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 1.52 2005/06/17 18:58:02 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 | M_PHCACHE;
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 mclmeta_dtor(void *obj, void *private)
282 struct mbcluster *mcl = obj;
284 KKASSERT(mcl->mcl_refs == 0);
285 free(mcl->mcl_data, M_MBUFCL);
289 linkcluster(struct mbuf *m, struct mbcluster *cl)
292 * Add the cluster to the mbuf. The caller will detect that the
293 * mbuf now has an attached cluster.
295 m->m_ext.ext_arg = cl;
296 m->m_ext.ext_buf = cl->mcl_data;
297 m->m_ext.ext_ref = m_mclref;
298 m->m_ext.ext_free = m_mclfree;
299 m->m_ext.ext_size = MCLBYTES;
302 m->m_data = m->m_ext.ext_buf;
303 m->m_flags |= M_EXT | M_EXT_CLUSTER;
307 mbufphdrcluster_ctor(void *obj, void *private, int ocflags)
309 struct mbuf *m = obj;
310 struct mbcluster *cl;
312 mbufphdr_ctor(obj, private, ocflags);
313 cl = objcache_get(mclmeta_cache, ocflags);
316 m->m_flags |= M_CLCACHE;
322 mbufcluster_ctor(void *obj, void *private, int ocflags)
324 struct mbuf *m = obj;
325 struct mbcluster *cl;
327 mbuf_ctor(obj, private, ocflags);
328 cl = objcache_get(mclmeta_cache, ocflags);
331 m->m_flags |= M_CLCACHE;
337 * Used for both the cluster and cluster PHDR caches.
339 * The mbuf may have lost its cluster due to sharing, deal
340 * with the situation by checking M_EXT.
343 mbufcluster_dtor(void *obj, void *private)
345 struct mbuf *m = obj;
346 struct mbcluster *mcl;
348 if (m->m_flags & M_EXT) {
349 KKASSERT((m->m_flags & M_EXT_CLUSTER) != 0);
350 mcl = m->m_ext.ext_arg;
351 KKASSERT(mcl->mcl_refs == 1);
353 objcache_put(mclmeta_cache, mcl);
357 struct objcache_malloc_args mbuf_malloc_args = { MSIZE, M_MBUF };
358 struct objcache_malloc_args mclmeta_malloc_args =
359 { sizeof(struct mbcluster), M_MCLMETA };
365 mbstat.m_msize = MSIZE;
366 mbstat.m_mclbytes = MCLBYTES;
367 mbstat.m_minclsize = MINCLSIZE;
368 mbstat.m_mlen = MLEN;
369 mbstat.m_mhlen = MHLEN;
371 mbuf_cache = objcache_create("mbuf", nmbufs, 0,
372 mbuf_ctor, null_dtor, NULL,
373 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
374 mbufphdr_cache = objcache_create("mbuf pkt hdr", nmbufs, 64,
375 mbufphdr_ctor, null_dtor, NULL,
376 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
377 mclmeta_cache = objcache_create("cluster mbuf", nmbclusters , 0,
378 mclmeta_ctor, mclmeta_dtor, NULL,
379 objcache_malloc_alloc, objcache_malloc_free, &mclmeta_malloc_args);
380 mbufcluster_cache = objcache_create("mbuf + cluster", nmbclusters, 0,
381 mbufcluster_ctor, mbufcluster_dtor, NULL,
382 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
383 mbufphdrcluster_cache = objcache_create("mbuf pkt hdr + cluster",
384 nmbclusters, 64, mbufphdrcluster_ctor, mbufcluster_dtor, NULL,
385 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
390 * Return the number of references to this mbuf's data. 0 is returned
391 * if the mbuf is not M_EXT, a reference count is returned if it is
392 * M_EXT | M_EXT_CLUSTER, and 99 is returned if it is a special M_EXT.
