2 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved.
37 * License terms: all terms for the DragonFly license above plus the following:
39 * 4. All advertising materials mentioning features or use of this software
40 * must display the following acknowledgement:
42 * This product includes software developed by Jeffrey M. Hsu
43 * for the DragonFly Project.
45 * This requirement may be waived with permission from Jeffrey Hsu.
46 * This requirement will sunset and may be removed on July 8 2005,
47 * after which the standard DragonFly license (as shown above) will
52 * Copyright (c) 1982, 1986, 1988, 1991, 1993
53 * The Regents of the University of California. All rights reserved.
55 * Redistribution and use in source and binary forms, with or without
56 * modification, are permitted provided that the following conditions
58 * 1. Redistributions of source code must retain the above copyright
59 * notice, this list of conditions and the following disclaimer.
60 * 2. Redistributions in binary form must reproduce the above copyright
61 * notice, this list of conditions and the following disclaimer in the
62 * documentation and/or other materials provided with the distribution.
63 * 3. All advertising materials mentioning features or use of this software
64 * must display the following acknowledgement:
65 * This product includes software developed by the University of
66 * California, Berkeley and its contributors.
67 * 4. Neither the name of the University nor the names of its contributors
68 * may be used to endorse or promote products derived from this software
69 * without specific prior written permission.
71 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
72 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
73 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
74 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
75 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
76 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
77 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
78 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
79 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
80 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
83 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94
84 * $FreeBSD: src/sys/kern/uipc_mbuf.c,v 1.51.2.24 2003/04/15 06:59:29 silby Exp $
85 * $DragonFly: src/sys/kern/uipc_mbuf.c,v 1.42 2005/06/08 16:30:47 dillon Exp $
88 #include "opt_param.h"
90 #include "opt_mbuf_stress_test.h"
91 #include <sys/param.h>
92 #include <sys/systm.h>
93 #include <sys/malloc.h>
95 #include <sys/kernel.h>
96 #include <sys/sysctl.h>
97 #include <sys/domain.h>
98 #include <sys/objcache.h>
99 #include <sys/protosw.h>
101 #include <sys/thread.h>
102 #include <sys/globaldata.h>
103 #include <sys/thread2.h>
106 #include <vm/vm_kern.h>
107 #include <vm/vm_extern.h>
110 #include <machine/cpu.h>
114 * mbuf cluster meta-data
121 static void mbinit(void *);
122 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbinit, NULL)
124 static u_long mbtypes[MT_NTYPES];
126 struct mbstat mbstat;
135 #ifdef MBUF_STRESS_TEST
136 int m_defragrandomfailures;
139 struct objcache *mbuf_cache, *mbufphdr_cache;
140 struct objcache *mclmeta_cache;
141 struct objcache *mbufcluster_cache, *mbufphdrcluster_cache;
146 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW,
147 &max_linkhdr, 0, "");
148 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW,
149 &max_protohdr, 0, "");
150 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, "");
151 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW,
152 &max_datalen, 0, "");
153 SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW,
155 SYSCTL_STRUCT(_kern_ipc, KIPC_MBSTAT, mbstat, CTLFLAG_RW, &mbstat, mbstat, "");
156 SYSCTL_OPAQUE(_kern_ipc, OID_AUTO, mbtypes, CTLFLAG_RD, mbtypes,
157 sizeof(mbtypes), "LU", "");
158 SYSCTL_INT(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLFLAG_RW,
159 &nmbclusters, 0, "Maximum number of mbuf clusters available");
160 SYSCTL_INT(_kern_ipc, OID_AUTO, nmbufs, CTLFLAG_RW, &nmbufs, 0,
161 "Maximum number of mbufs available");
163 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
164 &m_defragpackets, 0, "");
165 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
166 &m_defragbytes, 0, "");
167 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
168 &m_defraguseless, 0, "");
169 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
170 &m_defragfailure, 0, "");
171 #ifdef MBUF_STRESS_TEST
172 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
173 &m_defragrandomfailures, 0, "");
176 static MALLOC_DEFINE(M_MBUF, "mbuf", "mbuf");
177 static MALLOC_DEFINE(M_MBUFCL, "mbufcl", "mbufcl");
178 static MALLOC_DEFINE(M_MCLMETA, "mclmeta", "mclmeta");
180 static void m_reclaim (void);
181 static void m_mclref(void *arg);
182 static void m_mclfree(void *arg);
185 #define NMBCLUSTERS (512 + maxusers * 16)
188 #define NMBUFS (nmbclusters * 2)
192 * Perform sanity checks of tunables declared above.
