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 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94
34 * $FreeBSD: src/sys/kern/uipc_mbuf.c,v 1.51.2.24 2003/04/15 06:59:29 silby Exp $
37 #include "opt_param.h"
38 #include "opt_mbuf_stress_test.h"
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
45 #include <sys/domain.h>
46 #include <sys/protosw.h>
49 #include <vm/vm_kern.h>
50 #include <vm/vm_extern.h>
53 #include <machine/cpu.h>
56 static void mbinit __P((void *));
57 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbinit, NULL)
62 u_long mbtypes[MT_NTYPES];
64 union mcluster *mclfree;
73 #ifdef MBUF_STRESS_TEST
74 int m_defragrandomfailures;
79 u_int m_mballoc_wid = 0;
80 u_int m_clalloc_wid = 0;
82 SYSCTL_DECL(_kern_ipc);
83 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW,
85 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW,
86 &max_protohdr, 0, "");
87 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, "");
88 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW,
90 SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW,
92 SYSCTL_STRUCT(_kern_ipc, KIPC_MBSTAT, mbstat, CTLFLAG_RW, &mbstat, mbstat, "");
93 SYSCTL_OPAQUE(_kern_ipc, OID_AUTO, mbtypes, CTLFLAG_RD, mbtypes,
94 sizeof(mbtypes), "LU", "");
95 SYSCTL_INT(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLFLAG_RD,
96 &nmbclusters, 0, "Maximum number of mbuf clusters available");
97 SYSCTL_INT(_kern_ipc, OID_AUTO, nmbufs, CTLFLAG_RD, &nmbufs, 0,
98 "Maximum number of mbufs available");
99 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
100 &m_defragpackets, 0, "");
101 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
102 &m_defragbytes, 0, "");
103 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
104 &m_defraguseless, 0, "");
105 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
106 &m_defragfailure, 0, "");
107 #ifdef MBUF_STRESS_TEST
108 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
109 &m_defragrandomfailures, 0, "");
112 static void m_reclaim __P((void));
115 #define NMBCLUSTERS (512 + maxusers * 16)
118 #define NMBUFS (nmbclusters * 4)
122 * Perform sanity checks of tunables declared above.
125 tunable_mbinit(void *dummy)
129 * This has to be done before VM init.
131 nmbclusters = NMBCLUSTERS;
132 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
134 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs);
136 if (nmbufs < nmbclusters * 2)
137 nmbufs = nmbclusters * 2;
141 SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_mbinit, NULL);
143 /* "number of clusters of pages" */
155 mmbfree = NULL; mclfree = NULL;
156 mbstat.m_msize = MSIZE;
157 mbstat.m_mclbytes = MCLBYTES;
158 mbstat.m_minclsize = MINCLSIZE;
159 mbstat.m_mlen = MLEN;
160 mbstat.m_mhlen = MHLEN;
163 if (m_mballoc(NMB_INIT, M_DONTWAIT) == 0)
165 #if MCLBYTES <= PAGE_SIZE
166 if (m_clalloc(NCL_INIT, M_DONTWAIT) == 0)
169 /* It's OK to call contigmalloc in this context. */
170 if (m_clalloc(16, M_WAIT) == 0)
180 * Allocate at least nmb mbufs and place on mbuf free list.
181 * Must be called at splimp.
194 * If we've hit the mbuf limit, stop allocating from mb_map,
195 * (or trying to) in order to avoid dipping into the section of
196 * mb_map which we've "reserved" for clusters.
198 if ((nmb + mbstat.m_mbufs) > nmbufs)
202 * Once we run out of map space, it will be impossible to get
203 * any more (nothing is ever freed back to the map)
204 * -- however you are not dead as m_reclaim might
205 * still be able to free a substantial amount of space.
207 * XXX Furthermore, we can also work with "recycled" mbufs (when
208 * we're calling with M_WAIT the sleep procedure will be woken
209 * up when an mbuf is freed. See m_mballoc_wait()).
