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 $
35 * $DragonFly: src/sys/kern/uipc_mbuf.c,v 1.14 2003/12/28 06:11:32 dillon Exp $
38 #include "opt_param.h"
39 #include "opt_mbuf_stress_test.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/malloc.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/domain.h>
47 #include <sys/protosw.h>
48 #include <sys/thread.h>
49 #include <sys/globaldata.h>
52 #include <vm/vm_kern.h>
53 #include <vm/vm_extern.h>
56 #include <machine/cpu.h>
59 static void mbinit (void *);
60 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbinit, NULL)
66 u_long mbtypes[MT_NTYPES];
68 union mcluster *mclfree;
77 #ifdef MBUF_STRESS_TEST
78 int m_defragrandomfailures;
83 u_int m_mballoc_wid = 0;
84 u_int m_clalloc_wid = 0;
86 SYSCTL_DECL(_kern_ipc);
87 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW,
89 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW,
90 &max_protohdr, 0, "");
91 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, "");
92 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW,
94 SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW,
96 SYSCTL_STRUCT(_kern_ipc, KIPC_MBSTAT, mbstat, CTLFLAG_RW, &mbstat, mbstat, "");
97 SYSCTL_OPAQUE(_kern_ipc, OID_AUTO, mbtypes, CTLFLAG_RD, mbtypes,
98 sizeof(mbtypes), "LU", "");
99 SYSCTL_INT(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLFLAG_RD,
100 &nmbclusters, 0, "Maximum number of mbuf clusters available");
101 SYSCTL_INT(_kern_ipc, OID_AUTO, nmbufs, CTLFLAG_RD, &nmbufs, 0,
102 "Maximum number of mbufs available");
103 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
104 &m_defragpackets, 0, "");
105 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
106 &m_defragbytes, 0, "");
107 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
108 &m_defraguseless, 0, "");
109 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
110 &m_defragfailure, 0, "");
111 #ifdef MBUF_STRESS_TEST
112 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
113 &m_defragrandomfailures, 0, "");
116 static void m_reclaim (void);
119 #define NMBCLUSTERS (512 + maxusers * 16)
122 #define NMBUFS (nmbclusters * 4)
126 * Perform sanity checks of tunables declared above.
129 tunable_mbinit(void *dummy)
133 * This has to be done before VM init.
135 nmbclusters = NMBCLUSTERS;
136 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
138 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs);
140 if (nmbufs < nmbclusters * 2)
141 nmbufs = nmbclusters * 2;
145 SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_mbinit, NULL);
147 /* "number of clusters of pages" */
159 mmbfree = NULL; mclfree = NULL;
160 mbstat.m_msize = MSIZE;
161 mbstat.m_mclbytes = MCLBYTES;
162 mbstat.m_minclsize = MINCLSIZE;
163 mbstat.m_mlen = MLEN;
164 mbstat.m_mhlen = MHLEN;
167 if (m_mballoc(NMB_INIT, M_DONTWAIT) == 0)
169 #if MCLBYTES <= PAGE_SIZE
170 if (m_clalloc(NCL_INIT, M_DONTWAIT) == 0)
173 /* It's OK to call contigmalloc in this context. */
174 if (m_clalloc(16, M_WAIT) == 0)
184 * Allocate at least nmb mbufs and place on mbuf free list.
185 * Must be called at splimp.
198 * If we've hit the mbuf limit, stop allocating from mb_map,
199 * (or trying to) in order to avoid dipping into the section of
200 * mb_map which we've "reserved" for clusters.
202 if ((nmb + mbstat.m_mbufs) > nmbufs)
206 * Once we run out of map space, it will be impossible to get
207 * any more (nothing is ever freed back to the map)
208 * -- however you are not dead as m_reclaim might
209 * still be able to free a substantial amount of space.
211 * XXX Furthermore, we can also work with "recycled" mbufs (when
212 * we're calling with M_WAIT the sleep procedure will be woken
213 * up when an mbuf is freed. See m_mballoc_wait()).