195 tunable_mbinit(void *dummy)
199 * This has to be done before VM init.
201 nmbclusters = NMBCLUSTERS;
202 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
204 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs);
206 if (nmbufs < nmbclusters * 2)
207 nmbufs = nmbclusters * 2;
211 SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_mbinit, NULL);
213 /* "number of clusters of pages" */
219 * The mbuf object cache only guarantees that m_next and m_nextpkt are
220 * NULL and that m_data points to the beginning of the data area. In
221 * particular, m_len and m_pkthdr.len are uninitialized. It is the
222 * responsibility of the caller to initialize those fields before use.
225 static boolean_t __inline
226 mbuf_ctor(void *obj, void *private, int ocflags)
228 struct mbuf *m = obj;
232 m->m_data = m->m_dat;
239 * Initialize the mbuf and the packet header fields.
242 mbufphdr_ctor(void *obj, void *private, int ocflags)
244 struct mbuf *m = obj;
248 m->m_data = m->m_pktdat;
249 m->m_flags = M_PKTHDR;
251 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */
252 SLIST_INIT(&m->m_pkthdr.tags);
253 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */
254 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */
260 * A mbcluster object consists of 2K (MCLBYTES) cluster and a refcount.
263 mclmeta_ctor(void *obj, void *private, int ocflags)
265 struct mbcluster *cl = obj;
268 if (ocflags & M_NOWAIT)
269 buf = malloc(MCLBYTES, M_MBUFCL, M_NOWAIT | M_ZERO);
271 buf = malloc(MCLBYTES, M_MBUFCL, M_INTWAIT | M_ZERO);
280 linkcluster(struct mbuf *m, struct mbcluster *cl)
283 * Add the cluster to the mbuf. The caller will detect that the
284 * mbuf now has an attached cluster.
286 m->m_ext.ext_arg = cl;
287 m->m_ext.ext_buf = cl->mcl_data;
288 m->m_ext.ext_ref = m_mclref;
289 m->m_ext.ext_free = m_mclfree;
290 m->m_ext.ext_size = MCLBYTES;
292 m->m_data = m->m_ext.ext_buf;
293 m->m_flags |= M_EXT | M_EXT_CLUSTER;
297 mbufphdrcluster_ctor(void *obj, void *private, int ocflags)
299 struct mbuf *m = obj;
300 struct mbcluster *cl;
302 mbufphdr_ctor(obj, private, ocflags);
303 cl = objcache_get(mclmeta_cache, ocflags);
311 mbufcluster_ctor(void *obj, void *private, int ocflags)
313 struct mbuf *m = obj;
314 struct mbcluster *cl;
316 mbuf_ctor(obj, private, ocflags);
317 cl = objcache_get(mclmeta_cache, ocflags);
325 mclmeta_dtor(void *obj, void *private)
327 struct mbcluster *mcl = obj;
329 KKASSERT(mcl->mcl_refs == 1);
330 free(mcl->mcl_data, M_MBUFCL);
334 mbufcluster_dtor(void *obj, void *private)
336 struct mbuf *m = obj;
338 objcache_put(mclmeta_cache, m->m_ext.ext_arg);
342 mbufphdrcluster_dtor(void *obj, void *private)
344 struct mbuf *m = obj;
346 objcache_put(mclmeta_cache, m->m_ext.ext_arg);
349 struct objcache_malloc_args mbuf_malloc_args = { MSIZE, M_MBUF };
350 struct objcache_malloc_args mclmeta_malloc_args =
351 { sizeof(struct mbcluster), M_MCLMETA };
357 mbstat.m_msize = MSIZE;
358 mbstat.m_mclbytes = MCLBYTES;
359 mbstat.m_minclsize = MINCLSIZE;
360 mbstat.m_mlen = MLEN;
361 mbstat.m_mhlen = MHLEN;
363 mbuf_cache = objcache_create("mbuf", nmbufs, 0,
364 mbuf_ctor, null_dtor, NULL,
365 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
366 mbufphdr_cache = objcache_create("mbuf pkt hdr", nmbufs, 64,
367 mbufphdr_ctor, null_dtor, NULL,
368 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
369 mclmeta_cache = objcache_create("cluster mbuf", nmbclusters , 0,
370 mclmeta_ctor, mclmeta_dtor, NULL,
371 objcache_malloc_alloc, objcache_malloc_free, &mclmeta_malloc_args);
372 mbufcluster_cache = objcache_create("mbuf + cluster", nmbclusters, 0,
373 mbufcluster_ctor, mbufcluster_dtor, NULL,
374 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
375 mbufphdrcluster_cache = objcache_create("mbuf pkt hdr + cluster",
376 nmbclusters, 64, mbufphdrcluster_ctor, mbufphdrcluster_dtor, NULL,
377 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
382 * Return the number of references to this mbuf's data. 0 is returned
383 * if the mbuf is not M_EXT, a reference count is returned if it is
384 * M_EXT | M_EXT_CLUSTER, and 99 is returned if it is a special M_EXT.