214 nbytes = round_page(nmb * MSIZE);
215 p = (caddr_t)kmem_malloc(mb_map, nbytes, M_NOWAIT);
216 if (p == 0 && how == M_WAIT) {
218 p = (caddr_t)kmem_malloc(mb_map, nbytes, M_WAITOK);
222 * Either the map is now full, or `how' is M_NOWAIT and there
228 nmb = nbytes / MSIZE;
229 for (i = 0; i < nmb; i++) {
230 ((struct mbuf *)p)->m_next = mmbfree;
231 mmbfree = (struct mbuf *)p;
234 mbstat.m_mbufs += nmb;
235 mbtypes[MT_FREE] += nmb;
240 * Once the mb_map has been exhausted and if the call to the allocation macros
241 * (or, in some cases, functions) is with M_WAIT, then it is necessary to rely
242 * solely on reclaimed mbufs. Here we wait for an mbuf to be freed for a
243 * designated (mbuf_wait) time.
246 m_mballoc_wait(int caller, int type)
253 if ((tsleep(&m_mballoc_wid, PVM, "mballc", mbuf_wait)) == EWOULDBLOCK)
258 * Now that we (think) that we've got something, we will redo an
259 * MGET, but avoid getting into another instance of m_mballoc_wait()
260 * XXX: We retry to fetch _even_ if the sleep timed out. This is left
261 * this way, purposely, in the [unlikely] case that an mbuf was
262 * freed but the sleep was not awakened in time.
267 MGET(p, M_DONTWAIT, type);
270 MGETHDR(p, M_DONTWAIT, type);
273 panic("m_mballoc_wait: invalid caller (%d)", caller);
277 if (p != NULL) { /* We waited and got something... */
279 /* Wake up another if we have more free. */
287 #if MCLBYTES > PAGE_SIZE
288 static int i_want_my_mcl;
296 tsleep(&i_want_my_mcl, PVM, "mclalloc", 0);
298 for (; i_want_my_mcl; i_want_my_mcl--) {
299 if (m_clalloc(1, M_WAIT) == 0)
300 printf("m_clalloc failed even in process context!\n");
305 static struct proc *mclallocproc;
306 static struct kproc_desc mclalloc_kp = {
311 SYSINIT(mclallocproc, SI_SUB_KTHREAD_UPDATE, SI_ORDER_ANY, kproc_start,
316 * Allocate some number of mbuf clusters
317 * and place on cluster free list.
318 * Must be called at splimp.
331 * If we've hit the mcluster number limit, stop allocating from
332 * mb_map, (or trying to) in order to avoid dipping into the section
333 * of mb_map which we've "reserved" for mbufs.
335 if ((ncl + mbstat.m_clusters) > nmbclusters)
339 * Once we run out of map space, it will be impossible
340 * to get any more (nothing is ever freed back to the
341 * map). From this point on, we solely rely on freed
347 #if MCLBYTES > PAGE_SIZE
349 i_want_my_mcl += ncl;
350 wakeup(&i_want_my_mcl);
354 p = contigmalloc1(MCLBYTES * ncl, M_DEVBUF, M_WAITOK, 0ul,
355 ~0ul, PAGE_SIZE, 0, mb_map);
359 p = (caddr_t)kmem_malloc(mb_map, ctob(npg),
360 how != M_WAIT ? M_NOWAIT : M_WAITOK);
361 ncl = ncl * PAGE_SIZE / MCLBYTES;
364 * Either the map is now full, or `how' is M_NOWAIT and there
368 static int last_report ; /* when we did that (in ticks) */
371 if (ticks < last_report || (ticks - last_report) >= hz) {
373 printf("All mbuf clusters exhausted, please see tuning(7).\n");
378 for (i = 0; i < ncl; i++) {
379 ((union mcluster *)p)->mcl_next = mclfree;
380 mclfree = (union mcluster *)p;
384 mbstat.m_clusters += ncl;
389 * Once the mb_map submap has been exhausted and the allocation is called with
390 * M_WAIT, we rely on the mclfree union pointers. If nothing is free, we will
391 * sleep for a designated amount of time (mbuf_wait) or until we're woken up
392 * due to sudden mcluster availability.
