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 1.58 2006/12/17 19:28:30 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/serialize.h>
104 #include <sys/thread2.h>
107 #include <vm/vm_kern.h>
108 #include <vm/vm_extern.h>
111 #include <machine/cpu.h>
115 * mbuf cluster meta-data
120 struct lwkt_serialize mcl_serializer;
123 static void mbinit(void *);
124 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbinit, NULL)
126 static u_long mbtypes[MT_NTYPES];
128 struct mbstat mbstat;
137 #ifdef MBUF_STRESS_TEST
138 int m_defragrandomfailures;
141 struct objcache *mbuf_cache, *mbufphdr_cache;
142 struct objcache *mclmeta_cache;
143 struct objcache *mbufcluster_cache, *mbufphdrcluster_cache;
148 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW,
149 &max_linkhdr, 0, "");
150 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW,
151 &max_protohdr, 0, "");
152 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, "");
153 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW,
154 &max_datalen, 0, "");
155 SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW,
157 SYSCTL_STRUCT(_kern_ipc, KIPC_MBSTAT, mbstat, CTLFLAG_RW, &mbstat, mbstat, "");
158 SYSCTL_OPAQUE(_kern_ipc, OID_AUTO, mbtypes, CTLFLAG_RD, mbtypes,
159 sizeof(mbtypes), "LU", "");
160 SYSCTL_INT(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLFLAG_RW,
161 &nmbclusters, 0, "Maximum number of mbuf clusters available");
162 SYSCTL_INT(_kern_ipc, OID_AUTO, nmbufs, CTLFLAG_RW, &nmbufs, 0,
163 "Maximum number of mbufs available");
165 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
166 &m_defragpackets, 0, "");
167 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
168 &m_defragbytes, 0, "");
169 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
170 &m_defraguseless, 0, "");
171 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
172 &m_defragfailure, 0, "");
173 #ifdef MBUF_STRESS_TEST
174 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
175 &m_defragrandomfailures, 0, "");
178 static MALLOC_DEFINE(M_MBUF, "mbuf", "mbuf");
179 static MALLOC_DEFINE(M_MBUFCL, "mbufcl", "mbufcl");
180 static MALLOC_DEFINE(M_MCLMETA, "mclmeta", "mclmeta");
182 static void m_reclaim (void);
183 static void m_mclref(void *arg);
184 static void m_mclfree(void *arg);
187 #define NMBCLUSTERS (512 + maxusers * 16)
190 #define NMBUFS (nmbclusters * 2)
194 * Perform sanity checks of tunables declared above.
197 tunable_mbinit(void *dummy)
201 * This has to be done before VM init.
203 nmbclusters = NMBCLUSTERS;
204 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
206 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs);
208 if (nmbufs < nmbclusters * 2)
209 nmbufs = nmbclusters * 2;
213 SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_mbinit, NULL);
215 /* "number of clusters of pages" */
221 * The mbuf object cache only guarantees that m_next and m_nextpkt are
222 * NULL and that m_data points to the beginning of the data area. In
223 * particular, m_len and m_pkthdr.len are uninitialized. It is the
224 * responsibility of the caller to initialize those fields before use.
227 static boolean_t __inline
228 mbuf_ctor(void *obj, void *private, int ocflags)
230 struct mbuf *m = obj;
234 m->m_data = m->m_dat;
241 * Initialize the mbuf and the packet header fields.
244 mbufphdr_ctor(void *obj, void *private, int ocflags)
246 struct mbuf *m = obj;
250 m->m_data = m->m_pktdat;
251 m->m_flags = M_PKTHDR | M_PHCACHE;
253 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */
254 SLIST_INIT(&m->m_pkthdr.tags);
255 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */
256 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */
262 * A mbcluster object consists of 2K (MCLBYTES) cluster and a refcount.
265 mclmeta_ctor(void *obj, void *private, int ocflags)
267 struct mbcluster *cl = obj;
270 if (ocflags & M_NOWAIT)
271 buf = kmalloc(MCLBYTES, M_MBUFCL, M_NOWAIT | M_ZERO);
273 buf = kmalloc(MCLBYTES, M_MBUFCL, M_INTWAIT | M_ZERO);
278 lwkt_serialize_init(&cl->mcl_serializer);
283 mclmeta_dtor(void *obj, void *private)
285 struct mbcluster *mcl = obj;
287 KKASSERT(mcl->mcl_refs == 0);
288 kfree(mcl->mcl_data, M_MBUFCL);
292 linkcluster(struct mbuf *m, struct mbcluster *cl)
295 * Add the cluster to the mbuf. The caller will detect that the
296 * mbuf now has an attached cluster.