218 nbytes = round_page(nmb * MSIZE);
219 p = (caddr_t)kmem_malloc(mb_map, nbytes, M_NOWAIT);
220 if (p == 0 && how == M_WAIT) {
222 p = (caddr_t)kmem_malloc(mb_map, nbytes, M_WAITOK);
226 * Either the map is now full, or `how' is M_NOWAIT and there
232 nmb = nbytes / MSIZE;
233 for (i = 0; i < nmb; i++) {
234 ((struct mbuf *)p)->m_next = mmbfree;
235 mmbfree = (struct mbuf *)p;
238 mbstat.m_mbufs += nmb;
239 mbtypes[MT_FREE] += nmb;
244 * Once the mb_map has been exhausted and if the call to the allocation macros
245 * (or, in some cases, functions) is with M_WAIT, then it is necessary to rely
246 * solely on reclaimed mbufs. Here we wait for an mbuf to be freed for a
247 * designated (mbuf_wait) time.
250 m_mballoc_wait(int caller, int type)
257 if ((tsleep(&m_mballoc_wid, 0, "mballc", mbuf_wait)) == EWOULDBLOCK)
262 * Now that we (think) that we've got something, we will redo an
263 * MGET, but avoid getting into another instance of m_mballoc_wait()
264 * XXX: We retry to fetch _even_ if the sleep timed out. This is left
265 * this way, purposely, in the [unlikely] case that an mbuf was
266 * freed but the sleep was not awakened in time.
271 MGET(p, M_DONTWAIT, type);
274 MGETHDR(p, M_DONTWAIT, type);
277 panic("m_mballoc_wait: invalid caller (%d)", caller);
281 if (p != NULL) { /* We waited and got something... */
283 /* Wake up another if we have more free. */
291 #if MCLBYTES > PAGE_SIZE
292 static int i_want_my_mcl;
300 tsleep(&i_want_my_mcl, 0, "mclalloc", 0);
302 for (; i_want_my_mcl; i_want_my_mcl--) {
303 if (m_clalloc(1, M_WAIT) == 0)
304 printf("m_clalloc failed even in process context!\n");
309 static struct thread *mclallocthread;
310 static struct kproc_desc mclalloc_kp = {
315 SYSINIT(mclallocthread, SI_SUB_KTHREAD_UPDATE, SI_ORDER_ANY, kproc_start,
320 * Allocate some number of mbuf clusters
321 * and place on cluster free list.
322 * Must be called at splimp.
335 * If we've hit the mcluster number limit, stop allocating from
336 * mb_map, (or trying to) in order to avoid dipping into the section
337 * of mb_map which we've "reserved" for mbufs.
339 if ((ncl + mbstat.m_clusters) > nmbclusters)
343 * Once we run out of map space, it will be impossible
344 * to get any more (nothing is ever freed back to the
345 * map). From this point on, we solely rely on freed
351 #if MCLBYTES > PAGE_SIZE
353 i_want_my_mcl += ncl;
354 wakeup(&i_want_my_mcl);
358 p = contigmalloc_map(MCLBYTES * ncl, M_DEVBUF, M_WAITOK, 0ul,
359 ~0ul, PAGE_SIZE, 0, mb_map);
363 p = (caddr_t)kmem_malloc(mb_map, ctob(npg),
364 how != M_WAIT ? M_NOWAIT : M_WAITOK);
365 ncl = ncl * PAGE_SIZE / MCLBYTES;
368 * Either the map is now full, or `how' is M_NOWAIT and there
372 static int last_report ; /* when we did that (in ticks) */
375 if (ticks < last_report || (ticks - last_report) >= hz) {
377 printf("All mbuf clusters exhausted, please see tuning(7).\n");
382 for (i = 0; i < ncl; i++) {
383 ((union mcluster *)p)->mcl_next = mclfree;
384 mclfree = (union mcluster *)p;
388 mbstat.m_clusters += ncl;
393 * Once the mb_map submap has been exhausted and the allocation is called with
394 * M_WAIT, we rely on the mclfree union pointers. If nothing is free, we will
395 * sleep for a designated amount of time (mbuf_wait) or until we're woken up
396 * due to sudden mcluster availability.
404 /* If in interrupt context, and INVARIANTS, maintain sanity and die. */
405 KASSERT(mycpu->gd_intr_nesting_level == 0, ("CLALLOC: CANNOT WAIT IN INTERRUPT"));
407 /* Sleep until something's available or until we expire. */
409 if ((tsleep(&m_clalloc_wid, 0, "mclalc", mbuf_wait)) == EWOULDBLOCK)
413 * Now that we (think) that we've got something, we will redo and
414 * MGET, but avoid getting into another instance of m_clalloc_wait()
416 p = m_mclalloc(M_DONTWAIT);
419 if (p != NULL) { /* We waited and got something... */
421 /* Wake up another if we have more free. */
431 * When MGET fails, ask protocols to free space when short of memory,
432 * then re-attempt to allocate an mbuf.