401 /* If in interrupt context, and INVARIANTS, maintain sanity and die. */
402 KASSERT(intr_nesting_level == 0, ("CLALLOC: CANNOT WAIT IN INTERRUPT"));
405 /* Sleep until something's available or until we expire. */
407 if ((tsleep(&m_clalloc_wid, PVM, "mclalc", mbuf_wait)) == EWOULDBLOCK)
411 * Now that we (think) that we've got something, we will redo and
412 * MGET, but avoid getting into another instance of m_clalloc_wait()
415 MCLALLOC(p, M_DONTWAIT);
418 if (p != NULL) { /* We waited and got something... */
420 /* Wake up another if we have more free. */
430 * When MGET fails, ask protocols to free space when short of memory,
431 * then re-attempt to allocate an mbuf.
437 register struct mbuf *m;
440 * Must only do the reclaim if not in an interrupt context.
444 KASSERT(intr_nesting_level == 0,
445 ("MBALLOC: CANNOT WAIT IN INTERRUPT"));
451 * Both m_mballoc_wait and m_retry must be nulled because
452 * when the MGET macro is run from here, we deffinately do _not_
453 * want to enter an instance of m_mballoc_wait() or m_retry() (again!)
455 #define m_mballoc_wait(caller,type) (struct mbuf *)0
456 #define m_retry(i, t) (struct mbuf *)0
459 #undef m_mballoc_wait
464 static int last_report ; /* when we did that (in ticks) */
466 if (ticks < last_report || (ticks - last_report) >= hz) {
468 printf("All mbufs exhausted, please see tuning(7).\n");
476 * As above; retry an MGETHDR.
482 register struct mbuf *m;
485 * Must only do the reclaim if not in an interrupt context.
489 KASSERT(intr_nesting_level == 0,
490 ("MBALLOC: CANNOT WAIT IN INTERRUPT"));
495 #define m_mballoc_wait(caller,type) (struct mbuf *)0
496 #define m_retryhdr(i, t) (struct mbuf *)0
499 #undef m_mballoc_wait
504 static int last_report ; /* when we did that (in ticks) */
506 if (ticks < last_report || (ticks - last_report) >= hz) {
508 printf("All mbufs exhausted, please see tuning(7).\n");
518 register struct domain *dp;
519 register struct protosw *pr;
522 for (dp = domains; dp; dp = dp->dom_next)
523 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
531 * Space allocation routines.
532 * These are also available as macros
533 * for critical paths.
539 register struct mbuf *m;
549 register struct mbuf *m;
551 MGETHDR(m, how, type);
559 register struct mbuf *m;
564 bzero(mtod(m, caddr_t), MLEN);
569 * m_getcl() returns an mbuf with an attached cluster.
570 * Because many network drivers use this kind of buffers a lot, it is
571 * convenient to keep a small pool of free buffers of this kind.
572 * Even a small size such as 10 gives about 10% improvement in the
573 * forwarding rate in a bridge or router.
574 * The size of this free list is controlled by the sysctl variable
575 * mcl_pool_max. The list is populated on m_freem(), and used in
576 * m_getcl() if elements are available.
578 static struct mbuf *mcl_pool;
579 static int mcl_pool_now;
580 static int mcl_pool_max = 0;
582 SYSCTL_INT(_kern_ipc, OID_AUTO, mcl_pool_max, CTLFLAG_RW, &mcl_pool_max, 0,
583 "Maximum number of mbufs+cluster in free list");
584 SYSCTL_INT(_kern_ipc, OID_AUTO, mcl_pool_now, CTLFLAG_RD, &mcl_pool_now, 0,
585 "Current number of mbufs+cluster in free list");
588 m_getcl(int how, short type, int flags)
593 if (flags & M_PKTHDR) {
594 if (type == MT_DATA && mcl_pool) {
596 mcl_pool = mp->m_nextpkt;
599 mp->m_nextpkt = NULL;
600 mp->m_data = mp->m_ext.ext_buf;
601 mp->m_flags = M_PKTHDR|M_EXT;
602 mp->m_pkthdr.rcvif = NULL;
603 mp->m_pkthdr.csum_flags = 0;
606 MGETHDR(mp, how, type);
611 if ( (mp->m_flags & M_EXT) == 0) {
622 * m_getm(m, len, how, type)
624 * This will allocate len-worth of mbufs and/or mbuf clusters (whatever fits
625 * best) and return a pointer to the top of the allocated chain. If m is
626 * non-null, then we assume that it is a single mbuf or an mbuf chain to
627 * which we want len bytes worth of mbufs and/or clusters attached, and so
628 * if we succeed in allocating it, we will just return a pointer to m.