298 m->m_ext.ext_arg = cl;
299 m->m_ext.ext_buf = cl->mcl_data;
300 m->m_ext.ext_ref = m_mclref;
301 m->m_ext.ext_free = m_mclfree;
302 m->m_ext.ext_size = MCLBYTES;
303 atomic_add_int(&cl->mcl_refs, 1);
305 m->m_data = m->m_ext.ext_buf;
306 m->m_flags |= M_EXT | M_EXT_CLUSTER;
310 mbufphdrcluster_ctor(void *obj, void *private, int ocflags)
312 struct mbuf *m = obj;
313 struct mbcluster *cl;
315 mbufphdr_ctor(obj, private, ocflags);
316 cl = objcache_get(mclmeta_cache, ocflags);
319 m->m_flags |= M_CLCACHE;
325 mbufcluster_ctor(void *obj, void *private, int ocflags)
327 struct mbuf *m = obj;
328 struct mbcluster *cl;
330 mbuf_ctor(obj, private, ocflags);
331 cl = objcache_get(mclmeta_cache, ocflags);
334 m->m_flags |= M_CLCACHE;
340 * Used for both the cluster and cluster PHDR caches.
342 * The mbuf may have lost its cluster due to sharing, deal
343 * with the situation by checking M_EXT.
346 mbufcluster_dtor(void *obj, void *private)
348 struct mbuf *m = obj;
349 struct mbcluster *mcl;
351 if (m->m_flags & M_EXT) {
352 KKASSERT((m->m_flags & M_EXT_CLUSTER) != 0);
353 mcl = m->m_ext.ext_arg;
354 KKASSERT(mcl->mcl_refs == 1);
356 objcache_put(mclmeta_cache, mcl);
360 struct objcache_malloc_args mbuf_malloc_args = { MSIZE, M_MBUF };
361 struct objcache_malloc_args mclmeta_malloc_args =
362 { sizeof(struct mbcluster), M_MCLMETA };
368 mbstat.m_msize = MSIZE;
369 mbstat.m_mclbytes = MCLBYTES;
370 mbstat.m_minclsize = MINCLSIZE;
371 mbstat.m_mlen = MLEN;
372 mbstat.m_mhlen = MHLEN;
374 mbuf_cache = objcache_create("mbuf", nmbufs, 0,
375 mbuf_ctor, NULL, NULL,
376 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
377 mbufphdr_cache = objcache_create("mbuf pkt hdr", nmbufs, 64,
378 mbufphdr_ctor, NULL, NULL,
379 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
380 mclmeta_cache = objcache_create("cluster mbuf", nmbclusters , 0,
381 mclmeta_ctor, mclmeta_dtor, NULL,
382 objcache_malloc_alloc, objcache_malloc_free, &mclmeta_malloc_args);
383 mbufcluster_cache = objcache_create("mbuf + cluster", nmbclusters, 0,
384 mbufcluster_ctor, mbufcluster_dtor, NULL,
385 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
386 mbufphdrcluster_cache = objcache_create("mbuf pkt hdr + cluster",
387 nmbclusters, 64, mbufphdrcluster_ctor, mbufcluster_dtor, NULL,
388 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
393 * Return the number of references to this mbuf's data. 0 is returned
394 * if the mbuf is not M_EXT, a reference count is returned if it is
395 * M_EXT | M_EXT_CLUSTER, and 99 is returned if it is a special M_EXT.
398 m_sharecount(struct mbuf *m)
400 switch (m->m_flags & (M_EXT | M_EXT_CLUSTER)) {
405 case M_EXT | M_EXT_CLUSTER:
406 return (((struct mbcluster *)m->m_ext.ext_arg)->mcl_refs);
409 return (0); /* to shut up compiler */
413 * change mbuf to new type
416 m_chtype(struct mbuf *m, int type)
420 --mbtypes[m->m_type];
432 SLIST_FOREACH(dp, &domains, dom_next) {
433 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
443 updatestats(struct mbuf *m, int type)
457 m_get(int how, int type)
461 int ocf = MBTOM(how);
465 m = objcache_get(mbuf_cache, ocf);
468 if ((how & MB_TRYWAIT) && ntries++ == 0) {
469 struct objcache *reclaimlist[] = {
471 mbufcluster_cache, mbufphdrcluster_cache
473 const int nreclaims = __arysize(reclaimlist);
475 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf))
482 updatestats(m, type);
487 m_gethdr(int how, int type)
490 int ocf = MBTOM(how);
495 m = objcache_get(mbufphdr_cache, ocf);
498 if ((how & MB_TRYWAIT) && ntries++ == 0) {
499 struct objcache *reclaimlist[] = {
501 mbufcluster_cache, mbufphdrcluster_cache
503 const int nreclaims = __arysize(reclaimlist);
505 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf))
512 updatestats(m, type);
517 * Get a mbuf (not a mbuf cluster!) and zero it.