442 * Must only do the reclaim if not in an interrupt context.
445 KASSERT(mycpu->gd_intr_nesting_level == 0,
446 ("MBALLOC: CANNOT WAIT IN INTERRUPT"));
452 (void)m_mballoc(1, i);
461 m->m_data = m->m_dat;
466 static int last_report ; /* when we did that (in ticks) */
470 if (ticks < last_report || (ticks - last_report) >= hz) {
472 printf("All mbufs exhausted, please see tuning(7).\n");
480 * As above; retry an MGETHDR.
490 * Must only do the reclaim if not in an interrupt context.
493 KASSERT(mycpu->gd_intr_nesting_level == 0,
494 ("MBALLOC: CANNOT WAIT IN INTERRUPT"));
500 (void)m_mballoc(1, i);
509 m->m_data = m->m_pktdat;
510 m->m_flags = M_PKTHDR;
511 m->m_pkthdr.rcvif = NULL;
512 SLIST_INIT(&m->m_pkthdr.tags);
513 m->m_pkthdr.csum_flags = 0;
517 static int last_report ; /* when we did that (in ticks) */
521 if (ticks < last_report || (ticks - last_report) >= hz) {
523 printf("All mbufs exhausted, please see tuning(7).\n");
537 for (dp = domains; dp; dp = dp->dom_next)
538 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
546 * Space allocation routines.
547 * These are also available as macros
548 * for critical paths.
559 (void)m_mballoc(1, how);
568 m->m_data = m->m_dat;
573 m = m_retry(how, type);
574 if (m == NULL && how == M_WAIT)
575 m = m_mballoc_wait(MGET_C, type);
589 (void)m_mballoc(1, how);
598 m->m_data = m->m_pktdat;
599 m->m_flags = M_PKTHDR;
600 m->m_pkthdr.rcvif = NULL;
601 SLIST_INIT(&m->m_pkthdr.tags);
602 m->m_pkthdr.csum_flags = 0;
606 m = m_retryhdr(how, type);
607 if (m == NULL && how == M_WAIT)
608 m = m_mballoc_wait(MGETHDR_C, type);
622 bzero(mtod(m, caddr_t), MLEN);
627 * m_getcl() returns an mbuf with an attached cluster.
628 * Because many network drivers use this kind of buffers a lot, it is
629 * convenient to keep a small pool of free buffers of this kind.
630 * Even a small size such as 10 gives about 10% improvement in the
631 * forwarding rate in a bridge or router.
632 * The size of this free list is controlled by the sysctl variable
633 * mcl_pool_max. The list is populated on m_freem(), and used in
634 * m_getcl() if elements are available.
636 static struct mbuf *mcl_pool;
637 static int mcl_pool_now;
638 static int mcl_pool_max = 0;
640 SYSCTL_INT(_kern_ipc, OID_AUTO, mcl_pool_max, CTLFLAG_RW, &mcl_pool_max, 0,
641 "Maximum number of mbufs+cluster in free list");
642 SYSCTL_INT(_kern_ipc, OID_AUTO, mcl_pool_now, CTLFLAG_RD, &mcl_pool_now, 0,
643 "Current number of mbufs+cluster in free list");
646 m_getcl(int how, short type, int flags)
651 if (flags & M_PKTHDR) {
652 if (type == MT_DATA && mcl_pool) {
654 mcl_pool = mp->m_nextpkt;
657 mp->m_nextpkt = NULL;
658 mp->m_data = mp->m_ext.ext_buf;
659 mp->m_flags = M_PKTHDR|M_EXT;
660 mp->m_pkthdr.rcvif = NULL;
661 mp->m_pkthdr.csum_flags = 0;
664 MGETHDR(mp, how, type);
669 if ( (mp->m_flags & M_EXT) == 0) {
680 * m_getm(m, len, how, type)
682 * This will allocate len-worth of mbufs and/or mbuf clusters (whatever fits
683 * best) and return a pointer to the top of the allocated chain. If m is
684 * non-null, then we assume that it is a single mbuf or an mbuf chain to
685 * which we want len bytes worth of mbufs and/or clusters attached, and so
686 * if we succeed in allocating it, we will just return a pointer to m.