387 m_sharecount(struct mbuf *m)
389 switch (m->m_flags & (M_EXT | M_EXT_CLUSTER)) {
394 case M_EXT | M_EXT_CLUSTER:
395 return (((struct mbcluster *)m->m_ext.ext_arg)->mcl_refs);
398 return (0); /* to shut up compiler */
402 * change mbuf to new type
405 m_chtype(struct mbuf *m, int type)
409 --mbtypes[m->m_type];
421 SLIST_FOREACH(dp, &domains, dom_next) {
422 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
432 updatestats(struct mbuf *m, int type)
446 m_get(int how, int type)
450 int ocf = MBTOM(how);
454 m = objcache_get(mbuf_cache, ocf);
457 if ((how & MB_TRYWAIT) && ntries++ == 0) {
458 struct objcache *reclaimlist[] = {
460 mbufcluster_cache, mbufphdrcluster_cache
462 const int nreclaims = __arysize(reclaimlist);
464 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf))
471 updatestats(m, type);
476 m_gethdr(int how, int type)
479 int ocf = MBTOM(how);
484 m = objcache_get(mbufphdr_cache, ocf);
487 if ((how & MB_TRYWAIT) && ntries++ == 0) {
488 struct objcache *reclaimlist[] = {
490 mbufcluster_cache, mbufphdrcluster_cache
492 const int nreclaims = __arysize(reclaimlist);
494 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf))
501 updatestats(m, type);
506 * Get a mbuf (not a mbuf cluster!) and zero it.
510 m_getclr(int how, int type)
514 m = m_get(how, type);
516 bzero(m->m_data, MLEN);
521 * Returns an mbuf with an attached cluster.
522 * Because many network drivers use this kind of buffers a lot, it is
523 * convenient to keep a small pool of free buffers of this kind.
524 * Even a small size such as 10 gives about 10% improvement in the
525 * forwarding rate in a bridge or router.
528 m_getcl(int how, short type, int flags)
531 int ocflags = MBTOM(how);
536 if (flags & M_PKTHDR)
537 m = objcache_get(mbufphdrcluster_cache, ocflags);
539 m = objcache_get(mbufcluster_cache, ocflags);
542 if ((how & MB_TRYWAIT) && ntries++ == 0) {
543 struct objcache *reclaimlist[1];
545 if (flags & M_PKTHDR)
546 reclaimlist[0] = mbufcluster_cache;
548 reclaimlist[0] = mbufphdrcluster_cache;
549 if (!objcache_reclaimlist(reclaimlist, 1, ocflags))
566 * Allocate chain of requested length.
569 m_getc(int len, int how, int type)
571 struct mbuf *n, *nfirst = NULL, **ntail = &nfirst;
575 n = m_getl(len, how, type, 0, &nsize);
591 * Allocate len-worth of mbufs and/or mbuf clusters (whatever fits best)
592 * and return a pointer to the head of the allocated chain. If m0 is
593 * non-null, then we assume that it is a single mbuf or an mbuf chain to
594 * which we want len bytes worth of mbufs and/or clusters attached, and so
595 * if we succeed in allocating it, we will just return a pointer to m0.