395 m_sharecount(struct mbuf *m)
397 switch (m->m_flags & (M_EXT | M_EXT_CLUSTER)) {
402 case M_EXT | M_EXT_CLUSTER:
403 return (((struct mbcluster *)m->m_ext.ext_arg)->mcl_refs);
406 return (0); /* to shut up compiler */
410 * change mbuf to new type
413 m_chtype(struct mbuf *m, int type)
417 --mbtypes[m->m_type];
429 SLIST_FOREACH(dp, &domains, dom_next) {
430 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
440 updatestats(struct mbuf *m, int type)
454 m_get(int how, int type)
458 int ocf = MBTOM(how);
462 m = objcache_get(mbuf_cache, ocf);
465 if ((how & MB_TRYWAIT) && ntries++ == 0) {
466 struct objcache *reclaimlist[] = {
468 mbufcluster_cache, mbufphdrcluster_cache
470 const int nreclaims = __arysize(reclaimlist);
472 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf))
479 updatestats(m, type);
484 m_gethdr(int how, int type)
487 int ocf = MBTOM(how);
492 m = objcache_get(mbufphdr_cache, ocf);
495 if ((how & MB_TRYWAIT) && ntries++ == 0) {
496 struct objcache *reclaimlist[] = {
498 mbufcluster_cache, mbufphdrcluster_cache
500 const int nreclaims = __arysize(reclaimlist);
502 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf))
509 updatestats(m, type);
514 * Get a mbuf (not a mbuf cluster!) and zero it.
518 m_getclr(int how, int type)
522 m = m_get(how, type);
524 bzero(m->m_data, MLEN);
529 * Returns an mbuf with an attached cluster.
530 * Because many network drivers use this kind of buffers a lot, it is
531 * convenient to keep a small pool of free buffers of this kind.
532 * Even a small size such as 10 gives about 10% improvement in the
533 * forwarding rate in a bridge or router.
536 m_getcl(int how, short type, int flags)
539 int ocflags = MBTOM(how);
544 if (flags & M_PKTHDR)
545 m = objcache_get(mbufphdrcluster_cache, ocflags);
547 m = objcache_get(mbufcluster_cache, ocflags);
550 if ((how & MB_TRYWAIT) && ntries++ == 0) {
551 struct objcache *reclaimlist[1];
553 if (flags & M_PKTHDR)
554 reclaimlist[0] = mbufcluster_cache;
556 reclaimlist[0] = mbufphdrcluster_cache;
557 if (!objcache_reclaimlist(reclaimlist, 1, ocflags))
574 * Allocate chain of requested length.
577 m_getc(int len, int how, int type)
579 struct mbuf *n, *nfirst = NULL, **ntail = &nfirst;
583 n = m_getl(len, how, type, 0, &nsize);
599 * Allocate len-worth of mbufs and/or mbuf clusters (whatever fits best)
600 * and return a pointer to the head of the allocated chain. If m0 is
601 * non-null, then we assume that it is a single mbuf or an mbuf chain to
602 * which we want len bytes worth of mbufs and/or clusters attached, and so
603 * if we succeed in allocating it, we will just return a pointer to m0.
605 * If we happen to fail at any point during the allocation, we will free
606 * up everything we have already allocated and return NULL.
608 * Deprecated. Use m_getc() and m_cat() instead.
611 m_getm(struct mbuf *m0, int len, int how, int type)
615 nfirst = m_getc(len, how, type);
618 m_last(m0)->m_next = nfirst;
626 * Adds a cluster to a normal mbuf, M_EXT is set on success.
627 * Deprecated. Use m_getcl() instead.
630 m_mclget(struct mbuf *m, int how)
632 struct mbcluster *mcl;
634 KKASSERT((m->m_flags & M_EXT) == 0);
635 mcl = objcache_get(mclmeta_cache, MBTOM(how));
640 /* leave the m_mbufs count intact for original mbuf */
648 struct mbcluster *mcl = arg;
650 atomic_add_int(&mcl->mcl_refs, 1);
656 struct mbcluster *mcl = arg;
658 /* XXX interrupt race. Currently called from a critical section */
659 if (mcl->mcl_refs > 1) {
660 atomic_subtract_int(&mcl->mcl_refs, 1);
662 KKASSERT(mcl->mcl_refs == 1);
664 objcache_put(mclmeta_cache, mcl);
668 extern void db_print_backtrace(void);
671 * Free a single mbuf and any associated external storage. The successor,
672 * if any, is returned.
674 * We do need to check non-first mbuf for m_aux, since some of existing
675 * code does not call M_PREPEND properly.