688 * If we happen to fail at any point during the allocation, we will free
689 * up everything we have already allocated and return NULL.
693 m_getm(struct mbuf *m, int len, int how, int type)
695 struct mbuf *top, *tail, *mp, *mtail = NULL;
697 KASSERT(len >= 0, ("len is < 0 in m_getm"));
702 else if (len > MINCLSIZE) {
704 if ((mp->m_flags & M_EXT) == 0) {
710 len -= M_TRAILINGSPACE(mp);
713 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next);
725 if (len > MINCLSIZE) {
727 if ((mp->m_flags & M_EXT) == 0)
732 len -= M_TRAILINGSPACE(mp);
745 * m_mclalloc() - Allocates an mbuf cluster.
757 mp = (caddr_t)mclfree;
759 KKASSERT((struct mbuf *)mp >= mbutl &&
760 (struct mbuf *)mp < mbute);
761 mclrefcnt[mtocl(mp)]++;
763 mclfree = ((union mcluster *)mp)->mcl_next;
769 return(m_clalloc_wait());
774 * m_mclget() - Adds a cluster to a normal mbuf, M_EXT is set on success.
777 m_mclget(struct mbuf *m, int how)
779 m->m_ext.ext_buf = m_mclalloc(how);
780 if (m->m_ext.ext_buf != NULL) {
781 m->m_data = m->m_ext.ext_buf;
783 m->m_ext.ext_free = NULL;
784 m->m_ext.ext_ref = NULL;
785 m->m_ext.ext_size = MCLBYTES;
790 _m_mclfree(caddr_t data)
792 union mcluster *mp = (union mcluster *)data;
794 KASSERT(mclrefcnt[mtocl(mp)] > 0, ("freeing free cluster"));
795 KKASSERT((struct mbuf *)mp >= mbutl &&
796 (struct mbuf *)mp < mbute);
797 if (--mclrefcnt[mtocl(mp)] == 0) {
798 mp->mcl_next = mclfree;
806 m_mclfree(caddr_t mp)
816 * Free a single mbuf and any associated external storage. The successor,
817 * if any, is returned.
819 * We do need to check non-first mbuf for m_aux, since some of existing
820 * code does not call M_PREPEND properly.
821 * (example: call to bpf_mtap from drivers)
824 m_free(struct mbuf *m)
830 KASSERT(m->m_type != MT_FREE, ("freeing free mbuf"));
831 mbtypes[m->m_type]--;
832 if ((m->m_flags & M_PKTHDR) != 0)
833 m_tag_delete_chain(m, NULL);
834 if (m->m_flags & M_EXT) {
835 if (m->m_ext.ext_free != NULL) {
836 m->m_ext.ext_free(m->m_ext.ext_buf, m->m_ext.ext_size);
838 _m_mclfree(m->m_ext.ext_buf); /* inlined */
853 m_freem(struct mbuf *m)
858 * Try to keep a small pool of mbuf+cluster for quick use in
859 * device drivers. A good candidate is a M_PKTHDR buffer with
860 * only one cluster attached. Other mbufs, or those exceeding
861 * the pool size, are just m_free'd in the usual way.
862 * The following code makes sure that m_next, m_type,
863 * m_pkthdr.aux and m_ext.* are properly initialized.
864 * Other fields in the mbuf are initialized in m_getcl()
867 if (mcl_pool_now < mcl_pool_max && m && m->m_next == NULL &&
868 (m->m_flags & (M_PKTHDR|M_EXT)) == (M_PKTHDR|M_EXT) &&
869 m->m_type == MT_DATA && M_EXT_WRITABLE(m) ) {
870 m_tag_delete_chain(m, NULL);
871 m->m_nextpkt = mcl_pool;
882 * Mbuffer utility routines.
886 * Lesser-used path for M_PREPEND:
887 * allocate new mbuf to prepend to chain,
891 m_prepend(m, len, how)
897 MGET(mn, how, m->m_type);
898 if (mn == (struct mbuf *)NULL) {
900 return ((struct mbuf *)NULL);
902 if (m->m_flags & M_PKTHDR)
903 M_MOVE_PKTHDR(mn, m);
913 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
914 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
915 * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
916 * Note that the copy is read-only, because clusters are not copied,
917 * only their reference counts are incremented.