597 * If we happen to fail at any point during the allocation, we will free
598 * up everything we have already allocated and return NULL.
600 * Deprecated. Use m_getc() and m_cat() instead.
603 m_getm(struct mbuf *m0, int len, int how, int type)
607 nfirst = m_getc(len, how, type);
610 m_last(m0)->m_next = nfirst;
618 * Adds a cluster to a normal mbuf, M_EXT is set on success.
619 * Deprecated. Use m_getcl() instead.
622 m_mclget(struct mbuf *m, int how)
624 struct mbcluster *mcl;
626 mcl = objcache_get(mclmeta_cache, MBTOM(how));
640 struct mbcluster *mcl = arg;
642 atomic_add_int(&mcl->mcl_refs, 1);
648 struct mbcluster *mcl = arg;
650 KKASSERT(mcl->mcl_refs > 1);
651 atomic_subtract_int(&mcl->mcl_refs, 1);
654 extern void db_print_backtrace(void);
657 * Free a single mbuf and any associated external storage. The successor,
658 * if any, is returned.
660 * We do need to check non-first mbuf for m_aux, since some of existing
661 * code does not call M_PREPEND properly.
662 * (example: call to bpf_mtap from drivers)
665 m_free(struct mbuf *m)
669 KASSERT(m->m_type != MT_FREE, ("freeing free mbuf %p", m));
670 --mbtypes[m->m_type];
675 * Make sure the mbuf is in constructed state before returning it
680 KKASSERT(m->m_nextpkt == NULL);
682 if (m->m_nextpkt != NULL) {
684 static int afewtimes = 10;
686 if (afewtimes-- > 0) {
687 printf("mfree: m->m_nextpkt != NULL\n");
688 db_print_backtrace();
694 if (m->m_flags & M_PKTHDR) {
695 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */
696 m_tag_delete_chain(m); /* eliminate XXX JH */
697 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */
698 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */
700 m->m_flags &= (M_PKTHDR | M_EXT | M_EXT_CLUSTER);
702 if (m->m_flags & M_EXT) {
703 if (m_sharecount(m) > 1) {
704 m->m_ext.ext_free(m->m_ext.ext_arg);
705 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER);
708 KKASSERT(((struct mbcluster *)m->m_ext.ext_arg)->mcl_refs == 1);
709 m->m_data = m->m_ext.ext_buf;
710 if (m->m_flags & M_PKTHDR)
711 objcache_put(mbufphdrcluster_cache, m);
713 objcache_put(mbufcluster_cache, m);
719 if (m->m_flags & M_PKTHDR) {
720 m->m_data = m->m_pktdat;
721 objcache_put(mbufphdr_cache, m);
723 m->m_data = m->m_dat;
724 objcache_put(mbuf_cache, m);
734 m_freem(struct mbuf *m)
743 * mbuf utility routines
747 * Lesser-used path for M_PREPEND: allocate new mbuf to prepend to chain and
751 m_prepend(struct mbuf *m, int len, int how)
755 mn = m_get(how, m->m_type);
760 if (m->m_flags & M_PKTHDR)
761 M_MOVE_PKTHDR(mn, m);
771 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
772 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
773 * The wait parameter is a choice of MB_WAIT/MB_DONTWAIT from caller.
774 * Note that the copy is read-only, because clusters are not copied,
775 * only their reference counts are incremented.
778 m_copym(const struct mbuf *m, int off0, int len, int wait)
780 struct mbuf *n, **np;
785 KASSERT(off >= 0, ("m_copym, negative off %d", off));
786 KASSERT(len >= 0, ("m_copym, negative len %d", len));
787 if (off == 0 && m->m_flags & M_PKTHDR)
790 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
800 KASSERT(len == M_COPYALL,
801 ("m_copym, length > size of mbuf chain"));
804 n = m_get(wait, m->m_type);
809 if (!m_dup_pkthdr(n, m, wait))
811 if (len == M_COPYALL)
812 n->m_pkthdr.len -= off0;
814 n->m_pkthdr.len = len;
817 n->m_len = min(len, m->m_len - off);
818 if (m->m_flags & M_EXT) {
819 n->m_data = m->m_data + off;
820 m->m_ext.ext_ref(m->m_ext.ext_arg);
822 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
824 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
827 if (len != M_COPYALL)
843 * Copy an entire packet, including header (which must be present).