676 * (example: call to bpf_mtap from drivers)
679 m_free(struct mbuf *m)
683 KASSERT(m->m_type != MT_FREE, ("freeing free mbuf %p", m));
684 --mbtypes[m->m_type];
689 * Make sure the mbuf is in constructed state before returning it
694 KKASSERT(m->m_nextpkt == NULL);
696 if (m->m_nextpkt != NULL) {
698 static int afewtimes = 10;
700 if (afewtimes-- > 0) {
701 printf("mfree: m->m_nextpkt != NULL\n");
702 db_print_backtrace();
708 if (m->m_flags & M_PKTHDR) {
709 m_tag_delete_chain(m); /* eliminate XXX JH */
712 m->m_flags &= (M_EXT | M_EXT_CLUSTER | M_CLCACHE | M_PHCACHE);
715 * Clean the M_PKTHDR state so we can return the mbuf to its original
716 * cache. This is based on the PHCACHE flag which tells us whether
717 * the mbuf was originally allocated out of a packet-header cache
718 * or a non-packet-header cache.
720 if (m->m_flags & M_PHCACHE) {
721 m->m_flags |= M_PKTHDR;
722 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */
723 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */
724 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */
725 SLIST_INIT(&m->m_pkthdr.tags);
729 * Handle remaining flags combinations. M_CLCACHE tells us whether
730 * the mbuf was originally allocated from a cluster cache or not,
731 * and is totally separate from whether the mbuf is currently
732 * associated with a cluster.
735 switch(m->m_flags & (M_CLCACHE | M_EXT | M_EXT_CLUSTER)) {
736 case M_CLCACHE | M_EXT | M_EXT_CLUSTER:
738 * mbuf+cluster cache case. The mbuf was allocated from the
739 * combined mbuf_cluster cache and can be returned to the
740 * cache if the cluster hasn't been shared.
742 if (m_sharecount(m) == 1) {
744 * The cluster has not been shared, we can just
745 * reset the data pointer and return the mbuf
746 * to the cluster cache. Note that the reference
747 * count is left intact (it is still associated with
750 m->m_data = m->m_ext.ext_buf;
751 if (m->m_flags & M_PHCACHE)
752 objcache_put(mbufphdrcluster_cache, m);
754 objcache_put(mbufcluster_cache, m);
757 * Hell. Someone else has a ref on this cluster,
758 * we have to disconnect it which means we can't
759 * put it back into the mbufcluster_cache, we
760 * have to destroy the mbuf.
762 * XXX we could try to connect another cluster to
765 m->m_ext.ext_free(m->m_ext.ext_arg);
766 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER);
767 if (m->m_flags & M_PHCACHE)
768 objcache_dtor(mbufphdrcluster_cache, m);
770 objcache_dtor(mbufcluster_cache, m);
774 case M_EXT | M_EXT_CLUSTER:
776 * Normal cluster associated with an mbuf that was allocated
777 * from the normal mbuf pool rather then the cluster pool.
778 * The cluster has to be independantly disassociated from the
785 * Normal cluster association case, disconnect the cluster from
786 * the mbuf. The cluster may or may not be custom.
788 m->m_ext.ext_free(m->m_ext.ext_arg);
789 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER);
793 * return the mbuf to the mbuf cache.
795 if (m->m_flags & M_PHCACHE) {
796 m->m_data = m->m_pktdat;
797 objcache_put(mbufphdr_cache, m);
799 m->m_data = m->m_dat;
800 objcache_put(mbuf_cache, m);
806 panic("bad mbuf flags %p %08x\n", m, m->m_flags);
814 m_freem(struct mbuf *m)
823 * mbuf utility routines
827 * Lesser-used path for M_PREPEND: allocate new mbuf to prepend to chain and
831 m_prepend(struct mbuf *m, int len, int how)
835 if (m->m_flags & M_PKTHDR)
836 mn = m_gethdr(how, m->m_type);
838 mn = m_get(how, m->m_type);
843 if (m->m_flags & M_PKTHDR)
844 M_MOVE_PKTHDR(mn, m);
854 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
855 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
856 * The wait parameter is a choice of MB_WAIT/MB_DONTWAIT from caller.
857 * Note that the copy is read-only, because clusters are not copied,
858 * only their reference counts are incremented.
861 m_copym(const struct mbuf *m, int off0, int len, int wait)
863 struct mbuf *n, **np;
868 KASSERT(off >= 0, ("m_copym, negative off %d", off));
869 KASSERT(len >= 0, ("m_copym, negative len %d", len));
870 if (off == 0 && m->m_flags & M_PKTHDR)
873 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
883 KASSERT(len == M_COPYALL,
884 ("m_copym, length > size of mbuf chain"));
888 * Because we are sharing any cluster attachment below,
889 * be sure to get an mbuf that does not have a cluster
890 * associated with it.