630 * If we happen to fail at any point during the allocation, we will free
631 * up everything we have already allocated and return NULL.
635 m_getm(struct mbuf *m, int len, int how, int type)
637 struct mbuf *top, *tail, *mp, *mtail = NULL;
639 KASSERT(len >= 0, ("len is < 0 in m_getm"));
644 else if (len > MINCLSIZE) {
646 if ((mp->m_flags & M_EXT) == 0) {
652 len -= M_TRAILINGSPACE(mp);
655 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next);
667 if (len > MINCLSIZE) {
669 if ((mp->m_flags & M_EXT) == 0)
674 len -= M_TRAILINGSPACE(mp);
687 * MFREE(struct mbuf *m, struct mbuf *n)
688 * Free a single mbuf and associated external storage.
689 * Place the successor, if any, in n.
691 * we do need to check non-first mbuf for m_aux, since some of existing
692 * code does not call M_PREPEND properly.
693 * (example: call to bpf_mtap from drivers)
695 #define MFREE(m, n) MBUFLOCK( \
696 struct mbuf *_mm = (m); \
698 KASSERT(_mm->m_type != MT_FREE, ("freeing free mbuf")); \
699 mbtypes[_mm->m_type]--; \
700 if ((_mm->m_flags & M_PKTHDR) != 0) \
701 m_tag_delete_chain(_mm, NULL); \
702 if (_mm->m_flags & M_EXT) \
705 _mm->m_type = MT_FREE; \
706 mbtypes[MT_FREE]++; \
707 _mm->m_next = mmbfree; \
716 register struct mbuf *n;
729 * Try to keep a small pool of mbuf+cluster for quick use in
730 * device drivers. A good candidate is a M_PKTHDR buffer with
731 * only one cluster attached. Other mbufs, or those exceeding
732 * the pool size, are just m_free'd in the usual way.
733 * The following code makes sure that m_next, m_type,
734 * m_pkthdr.aux and m_ext.* are properly initialized.
735 * Other fields in the mbuf are initialized in m_getcl()
738 if (mcl_pool_now < mcl_pool_max && m && m->m_next == NULL &&
739 (m->m_flags & (M_PKTHDR|M_EXT)) == (M_PKTHDR|M_EXT) &&
740 m->m_type == MT_DATA && M_EXT_WRITABLE(m) ) {
741 m_tag_delete_chain(m, NULL);
742 m->m_nextpkt = mcl_pool;
753 * Mbuffer utility routines.
757 * Lesser-used path for M_PREPEND:
758 * allocate new mbuf to prepend to chain,
762 m_prepend(m, len, how)
763 register struct mbuf *m;
768 MGET(mn, how, m->m_type);
769 if (mn == (struct mbuf *)NULL) {
771 return ((struct mbuf *)NULL);
773 if (m->m_flags & M_PKTHDR)
774 M_MOVE_PKTHDR(mn, m);
784 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
785 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
786 * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
787 * Note that the copy is read-only, because clusters are not copied,
788 * only their reference counts are incremented.
790 #define MCFail (mbstat.m_mcfail)
793 m_copym(m, off0, len, wait)
794 register struct mbuf *m;
798 register struct mbuf *n, **np;
799 register int off = off0;
803 KASSERT(off >= 0, ("m_copym, negative off %d", off));
804 KASSERT(len >= 0, ("m_copym, negative len %d", len));
805 if (off == 0 && m->m_flags & M_PKTHDR)
808 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
818 KASSERT(len == M_COPYALL,
819 ("m_copym, length > size of mbuf chain"));
822 MGET(n, wait, m->m_type);
827 if (!m_dup_pkthdr(n, m, wait))
829 if (len == M_COPYALL)
830 n->m_pkthdr.len -= off0;
832 n->m_pkthdr.len = len;
835 n->m_len = min(len, m->m_len - off);
836 if (m->m_flags & M_EXT) {
837 n->m_data = m->m_data + off;
838 if (m->m_ext.ext_ref == NULL) {
840 &mclrefcnt[mtocl(m->m_ext.ext_buf)], 1);
844 (*m->m_ext.ext_ref)(m->m_ext.ext_buf,
851 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
853 if (len != M_COPYALL)
869 * Copy an entire packet, including header (which must be present).