919 #define MCFail (mbstat.m_mcfail)
922 m_copym(m, off0, len, wait)
923 const struct mbuf *m;
927 struct mbuf *n, **np;
932 KASSERT(off >= 0, ("m_copym, negative off %d", off));
933 KASSERT(len >= 0, ("m_copym, negative len %d", len));
934 if (off == 0 && m->m_flags & M_PKTHDR)
937 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
947 KASSERT(len == M_COPYALL,
948 ("m_copym, length > size of mbuf chain"));
951 MGET(n, wait, m->m_type);
956 if (!m_dup_pkthdr(n, m, wait))
958 if (len == M_COPYALL)
959 n->m_pkthdr.len -= off0;
961 n->m_pkthdr.len = len;
964 n->m_len = min(len, m->m_len - off);
965 if (m->m_flags & M_EXT) {
966 n->m_data = m->m_data + off;
967 if (m->m_ext.ext_ref == NULL) {
969 &mclrefcnt[mtocl(m->m_ext.ext_buf)], 1);
973 (*m->m_ext.ext_ref)(m->m_ext.ext_buf,
980 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
982 if (len != M_COPYALL)
998 * Copy an entire packet, including header (which must be present).
999 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
1000 * Note that the copy is read-only, because clusters are not copied,
1001 * only their reference counts are incremented.
1002 * Preserve alignment of the first mbuf so if the creator has left
1003 * some room at the beginning (e.g. for inserting protocol headers)
1004 * the copies also have the room available.
1007 m_copypacket(m, how)
1011 struct mbuf *top, *n, *o;
1013 MGET(n, how, m->m_type);
1018 if (!m_dup_pkthdr(n, m, how))
1020 n->m_len = m->m_len;
1021 if (m->m_flags & M_EXT) {
1022 n->m_data = m->m_data;
1023 if (m->m_ext.ext_ref == NULL)
1024 atomic_add_char(&mclrefcnt[mtocl(m->m_ext.ext_buf)], 1);
1028 (*m->m_ext.ext_ref)(m->m_ext.ext_buf,
1032 n->m_ext = m->m_ext;
1033 n->m_flags |= M_EXT;
1035 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
1036 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
1041 MGET(o, how, m->m_type);
1048 n->m_len = m->m_len;
1049 if (m->m_flags & M_EXT) {
1050 n->m_data = m->m_data;
1051 if (m->m_ext.ext_ref == NULL) {
1053 &mclrefcnt[mtocl(m->m_ext.ext_buf)], 1);
1057 (*m->m_ext.ext_ref)(m->m_ext.ext_buf,
1061 n->m_ext = m->m_ext;
1062 n->m_flags |= M_EXT;
1064 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
1077 * Copy data from an mbuf chain starting "off" bytes from the beginning,
1078 * continuing for "len" bytes, into the indicated buffer.
1081 m_copydata(m, off, len, cp)
1082 const struct mbuf *m;
1089 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
1090 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
1092 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
1099 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
1100 count = min(m->m_len - off, len);
1101 bcopy(mtod(m, caddr_t) + off, cp, count);
1110 * Copy a packet header mbuf chain into a completely new chain, including
1111 * copying any mbuf clusters. Use this instead of m_copypacket() when
1112 * you need a writable copy of an mbuf chain.
1119 struct mbuf **p, *top = NULL;
1120 int remain, moff, nsize;
1125 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __FUNCTION__));
1127 /* While there's more data, get a new mbuf, tack it on, and fill it */
1128 remain = m->m_pkthdr.len;
1131 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
1134 /* Get the next new mbuf */
1135 MGET(n, how, m->m_type);
1138 if (top == NULL) { /* first one, must be PKTHDR */
1139 if (!m_dup_pkthdr(n, m, how))
1142 } else /* not the first one */
1144 if (remain >= MINCLSIZE) {
1146 if ((n->m_flags & M_EXT) == 0) {
1154 /* Link it into the new chain */
1158 /* Copy data from original mbuf(s) into new mbuf */
1159 while (n->m_len < nsize && m != NULL) {
1160 int chunk = min(nsize - n->m_len, m->m_len - moff);
1162 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
1166 if (moff == m->m_len) {
1172 /* Check correct total mbuf length */
1173 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
1174 ("%s: bogus m_pkthdr.len", __FUNCTION__));
1185 * Concatenate mbuf chain n to m.
1186 * Both chains must be of the same type (e.g. MT_DATA).
1187 * Any m_pkthdr is not updated.