893 n = m_gethdr(wait, m->m_type);
895 n = m_get(wait, m->m_type);
900 if (!m_dup_pkthdr(n, m, wait))
902 if (len == M_COPYALL)
903 n->m_pkthdr.len -= off0;
905 n->m_pkthdr.len = len;
908 n->m_len = min(len, m->m_len - off);
909 if (m->m_flags & M_EXT) {
910 KKASSERT((n->m_flags & M_EXT) == 0);
911 n->m_data = m->m_data + off;
912 m->m_ext.ext_ref(m->m_ext.ext_arg);
914 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
916 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
919 if (len != M_COPYALL)
935 * Copy an entire packet, including header (which must be present).
936 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
937 * Note that the copy is read-only, because clusters are not copied,
938 * only their reference counts are incremented.
939 * Preserve alignment of the first mbuf so if the creator has left
940 * some room at the beginning (e.g. for inserting protocol headers)
941 * the copies also have the room available.
944 m_copypacket(struct mbuf *m, int how)
946 struct mbuf *top, *n, *o;
948 n = m_gethdr(how, m->m_type);
953 if (!m_dup_pkthdr(n, m, how))
956 if (m->m_flags & M_EXT) {
957 KKASSERT((n->m_flags & M_EXT) == 0);
958 n->m_data = m->m_data;
959 m->m_ext.ext_ref(m->m_ext.ext_arg);
961 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
963 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
964 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
969 o = m_get(how, m->m_type);
977 if (m->m_flags & M_EXT) {
978 KKASSERT((n->m_flags & M_EXT) == 0);
979 n->m_data = m->m_data;
980 m->m_ext.ext_ref(m->m_ext.ext_arg);
982 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
984 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
997 * Copy data from an mbuf chain starting "off" bytes from the beginning,
998 * continuing for "len" bytes, into the indicated buffer.
1001 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
1005 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
1006 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
1008 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
1015 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
1016 count = min(m->m_len - off, len);
1017 bcopy(mtod(m, caddr_t) + off, cp, count);
1026 * Copy a packet header mbuf chain into a completely new chain, including
1027 * copying any mbuf clusters. Use this instead of m_copypacket() when
1028 * you need a writable copy of an mbuf chain.
1031 m_dup(struct mbuf *m, int how)
1033 struct mbuf **p, *top = NULL;
1034 int remain, moff, nsize;
1039 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __func__));
1041 /* While there's more data, get a new mbuf, tack it on, and fill it */
1042 remain = m->m_pkthdr.len;
1045 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
1048 /* Get the next new mbuf */
1049 n = m_getl(remain, how, m->m_type, top == NULL ? M_PKTHDR : 0,
1054 if (!m_dup_pkthdr(n, m, how))
1057 /* Link it into the new chain */
1061 /* Copy data from original mbuf(s) into new mbuf */
1063 while (n->m_len < nsize && m != NULL) {
1064 int chunk = min(nsize - n->m_len, m->m_len - moff);
1066 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
1070 if (moff == m->m_len) {
1076 /* Check correct total mbuf length */
1077 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
1078 ("%s: bogus m_pkthdr.len", __func__));
1090 * Concatenate mbuf chain n to m.
1091 * Both chains must be of the same type (e.g. MT_DATA).
1092 * Any m_pkthdr is not updated.
1095 m_cat(struct mbuf *m, struct mbuf *n)
1099 if (m->m_flags & M_EXT ||
1100 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
1101 /* just join the two chains */
1105 /* splat the data from one into the other */
1106 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1108 m->m_len += n->m_len;
1114 m_adj(struct mbuf *mp, int req_len)
1120 if ((m = mp) == NULL)
1126 while (m != NULL && len > 0) {
1127 if (m->m_len <= len) {
1138 if (mp->m_flags & M_PKTHDR)
1139 m->m_pkthdr.len -= (req_len - len);
1142 * Trim from tail. Scan the mbuf chain,
1143 * calculating its length and finding the last mbuf.
1144 * If the adjustment only affects this mbuf, then just
1145 * adjust and return. Otherwise, rescan and truncate
1146 * after the remaining size.
1152 if (m->m_next == (struct mbuf *)0)
1156 if (m->m_len >= len) {
1158 if (mp->m_flags & M_PKTHDR)
1159 mp->m_pkthdr.len -= len;
1166 * Correct length for chain is "count".