870 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
871 * Note that the copy is read-only, because clusters are not copied,
872 * only their reference counts are incremented.
873 * Preserve alignment of the first mbuf so if the creator has left
874 * some room at the beginning (e.g. for inserting protocol headers)
875 * the copies also have the room available.
882 struct mbuf *top, *n, *o;
884 MGET(n, how, m->m_type);
889 if (!m_dup_pkthdr(n, m, how))
892 if (m->m_flags & M_EXT) {
893 n->m_data = m->m_data;
894 if (m->m_ext.ext_ref == NULL)
895 atomic_add_char(&mclrefcnt[mtocl(m->m_ext.ext_buf)], 1);
899 (*m->m_ext.ext_ref)(m->m_ext.ext_buf,
906 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
907 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
912 MGET(o, how, m->m_type);
920 if (m->m_flags & M_EXT) {
921 n->m_data = m->m_data;
922 if (m->m_ext.ext_ref == NULL) {
924 &mclrefcnt[mtocl(m->m_ext.ext_buf)], 1);
928 (*m->m_ext.ext_ref)(m->m_ext.ext_buf,
935 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
948 * Copy data from an mbuf chain starting "off" bytes from the beginning,
949 * continuing for "len" bytes, into the indicated buffer.
952 m_copydata(m, off, len, cp)
953 register struct mbuf *m;
958 register unsigned count;
960 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
961 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
963 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
970 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
971 count = min(m->m_len - off, len);
972 bcopy(mtod(m, caddr_t) + off, cp, count);
981 * Copy a packet header mbuf chain into a completely new chain, including
982 * copying any mbuf clusters. Use this instead of m_copypacket() when
983 * you need a writable copy of an mbuf chain.
990 struct mbuf **p, *top = NULL;
991 int remain, moff, nsize;
996 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __FUNCTION__));
998 /* While there's more data, get a new mbuf, tack it on, and fill it */
999 remain = m->m_pkthdr.len;
1002 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
1005 /* Get the next new mbuf */
1006 MGET(n, how, m->m_type);
1009 if (top == NULL) { /* first one, must be PKTHDR */
1010 if (!m_dup_pkthdr(n, m, how))
1013 } else /* not the first one */
1015 if (remain >= MINCLSIZE) {
1017 if ((n->m_flags & M_EXT) == 0) {
1025 /* Link it into the new chain */
1029 /* Copy data from original mbuf(s) into new mbuf */
1030 while (n->m_len < nsize && m != NULL) {
1031 int chunk = min(nsize - n->m_len, m->m_len - moff);
1033 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
1037 if (moff == m->m_len) {
1043 /* Check correct total mbuf length */
1044 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
1045 ("%s: bogus m_pkthdr.len", __FUNCTION__));
1056 * Concatenate mbuf chain n to m.
1057 * Both chains must be of the same type (e.g. MT_DATA).
1058 * Any m_pkthdr is not updated.
1062 register struct mbuf *m, *n;
1067 if (m->m_flags & M_EXT ||
1068 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
1069 /* just join the two chains */
1073 /* splat the data from one into the other */
1074 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1076 m->m_len += n->m_len;
1086 register int len = req_len;
1087 register struct mbuf *m;
1090 if ((m = mp) == NULL)
1096 while (m != NULL && len > 0) {
1097 if (m->m_len <= len) {
1108 if (mp->m_flags & M_PKTHDR)
1109 m->m_pkthdr.len -= (req_len - len);
1112 * Trim from tail. Scan the mbuf chain,
1113 * calculating its length and finding the last mbuf.
1114 * If the adjustment only affects this mbuf, then just
1115 * adjust and return. Otherwise, rescan and truncate
1116 * after the remaining size.