844 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
845 * Note that the copy is read-only, because clusters are not copied,
846 * only their reference counts are incremented.
847 * Preserve alignment of the first mbuf so if the creator has left
848 * some room at the beginning (e.g. for inserting protocol headers)
849 * the copies also have the room available.
852 m_copypacket(struct mbuf *m, int how)
854 struct mbuf *top, *n, *o;
856 n = m_get(how, m->m_type);
861 if (!m_dup_pkthdr(n, m, how))
864 if (m->m_flags & M_EXT) {
865 n->m_data = m->m_data;
866 m->m_ext.ext_ref(m->m_ext.ext_arg);
868 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
870 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
871 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
876 o = m_get(how, m->m_type);
884 if (m->m_flags & M_EXT) {
885 n->m_data = m->m_data;
886 m->m_ext.ext_ref(m->m_ext.ext_arg);
888 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
890 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
903 * Copy data from an mbuf chain starting "off" bytes from the beginning,
904 * continuing for "len" bytes, into the indicated buffer.
907 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
911 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
912 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
914 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
921 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
922 count = min(m->m_len - off, len);
923 bcopy(mtod(m, caddr_t) + off, cp, count);
932 * Copy a packet header mbuf chain into a completely new chain, including
933 * copying any mbuf clusters. Use this instead of m_copypacket() when
934 * you need a writable copy of an mbuf chain.
937 m_dup(struct mbuf *m, int how)
939 struct mbuf **p, *top = NULL;
940 int remain, moff, nsize;
945 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __func__));
947 /* While there's more data, get a new mbuf, tack it on, and fill it */
948 remain = m->m_pkthdr.len;
951 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
954 /* Get the next new mbuf */
955 n = m_getl(remain, how, m->m_type, top == NULL ? M_PKTHDR : 0,
960 if (!m_dup_pkthdr(n, m, how))
963 /* Link it into the new chain */
967 /* Copy data from original mbuf(s) into new mbuf */
969 while (n->m_len < nsize && m != NULL) {
970 int chunk = min(nsize - n->m_len, m->m_len - moff);
972 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
976 if (moff == m->m_len) {
982 /* Check correct total mbuf length */
983 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
984 ("%s: bogus m_pkthdr.len", __func__));
996 * Concatenate mbuf chain n to m.
997 * Both chains must be of the same type (e.g. MT_DATA).
998 * Any m_pkthdr is not updated.
1001 m_cat(struct mbuf *m, struct mbuf *n)
1005 if (m->m_flags & M_EXT ||
1006 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
1007 /* just join the two chains */
1011 /* splat the data from one into the other */
1012 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1014 m->m_len += n->m_len;
1020 m_adj(struct mbuf *mp, int req_len)
1026 if ((m = mp) == NULL)
1032 while (m != NULL && len > 0) {
1033 if (m->m_len <= len) {
1044 if (mp->m_flags & M_PKTHDR)
1045 m->m_pkthdr.len -= (req_len - len);
1048 * Trim from tail. Scan the mbuf chain,
1049 * calculating its length and finding the last mbuf.
1050 * If the adjustment only affects this mbuf, then just
1051 * adjust and return. Otherwise, rescan and truncate
1052 * after the remaining size.
1058 if (m->m_next == (struct mbuf *)0)
1062 if (m->m_len >= len) {
1064 if (mp->m_flags & M_PKTHDR)
1065 mp->m_pkthdr.len -= len;
1072 * Correct length for chain is "count".
1073 * Find the mbuf with last data, adjust its length,
1074 * and toss data from remaining mbufs on chain.
1077 if (m->m_flags & M_PKTHDR)
1078 m->m_pkthdr.len = count;
1079 for (; m; m = m->m_next) {
1080 if (m->m_len >= count) {
1087 (m = m->m_next) ->m_len = 0;
1092 * Rearrange an mbuf chain so that len bytes are contiguous
1093 * and in the data area of an mbuf (so that mtod will work for a structure
1094 * of size len). Returns the resulting mbuf chain on success, frees it and
1095 * returns null on failure. If there is room, it will add up to
1096 * max_protohdr-len extra bytes to the contiguous region in an attempt to
1097 * avoid being called next time.