1167 * Find the mbuf with last data, adjust its length,
1168 * and toss data from remaining mbufs on chain.
1171 if (m->m_flags & M_PKTHDR)
1172 m->m_pkthdr.len = count;
1173 for (; m; m = m->m_next) {
1174 if (m->m_len >= count) {
1181 (m = m->m_next) ->m_len = 0;
1186 * Rearrange an mbuf chain so that len bytes are contiguous
1187 * and in the data area of an mbuf (so that mtod will work for a structure
1188 * of size len). Returns the resulting mbuf chain on success, frees it and
1189 * returns null on failure. If there is room, it will add up to
1190 * max_protohdr-len extra bytes to the contiguous region in an attempt to
1191 * avoid being called next time.
1194 m_pullup(struct mbuf *n, int len)
1201 * If first mbuf has no cluster, and has room for len bytes
1202 * without shifting current data, pullup into it,
1203 * otherwise allocate a new mbuf to prepend to the chain.
1205 if (!(n->m_flags & M_EXT) &&
1206 n->m_data + len < &n->m_dat[MLEN] &&
1208 if (n->m_len >= len)
1216 if (n->m_flags & M_PKTHDR)
1217 m = m_gethdr(MB_DONTWAIT, n->m_type);
1219 m = m_get(MB_DONTWAIT, n->m_type);
1223 if (n->m_flags & M_PKTHDR)
1224 M_MOVE_PKTHDR(m, n);
1226 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1228 count = min(min(max(len, max_protohdr), space), n->m_len);
1229 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1239 } while (len > 0 && n);
1253 * Partition an mbuf chain in two pieces, returning the tail --
1254 * all but the first len0 bytes. In case of failure, it returns NULL and
1255 * attempts to restore the chain to its original state.
1257 * Note that the resulting mbufs might be read-only, because the new
1258 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1259 * the "breaking point" happens to lie within a cluster mbuf. Use the
1260 * M_WRITABLE() macro to check for this case.
1263 m_split(struct mbuf *m0, int len0, int wait)
1266 unsigned len = len0, remain;
1268 for (m = m0; m && len > m->m_len; m = m->m_next)
1272 remain = m->m_len - len;
1273 if (m0->m_flags & M_PKTHDR) {
1274 n = m_gethdr(wait, m0->m_type);
1277 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1278 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1279 m0->m_pkthdr.len = len0;
1280 if (m->m_flags & M_EXT)
1282 if (remain > MHLEN) {
1283 /* m can't be the lead packet */
1285 n->m_next = m_split(m, len, wait);
1286 if (n->m_next == NULL) {
1294 MH_ALIGN(n, remain);
1295 } else if (remain == 0) {
1300 n = m_get(wait, m->m_type);
1306 if (m->m_flags & M_EXT) {
1307 KKASSERT((n->m_flags & M_EXT) == 0);
1308 n->m_data = m->m_data + len;
1309 m->m_ext.ext_ref(m->m_ext.ext_arg);
1310 n->m_ext = m->m_ext;
1311 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1313 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1317 n->m_next = m->m_next;
1323 * Routine to copy from device local memory into mbufs.
1324 * Note: "offset" is ill-defined and always called as 0, so ignore it.
1327 m_devget(char *buf, int len, int offset, struct ifnet *ifp,
1328 void (*copy)(volatile const void *from, volatile void *to, size_t length))
1330 struct mbuf *m, *mfirst = NULL, **mtail;
1339 m = m_getl(len, MB_DONTWAIT, MT_DATA, flags, &nsize);
1344 m->m_len = min(len, nsize);
1346 if (flags & M_PKTHDR) {
1347 if (len + max_linkhdr <= nsize)
1348 m->m_data += max_linkhdr;
1349 m->m_pkthdr.rcvif = ifp;
1350 m->m_pkthdr.len = len;
1354 copy(buf, m->m_data, (unsigned)m->m_len);
1365 * Copy data from a buffer back into the indicated mbuf chain,
1366 * starting "off" bytes from the beginning, extending the mbuf
1367 * chain if necessary.