1122 if (m->m_next == (struct mbuf *)0)
1126 if (m->m_len >= len) {
1128 if (mp->m_flags & M_PKTHDR)
1129 mp->m_pkthdr.len -= len;
1136 * Correct length for chain is "count".
1137 * Find the mbuf with last data, adjust its length,
1138 * and toss data from remaining mbufs on chain.
1141 if (m->m_flags & M_PKTHDR)
1142 m->m_pkthdr.len = count;
1143 for (; m; m = m->m_next) {
1144 if (m->m_len >= count) {
1151 (m = m->m_next) ->m_len = 0;
1156 * Rearange an mbuf chain so that len bytes are contiguous
1157 * and in the data area of an mbuf (so that mtod and dtom
1158 * will work for a structure of size len). Returns the resulting
1159 * mbuf chain on success, frees it and returns null on failure.
1160 * If there is room, it will add up to max_protohdr-len extra bytes to the
1161 * contiguous region in an attempt to avoid being called next time.
1163 #define MPFail (mbstat.m_mpfail)
1167 register struct mbuf *n;
1170 register struct mbuf *m;
1175 * If first mbuf has no cluster, and has room for len bytes
1176 * without shifting current data, pullup into it,
1177 * otherwise allocate a new mbuf to prepend to the chain.
1179 if ((n->m_flags & M_EXT) == 0 &&
1180 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
1181 if (n->m_len >= len)
1189 MGET(m, M_DONTWAIT, n->m_type);
1193 if (n->m_flags & M_PKTHDR)
1194 M_MOVE_PKTHDR(m, n);
1196 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1198 count = min(min(max(len, max_protohdr), space), n->m_len);
1199 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1209 } while (len > 0 && n);
1223 * Partition an mbuf chain in two pieces, returning the tail --
1224 * all but the first len0 bytes. In case of failure, it returns NULL and
1225 * attempts to restore the chain to its original state.
1227 * Note that the resulting mbufs might be read-only, because the new
1228 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1229 * the "breaking point" happens to lie within a cluster mbuf. Use the
1230 * M_WRITABLE() macro to check for this case.
1233 m_split(m0, len0, wait)
1234 register struct mbuf *m0;
1237 register struct mbuf *m, *n;
1238 unsigned len = len0, remain;
1240 for (m = m0; m && len > m->m_len; m = m->m_next)
1244 remain = m->m_len - len;
1245 if (m0->m_flags & M_PKTHDR) {
1246 MGETHDR(n, wait, m0->m_type);
1249 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1250 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1251 m0->m_pkthdr.len = len0;
1252 if (m->m_flags & M_EXT)
1254 if (remain > MHLEN) {
1255 /* m can't be the lead packet */
1257 n->m_next = m_split(m, len, wait);
1258 if (n->m_next == 0) {
1266 MH_ALIGN(n, remain);
1267 } else if (remain == 0) {
1272 MGET(n, wait, m->m_type);
1278 if (m->m_flags & M_EXT) {
1279 n->m_flags |= M_EXT;
1280 n->m_ext = m->m_ext;
1281 if (m->m_ext.ext_ref == NULL)
1282 atomic_add_char(&mclrefcnt[mtocl(m->m_ext.ext_buf)], 1);
1286 (*m->m_ext.ext_ref)(m->m_ext.ext_buf,
1290 n->m_data = m->m_data + len;
1292 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1296 n->m_next = m->m_next;
1301 * Routine to copy from device local memory into mbufs.
1304 m_devget(buf, totlen, off0, ifp, copy)
1308 void (*copy) __P((char *from, caddr_t to, u_int len));
1310 register struct mbuf *m;
1311 struct mbuf *top = 0, **mp = ⊤
1312 register int off = off0, len;
1319 cp += off + 2 * sizeof(u_short);
1320 totlen -= 2 * sizeof(u_short);
1322 MGETHDR(m, M_DONTWAIT, MT_DATA);
1325 m->m_pkthdr.rcvif = ifp;
1326 m->m_pkthdr.len = totlen;
1329 while (totlen > 0) {
1331 MGET(m, M_DONTWAIT, MT_DATA);
1338 len = min(totlen, epkt - cp);
1339 if (len >= MINCLSIZE) {
1340 MCLGET(m, M_DONTWAIT);
1341 if (m->m_flags & M_EXT)
1342 m->m_len = len = min(len, MCLBYTES);
1347 * Place initial small packet/header at end of mbuf.