1100 m_pullup(struct mbuf *n, int len)
1107 * If first mbuf has no cluster, and has room for len bytes
1108 * without shifting current data, pullup into it,
1109 * otherwise allocate a new mbuf to prepend to the chain.
1111 if (!(n->m_flags & M_EXT) &&
1112 n->m_data + len < &n->m_dat[MLEN] &&
1114 if (n->m_len >= len)
1122 m = m_get(MB_DONTWAIT, n->m_type);
1126 if (n->m_flags & M_PKTHDR)
1127 M_MOVE_PKTHDR(m, n);
1129 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1131 count = min(min(max(len, max_protohdr), space), n->m_len);
1132 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1142 } while (len > 0 && n);
1156 * Partition an mbuf chain in two pieces, returning the tail --
1157 * all but the first len0 bytes. In case of failure, it returns NULL and
1158 * attempts to restore the chain to its original state.
1160 * Note that the resulting mbufs might be read-only, because the new
1161 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1162 * the "breaking point" happens to lie within a cluster mbuf. Use the
1163 * M_WRITABLE() macro to check for this case.
1166 m_split(struct mbuf *m0, int len0, int wait)
1169 unsigned len = len0, remain;
1171 for (m = m0; m && len > m->m_len; m = m->m_next)
1175 remain = m->m_len - len;
1176 if (m0->m_flags & M_PKTHDR) {
1177 n = m_gethdr(wait, m0->m_type);
1180 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1181 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1182 m0->m_pkthdr.len = len0;
1183 if (m->m_flags & M_EXT)
1185 if (remain > MHLEN) {
1186 /* m can't be the lead packet */
1188 n->m_next = m_split(m, len, wait);
1189 if (n->m_next == NULL) {
1197 MH_ALIGN(n, remain);
1198 } else if (remain == 0) {
1203 n = m_get(wait, m->m_type);
1209 if (m->m_flags & M_EXT) {
1210 n->m_data = m->m_data + len;
1211 m->m_ext.ext_ref(m->m_ext.ext_arg);
1212 n->m_ext = m->m_ext;
1213 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1215 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1219 n->m_next = m->m_next;
1225 * Routine to copy from device local memory into mbufs.
1226 * Note: "offset" is ill-defined and always called as 0, so ignore it.
1229 m_devget(char *buf, int len, int offset, struct ifnet *ifp,
1230 void (*copy)(volatile const void *from, volatile void *to, size_t length))
1232 struct mbuf *m, *mfirst = NULL, **mtail;
1241 m = m_getl(len, MB_DONTWAIT, MT_DATA, flags, &nsize);
1246 m->m_len = min(len, nsize);
1248 if (flags & M_PKTHDR) {
1249 if (len + max_linkhdr <= nsize)
1250 m->m_data += max_linkhdr;
1251 m->m_pkthdr.rcvif = ifp;
1252 m->m_pkthdr.len = len;
1256 copy(buf, m->m_data, (unsigned)m->m_len);
1267 * Copy data from a buffer back into the indicated mbuf chain,
1268 * starting "off" bytes from the beginning, extending the mbuf
1269 * chain if necessary.
1272 m_copyback(struct mbuf *m0, int off, int len, caddr_t cp)
1275 struct mbuf *m = m0, *n;
1280 while (off > (mlen = m->m_len)) {
1283 if (m->m_next == NULL) {
1284 n = m_getclr(MB_DONTWAIT, m->m_type);
1287 n->m_len = min(MLEN, len + off);
1293 mlen = min (m->m_len - off, len);
1294 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen);
1302 if (m->m_next == NULL) {
1303 n = m_get(MB_DONTWAIT, m->m_type);
1306 n->m_len = min(MLEN, len);
1311 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1312 m->m_pkthdr.len = totlen;
1316 m_print(const struct mbuf *m)
1319 const struct mbuf *m2;
1321 len = m->m_pkthdr.len;
1324 printf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-");
1332 * "Move" mbuf pkthdr from "from" to "to".