1196 if (m->m_flags & M_EXT ||
1197 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
1198 /* just join the two chains */
1202 /* splat the data from one into the other */
1203 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1205 m->m_len += n->m_len;
1219 if ((m = mp) == NULL)
1225 while (m != NULL && len > 0) {
1226 if (m->m_len <= len) {
1237 if (mp->m_flags & M_PKTHDR)
1238 m->m_pkthdr.len -= (req_len - len);
1241 * Trim from tail. Scan the mbuf chain,
1242 * calculating its length and finding the last mbuf.
1243 * If the adjustment only affects this mbuf, then just
1244 * adjust and return. Otherwise, rescan and truncate
1245 * after the remaining size.
1251 if (m->m_next == (struct mbuf *)0)
1255 if (m->m_len >= len) {
1257 if (mp->m_flags & M_PKTHDR)
1258 mp->m_pkthdr.len -= len;
1265 * Correct length for chain is "count".
1266 * Find the mbuf with last data, adjust its length,
1267 * and toss data from remaining mbufs on chain.
1270 if (m->m_flags & M_PKTHDR)
1271 m->m_pkthdr.len = count;
1272 for (; m; m = m->m_next) {
1273 if (m->m_len >= count) {
1280 (m = m->m_next) ->m_len = 0;
1285 * Rearange an mbuf chain so that len bytes are contiguous
1286 * and in the data area of an mbuf (so that mtod and dtom
1287 * will work for a structure of size len). Returns the resulting
1288 * mbuf chain on success, frees it and returns null on failure.
1289 * If there is room, it will add up to max_protohdr-len extra bytes to the
1290 * contiguous region in an attempt to avoid being called next time.
1292 #define MPFail (mbstat.m_mpfail)
1304 * If first mbuf has no cluster, and has room for len bytes
1305 * without shifting current data, pullup into it,
1306 * otherwise allocate a new mbuf to prepend to the chain.
1308 if ((n->m_flags & M_EXT) == 0 &&
1309 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
1310 if (n->m_len >= len)
1318 MGET(m, M_DONTWAIT, n->m_type);
1322 if (n->m_flags & M_PKTHDR)
1323 M_MOVE_PKTHDR(m, n);
1325 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1327 count = min(min(max(len, max_protohdr), space), n->m_len);
1328 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1338 } while (len > 0 && n);
1352 * Partition an mbuf chain in two pieces, returning the tail --
1353 * all but the first len0 bytes. In case of failure, it returns NULL and
1354 * attempts to restore the chain to its original state.
1356 * Note that the resulting mbufs might be read-only, because the new
1357 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1358 * the "breaking point" happens to lie within a cluster mbuf. Use the
1359 * M_WRITABLE() macro to check for this case.
1362 m_split(m0, len0, wait)
1367 unsigned len = len0, remain;
1369 for (m = m0; m && len > m->m_len; m = m->m_next)
1373 remain = m->m_len - len;
1374 if (m0->m_flags & M_PKTHDR) {
1375 MGETHDR(n, wait, m0->m_type);
1378 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1379 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1380 m0->m_pkthdr.len = len0;
1381 if (m->m_flags & M_EXT)
1383 if (remain > MHLEN) {
1384 /* m can't be the lead packet */
1386 n->m_next = m_split(m, len, wait);
1387 if (n->m_next == 0) {
1395 MH_ALIGN(n, remain);
1396 } else if (remain == 0) {
1401 MGET(n, wait, m->m_type);
1407 if (m->m_flags & M_EXT) {
1408 n->m_flags |= M_EXT;
1409 n->m_ext = m->m_ext;
1410 if (m->m_ext.ext_ref == NULL)
1411 atomic_add_char(&mclrefcnt[mtocl(m->m_ext.ext_buf)], 1);
1415 (*m->m_ext.ext_ref)(m->m_ext.ext_buf,
1419 n->m_data = m->m_data + len;
1421 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1425 n->m_next = m->m_next;
1430 * Routine to copy from device local memory into mbufs.