1349 if (len < m->m_len) {
1350 if (top == 0 && len + max_linkhdr <= m->m_len)
1351 m->m_data += max_linkhdr;
1357 copy(cp, mtod(m, caddr_t), (unsigned)len);
1359 bcopy(cp, mtod(m, caddr_t), (unsigned)len);
1371 * Copy data from a buffer back into the indicated mbuf chain,
1372 * starting "off" bytes from the beginning, extending the mbuf
1373 * chain if necessary.
1376 m_copyback(m0, off, len, cp)
1383 register struct mbuf *m = m0, *n;
1388 while (off > (mlen = m->m_len)) {
1391 if (m->m_next == 0) {
1392 n = m_getclr(M_DONTWAIT, m->m_type);
1395 n->m_len = min(MLEN, len + off);
1401 mlen = min (m->m_len - off, len);
1402 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen);
1410 if (m->m_next == 0) {
1411 n = m_get(M_DONTWAIT, m->m_type);
1414 n->m_len = min(MLEN, len);
1419 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1420 m->m_pkthdr.len = totlen;
1424 m_print(const struct mbuf *m)
1427 const struct mbuf *m2;
1429 len = m->m_pkthdr.len;
1432 printf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-");
1440 * "Move" mbuf pkthdr from "from" to "to".
1441 * "from" must have M_PKTHDR set, and "to" must be empty.
1444 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
1446 KASSERT((to->m_flags & M_EXT) == 0, ("m_move_pkthdr: to has cluster"));
1448 to->m_flags = from->m_flags & M_COPYFLAGS;
1449 to->m_data = to->m_pktdat;
1450 to->m_pkthdr = from->m_pkthdr; /* especially tags */
1451 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
1452 from->m_flags &= ~M_PKTHDR;
1456 * Duplicate "from"'s mbuf pkthdr in "to".
1457 * "from" must have M_PKTHDR set, and "to" must be empty.
1458 * In particular, this does a deep copy of the packet tags.
1461 m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how)
1463 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
1464 if ((to->m_flags & M_EXT) == 0)
1465 to->m_data = to->m_pktdat;
1466 to->m_pkthdr = from->m_pkthdr;
1467 SLIST_INIT(&to->m_pkthdr.tags);
1468 return (m_tag_copy_chain(to, from, how));
1472 * Defragment a mbuf chain, returning the shortest possible
1473 * chain of mbufs and clusters. If allocation fails and
1474 * this cannot be completed, NULL will be returned, but
1475 * the passed in chain will be unchanged. Upon success,
1476 * the original chain will be freed, and the new chain
1479 * If a non-packet header is passed in, the original
1480 * mbuf (chain?) will be returned unharmed.
1483 m_defrag(struct mbuf *m0, int how)
1485 struct mbuf *m_new = NULL, *m_final = NULL;
1486 int progress = 0, length;
1488 if (!(m0->m_flags & M_PKTHDR))
1491 #ifdef MBUF_STRESS_TEST
1492 if (m_defragrandomfailures) {
1493 int temp = arc4random() & 0xff;
1499 if (m0->m_pkthdr.len > MHLEN)
1500 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1502 m_final = m_gethdr(how, MT_DATA);
1504 if (m_final == NULL)
1507 if (m_dup_pkthdr(m_final, m0, how) == NULL)
1512 while (progress < m0->m_pkthdr.len) {
1513 length = m0->m_pkthdr.len - progress;
1514 if (length > MCLBYTES)
1517 if (m_new == NULL) {
1519 m_new = m_getcl(how, MT_DATA, 0);
1521 m_new = m_get(how, MT_DATA);
1526 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1528 m_new->m_len = length;
1529 if (m_new != m_final)
1530 m_cat(m_final, m_new);
1533 if (m0->m_next == NULL)
1538 m_defragbytes += m0->m_pkthdr.len;