1333 * "from" must have M_PKTHDR set, and "to" must be empty.
1336 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
1338 KASSERT(!(to->m_flags & M_EXT), ("m_move_pkthdr: to has cluster"));
1340 to->m_flags = from->m_flags & M_COPYFLAGS;
1341 to->m_data = to->m_pktdat;
1342 to->m_pkthdr = from->m_pkthdr; /* especially tags */
1343 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
1344 from->m_flags &= ~M_PKTHDR;
1348 * Duplicate "from"'s mbuf pkthdr in "to".
1349 * "from" must have M_PKTHDR set, and "to" must be empty.
1350 * In particular, this does a deep copy of the packet tags.
1353 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how)
1355 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
1356 if (!(to->m_flags & M_EXT))
1357 to->m_data = to->m_pktdat;
1358 to->m_pkthdr = from->m_pkthdr;
1359 SLIST_INIT(&to->m_pkthdr.tags);
1360 return (m_tag_copy_chain(to, from, how));
1364 * Defragment a mbuf chain, returning the shortest possible
1365 * chain of mbufs and clusters. If allocation fails and
1366 * this cannot be completed, NULL will be returned, but
1367 * the passed in chain will be unchanged. Upon success,
1368 * the original chain will be freed, and the new chain
1371 * If a non-packet header is passed in, the original
1372 * mbuf (chain?) will be returned unharmed.
1374 * m_defrag_nofree doesn't free the passed in mbuf.
1377 m_defrag(struct mbuf *m0, int how)
1381 if ((m_new = m_defrag_nofree(m0, how)) == NULL)
1389 m_defrag_nofree(struct mbuf *m0, int how)
1391 struct mbuf *m_new = NULL, *m_final = NULL;
1392 int progress = 0, length, nsize;
1394 if (!(m0->m_flags & M_PKTHDR))
1397 #ifdef MBUF_STRESS_TEST
1398 if (m_defragrandomfailures) {
1399 int temp = arc4random() & 0xff;
1405 m_final = m_getl(m0->m_pkthdr.len, how, MT_DATA, M_PKTHDR, &nsize);
1406 if (m_final == NULL)
1408 m_final->m_len = 0; /* in case m0->m_pkthdr.len is zero */
1410 if (m_dup_pkthdr(m_final, m0, how) == NULL)
1415 while (progress < m0->m_pkthdr.len) {
1416 length = m0->m_pkthdr.len - progress;
1417 if (length > MCLBYTES)
1420 if (m_new == NULL) {
1421 m_new = m_getl(length, how, MT_DATA, 0, &nsize);
1426 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1428 m_new->m_len = length;
1429 if (m_new != m_final)
1430 m_cat(m_final, m_new);
1433 if (m0->m_next == NULL)
1436 m_defragbytes += m_final->m_pkthdr.len;
1447 * Move data from uio into mbufs.
1450 m_uiomove(struct uio *uio)
1452 struct mbuf *m; /* current working mbuf */
1453 struct mbuf *head = NULL; /* result mbuf chain */
1454 struct mbuf **mp = &head;
1455 int resid = uio->uio_resid, nsize, flags = M_PKTHDR, error;
1458 m = m_getl(resid, MB_WAIT, MT_DATA, flags, &nsize);
1460 m->m_pkthdr.len = 0;
1461 /* Leave room for protocol headers. */
1466 m->m_len = min(nsize, resid);
1467 error = uiomove(mtod(m, caddr_t), m->m_len, uio);
1474 head->m_pkthdr.len += m->m_len;
1476 } while (resid > 0);
1486 m_last(struct mbuf *m)
1494 * Return the number of bytes in an mbuf chain.
1495 * If lastm is not NULL, also return the last mbuf.
1498 m_lengthm(struct mbuf *m, struct mbuf **lastm)
1501 struct mbuf *prev = m;
1514 * Like m_lengthm(), except also keep track of mbuf usage.
1517 m_countm(struct mbuf *m, struct mbuf **lastm, u_int *pmbcnt)
1519 u_int len = 0, mbcnt = 0;
1520 struct mbuf *prev = m;
1525 if (m->m_flags & M_EXT)
1526 mbcnt += m->m_ext.ext_size;