1433 m_devget(buf, totlen, off0, ifp, copy)
1437 void (*copy) (char *from, caddr_t to, u_int len);
1440 struct mbuf *top = 0, **mp = ⊤
1441 int off = off0, len;
1448 cp += off + 2 * sizeof(u_short);
1449 totlen -= 2 * sizeof(u_short);
1451 MGETHDR(m, M_DONTWAIT, MT_DATA);
1454 m->m_pkthdr.rcvif = ifp;
1455 m->m_pkthdr.len = totlen;
1458 while (totlen > 0) {
1460 MGET(m, M_DONTWAIT, MT_DATA);
1467 len = min(totlen, epkt - cp);
1468 if (len >= MINCLSIZE) {
1469 MCLGET(m, M_DONTWAIT);
1470 if (m->m_flags & M_EXT)
1471 m->m_len = len = min(len, MCLBYTES);
1476 * Place initial small packet/header at end of mbuf.
1478 if (len < m->m_len) {
1479 if (top == 0 && len + max_linkhdr <= m->m_len)
1480 m->m_data += max_linkhdr;
1486 copy(cp, mtod(m, caddr_t), (unsigned)len);
1488 bcopy(cp, mtod(m, caddr_t), (unsigned)len);
1500 * Copy data from a buffer back into the indicated mbuf chain,
1501 * starting "off" bytes from the beginning, extending the mbuf
1502 * chain if necessary.
1505 m_copyback(m0, off, len, cp)
1512 struct mbuf *m = m0, *n;
1517 while (off > (mlen = m->m_len)) {
1520 if (m->m_next == 0) {
1521 n = m_getclr(M_DONTWAIT, m->m_type);
1524 n->m_len = min(MLEN, len + off);
1530 mlen = min (m->m_len - off, len);
1531 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen);
1539 if (m->m_next == 0) {
1540 n = m_get(M_DONTWAIT, m->m_type);
1543 n->m_len = min(MLEN, len);
1548 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1549 m->m_pkthdr.len = totlen;
1553 m_print(const struct mbuf *m)
1556 const struct mbuf *m2;
1558 len = m->m_pkthdr.len;
1561 printf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-");
1569 * "Move" mbuf pkthdr from "from" to "to".
1570 * "from" must have M_PKTHDR set, and "to" must be empty.
1573 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
1575 KASSERT((to->m_flags & M_EXT) == 0, ("m_move_pkthdr: to has cluster"));
1577 to->m_flags = from->m_flags & M_COPYFLAGS;
1578 to->m_data = to->m_pktdat;
1579 to->m_pkthdr = from->m_pkthdr; /* especially tags */
1580 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
1581 from->m_flags &= ~M_PKTHDR;
1585 * Duplicate "from"'s mbuf pkthdr in "to".
1586 * "from" must have M_PKTHDR set, and "to" must be empty.
1587 * In particular, this does a deep copy of the packet tags.
1590 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how)
1592 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
1593 if ((to->m_flags & M_EXT) == 0)
1594 to->m_data = to->m_pktdat;
1595 to->m_pkthdr = from->m_pkthdr;
1596 SLIST_INIT(&to->m_pkthdr.tags);
1597 return (m_tag_copy_chain(to, from, how));
1601 * Defragment a mbuf chain, returning the shortest possible
1602 * chain of mbufs and clusters. If allocation fails and
1603 * this cannot be completed, NULL will be returned, but
1604 * the passed in chain will be unchanged. Upon success,
1605 * the original chain will be freed, and the new chain
1608 * If a non-packet header is passed in, the original
1609 * mbuf (chain?) will be returned unharmed.
1612 m_defrag(struct mbuf *m0, int how)
1614 struct mbuf *m_new = NULL, *m_final = NULL;
1615 int progress = 0, length;
1617 if (!(m0->m_flags & M_PKTHDR))
1620 #ifdef MBUF_STRESS_TEST
1621 if (m_defragrandomfailures) {
1622 int temp = arc4random() & 0xff;
1628 if (m0->m_pkthdr.len > MHLEN)
1629 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1631 m_final = m_gethdr(how, MT_DATA);
1633 if (m_final == NULL)
1636 if (m_dup_pkthdr(m_final, m0, how) == NULL)
1641 while (progress < m0->m_pkthdr.len) {
1642 length = m0->m_pkthdr.len - progress;
1643 if (length > MCLBYTES)
1646 if (m_new == NULL) {
1648 m_new = m_getcl(how, MT_DATA, 0);
1650 m_new = m_get(how, MT_DATA);
1655 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1657 m_new->m_len = length;
1658 if (m_new != m_final)
1659 m_cat(m_final, m_new);
1662 if (m0->m_next == NULL)
1667 m_defragbytes += m0->m_pkthdr.len;