1370 m_copyback(struct mbuf *m0, int off, int len, caddr_t cp)
1373 struct mbuf *m = m0, *n;
1378 while (off > (mlen = m->m_len)) {
1381 if (m->m_next == NULL) {
1382 n = m_getclr(MB_DONTWAIT, m->m_type);
1385 n->m_len = min(MLEN, len + off);
1391 mlen = min (m->m_len - off, len);
1392 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen);
1400 if (m->m_next == NULL) {
1401 n = m_get(MB_DONTWAIT, m->m_type);
1404 n->m_len = min(MLEN, len);
1409 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1410 m->m_pkthdr.len = totlen;
1414 m_print(const struct mbuf *m)
1417 const struct mbuf *m2;
1419 len = m->m_pkthdr.len;
1422 printf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-");
1430 * "Move" mbuf pkthdr from "from" to "to".
1431 * "from" must have M_PKTHDR set, and "to" must be empty.
1434 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
1436 KASSERT((to->m_flags & M_PKTHDR), ("m_move_pkthdr: not packet header"));
1438 to->m_flags |= from->m_flags & M_COPYFLAGS;
1439 to->m_pkthdr = from->m_pkthdr; /* especially tags */
1440 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
1444 * Duplicate "from"'s mbuf pkthdr in "to".
1445 * "from" must have M_PKTHDR set, and "to" must be empty.
1446 * In particular, this does a deep copy of the packet tags.
1449 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how)
1451 KASSERT((to->m_flags & M_PKTHDR), ("m_dup_pkthdr: not packet header"));
1453 to->m_flags = (from->m_flags & M_COPYFLAGS) |
1454 (to->m_flags & ~M_COPYFLAGS);
1455 to->m_pkthdr = from->m_pkthdr;
1456 SLIST_INIT(&to->m_pkthdr.tags);
1457 return (m_tag_copy_chain(to, from, how));
1461 * Defragment a mbuf chain, returning the shortest possible
1462 * chain of mbufs and clusters. If allocation fails and
1463 * this cannot be completed, NULL will be returned, but
1464 * the passed in chain will be unchanged. Upon success,
1465 * the original chain will be freed, and the new chain
1468 * If a non-packet header is passed in, the original
1469 * mbuf (chain?) will be returned unharmed.
1471 * m_defrag_nofree doesn't free the passed in mbuf.
1474 m_defrag(struct mbuf *m0, int how)
1478 if ((m_new = m_defrag_nofree(m0, how)) == NULL)
1486 m_defrag_nofree(struct mbuf *m0, int how)
1488 struct mbuf *m_new = NULL, *m_final = NULL;
1489 int progress = 0, length, nsize;
1491 if (!(m0->m_flags & M_PKTHDR))
1494 #ifdef MBUF_STRESS_TEST
1495 if (m_defragrandomfailures) {
1496 int temp = arc4random() & 0xff;
1502 m_final = m_getl(m0->m_pkthdr.len, how, MT_DATA, M_PKTHDR, &nsize);
1503 if (m_final == NULL)
1505 m_final->m_len = 0; /* in case m0->m_pkthdr.len is zero */
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) {
1518 m_new = m_getl(length, how, MT_DATA, 0, &nsize);
1523 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1525 m_new->m_len = length;
1526 if (m_new != m_final)
1527 m_cat(m_final, m_new);
1530 if (m0->m_next == NULL)
1533 m_defragbytes += m_final->m_pkthdr.len;
1544 * Move data from uio into mbufs.
1547 m_uiomove(struct uio *uio)
1549 struct mbuf *m; /* current working mbuf */
1550 struct mbuf *head = NULL; /* result mbuf chain */
1551 struct mbuf **mp = &head;
1552 int resid = uio->uio_resid, nsize, flags = M_PKTHDR, error;
1555 m = m_getl(resid, MB_WAIT, MT_DATA, flags, &nsize);
1557 m->m_pkthdr.len = 0;
1558 /* Leave room for protocol headers. */
1563 m->m_len = min(nsize, resid);
1564 error = uiomove(mtod(m, caddr_t), m->m_len, uio);
1571 head->m_pkthdr.len += m->m_len;
1573 } while (resid > 0);
1583 m_last(struct mbuf *m)
1591 * Return the number of bytes in an mbuf chain.
1592 * If lastm is not NULL, also return the last mbuf.
1595 m_lengthm(struct mbuf *m, struct mbuf **lastm)
1598 struct mbuf *prev = m;
1611 * Like m_lengthm(), except also keep track of mbuf usage.
1614 m_countm(struct mbuf *m, struct mbuf **lastm, u_int *pmbcnt)
1616 u_int len = 0, mbcnt = 0;
1617 struct mbuf *prev = m;
1622 if (m->m_flags & M_EXT)
1623 mbcnt += m->m_ext.ext_size;