can be
1914 * passed to crypto hardware in the most efficient manner possible.
1917 m_unshare(struct mbuf *m0, int how)
1919 struct mbuf *m, *mprev;
1920 struct mbuf *n, *mfirst, *mlast;
1924 for (m = m0; m != NULL; m = mprev->m_next) {
1926 * Regular mbufs are ignored unless there's a cluster
1927 * in front of it that we can use to coalesce. We do
1928 * the latter mainly so later clusters can be coalesced
1929 * also w/o having to handle them specially (i.e. convert
1930 * mbuf+cluster -> cluster). This optimization is heavily
1931 * influenced by the assumption that we're running over
1932 * Ethernet where MCLBYTES is large enough that the max
1933 * packet size will permit lots of coalescing into a
1934 * single cluster. This in turn permits efficient
1935 * crypto operations, especially when using hardware.
1937 if ((m->m_flags & M_EXT) == 0) {
1938 if (mprev && (mprev->m_flags & M_EXT) &&
1939 m->m_len <= M_TRAILINGSPACE(mprev)) {
1940 /* XXX: this ignores mbuf types */
1941 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1942 mtod(m, caddr_t), m->m_len);
1943 mprev->m_len += m->m_len;
1944 mprev->m_next = m->m_next; /* unlink from chain */
1945 m_free(m); /* reclaim mbuf */
1947 newipsecstat.ips_mbcoalesced++;
1955 * Writable mbufs are left alone (for now).
1957 if (M_WRITABLE(m)) {
1963 * Not writable, replace with a copy or coalesce with
1964 * the previous mbuf if possible (since we have to copy
1965 * it anyway, we try to reduce the number of mbufs and
1966 * clusters so that future work is easier).
1968 KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags));
1969 /* NB: we only coalesce into a cluster or larger */
1970 if (mprev != NULL && (mprev->m_flags & M_EXT) &&
1971 m->m_len <= M_TRAILINGSPACE(mprev)) {
1972 /* XXX: this ignores mbuf types */
1973 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1974 mtod(m, caddr_t), m->m_len);
1975 mprev->m_len += m->m_len;
1976 mprev->m_next = m->m_next; /* unlink from chain */
1977 m_free(m); /* reclaim mbuf */
1979 newipsecstat.ips_clcoalesced++;
1985 * Allocate new space to hold the copy and copy the data.
1986 * We deal with jumbo mbufs (i.e. m_len > MCLBYTES) by
1987 * splitting them into clusters. We could just malloc a
1988 * buffer and make it external but too many device drivers
1989 * don't know how to break up the non-contiguous memory when
1992 n = m_getcl(how, m->m_type, m->m_flags & M_COPYFLAGS);
2002 int cc = min(len, MCLBYTES);
2003 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc);
2009 newipsecstat.ips_clcopied++;
2017 n = m_getcl(how, m->m_type, m->m_flags & M_COPYFLAGS);
2024 n->m_next = m->m_next;
2026 m0 = mfirst; /* new head of chain */
2028 mprev->m_next = mfirst; /* replace old mbuf */
2029 m_free(m); /* release old mbuf */
2035 #ifdef MBUF_PROFILING
2037 #define MP_BUCKETS 32 /* don't just change this as things may overflow.*/
2038 struct mbufprofile {
2039 uintmax_t wasted[MP_BUCKETS];
2040 uintmax_t used[MP_BUCKETS];
2041 uintmax_t segments[MP_BUCKETS];
2044 #define MP_MAXDIGITS 21 /* strlen("16,000,000,000,000,000,000") == 21 */
2045 #define MP_NUMLINES 6
2046 #define MP_NUMSPERLINE 16
2047 #define MP_EXTRABYTES 64 /* > strlen("used:\nwasted:\nsegments:\n") */
2048 /* work out max space needed and add a bit of spare space too */
2049 #define MP_MAXLINE ((MP_MAXDIGITS+1) * MP_NUMSPERLINE)
2050 #define MP_BUFSIZE ((MP_MAXLINE * MP_NUMLINES) + 1 + MP_EXTRABYTES)
2052 char mbprofbuf[MP_BUFSIZE];
2055 m_profile(struct mbuf *m)
2064 if (m->m_flags & M_EXT) {
2065 wasted += MHLEN - sizeof(m->m_ext) +
2066 m->m_ext.ext_size - m->m_len;
2068 if (m->m_flags & M_PKTHDR)
2069 wasted += MHLEN - m->m_len;
2071 wasted += MLEN - m->m_len;
2075 /* be paranoid.. it helps */
2076 if (segments > MP_BUCKETS - 1)
2077 segments = MP_BUCKETS - 1;
2080 if (wasted > 100000)
2082 /* store in the appropriate bucket */
2083 /* don't bother locking. if it's slightly off, so what? */
2084 mbprof.segments[segments]++;
2085 mbprof.used[fls(used)]++;
2086 mbprof.wasted[fls(wasted)]++;
2090 mbprof_textify(void)
2097 p = &mbprof.wasted[0];
2099 offset = snprintf(c, MP_MAXLINE + 10,
2101 "%ju %ju %ju %ju %ju %ju %ju %ju "
2102 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2103 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2104 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2106 p = &mbprof.wasted[16];
2108 offset = snprintf(c, MP_MAXLINE,
2109 "%ju %ju %ju %ju %ju %ju %ju %ju "
2110 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2111 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2112 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2114 p = &mbprof.used[0];
2116 offset = snprintf(c, MP_MAXLINE + 10,
2118 "%ju %ju %ju %ju %ju %ju %ju %ju "
2119 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2120 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2121 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2123 p = &mbprof.used[16];
2125 offset = snprintf(c, MP_MAXLINE,
2126 "%ju %ju %ju %ju %ju %ju %ju %ju "
2127 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2128 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2129 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2131 p = &mbprof.segments[0];
2133 offset = snprintf(c, MP_MAXLINE + 10,
2135 "%ju %ju %ju %ju %ju %ju %ju %ju "
2136 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2137 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2138 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2140 p = &mbprof.segments[16];
2142 offset = snprintf(c, MP_MAXLINE,
2143 "%ju %ju %ju %ju %ju %ju %ju %ju "
2144 "%ju %ju %ju %ju %ju %ju %ju %jju",
2145 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2146 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2151 mbprof_handler(SYSCTL_HANDLER_ARGS)
2156 error = SYSCTL_OUT(req, mbprofbuf, strlen(mbprofbuf) + 1);
2161 mbprof_clr_handler(SYSCTL_HANDLER_ARGS)
2166 error = sysctl_handle_int(oidp, &clear, 0, req);
2167 if (error || !req->newptr)
2171 bzero(&mbprof, sizeof(mbprof));
2178 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofile, CTLTYPE_STRING|CTLFLAG_RD,
2179 NULL, 0, mbprof_handler, "A", "mbuf profiling statistics");
2181 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofileclr, CTLTYPE_INT|CTLFLAG_RW,
2182 NULL, 0, mbprof_clr_handler, "I", "clear mbuf profiling statistics");