521 m_getclr(int how, int type)
525 m = m_get(how, type);
527 bzero(m->m_data, MLEN);
532 * Returns an mbuf with an attached cluster.
533 * Because many network drivers use this kind of buffers a lot, it is
534 * convenient to keep a small pool of free buffers of this kind.
535 * Even a small size such as 10 gives about 10% improvement in the
536 * forwarding rate in a bridge or router.
539 m_getcl(int how, short type, int flags)
542 int ocflags = MBTOM(how);
547 if (flags & M_PKTHDR)
548 m = objcache_get(mbufphdrcluster_cache, ocflags);
550 m = objcache_get(mbufcluster_cache, ocflags);
553 if ((how & MB_TRYWAIT) && ntries++ == 0) {
554 struct objcache *reclaimlist[1];
556 if (flags & M_PKTHDR)
557 reclaimlist[0] = mbufcluster_cache;
559 reclaimlist[0] = mbufphdrcluster_cache;
560 if (!objcache_reclaimlist(reclaimlist, 1, ocflags))
577 * Allocate chain of requested length.
580 m_getc(int len, int how, int type)
582 struct mbuf *n, *nfirst = NULL, **ntail = &nfirst;
586 n = m_getl(len, how, type, 0, &nsize);
602 * Allocate len-worth of mbufs and/or mbuf clusters (whatever fits best)
603 * and return a pointer to the head of the allocated chain. If m0 is
604 * non-null, then we assume that it is a single mbuf or an mbuf chain to
605 * which we want len bytes worth of mbufs and/or clusters attached, and so
606 * if we succeed in allocating it, we will just return a pointer to m0.
608 * If we happen to fail at any point during the allocation, we will free
609 * up everything we have already allocated and return NULL.
611 * Deprecated. Use m_getc() and m_cat() instead.
614 m_getm(struct mbuf *m0, int len, int type, int how)
618 nfirst = m_getc(len, how, type);
621 m_last(m0)->m_next = nfirst;
629 * Adds a cluster to a normal mbuf, M_EXT is set on success.
630 * Deprecated. Use m_getcl() instead.
633 m_mclget(struct mbuf *m, int how)
635 struct mbcluster *mcl;
637 KKASSERT((m->m_flags & M_EXT) == 0);
638 mcl = objcache_get(mclmeta_cache, MBTOM(how));
643 /* leave the m_mbufs count intact for original mbuf */
649 * Updates to mbcluster must be MPSAFE. Only an entity which already has
650 * a reference to the cluster can ref it, so we are in no danger of
651 * racing an add with a subtract. But the operation must still be atomic
652 * since multiple entities may have a reference on the cluster.
654 * m_mclfree() is almost the same but it must contend with two entities
655 * freeing the cluster at the same time. If there is only one reference
656 * count we are the only entity referencing the cluster and no further
657 * locking is required. Otherwise we must protect against a race to 0
658 * with the serializer.
663 struct mbcluster *mcl = arg;
665 atomic_add_int(&mcl->mcl_refs, 1);
671 struct mbcluster *mcl = arg;
673 if (mcl->mcl_refs == 1) {
675 objcache_put(mclmeta_cache, mcl);
677 lwkt_serialize_enter(&mcl->mcl_serializer);
678 if (mcl->mcl_refs > 1) {
679 atomic_subtract_int(&mcl->mcl_refs, 1);
680 lwkt_serialize_exit(&mcl->mcl_serializer);
682 lwkt_serialize_exit(&mcl->mcl_serializer);
683 KKASSERT(mcl->mcl_refs == 1);
685 objcache_put(mclmeta_cache, mcl);
690 extern void db_print_backtrace(void);
693 * Free a single mbuf and any associated external storage. The successor,
694 * if any, is returned.
696 * We do need to check non-first mbuf for m_aux, since some of existing
697 * code does not call M_PREPEND properly.
698 * (example: call to bpf_mtap from drivers)
701 m_free(struct mbuf *m)
705 KASSERT(m->m_type != MT_FREE, ("freeing free mbuf %p", m));
706 --mbtypes[m->m_type];
711 * Make sure the mbuf is in constructed state before returning it
716 KKASSERT(m->m_nextpkt == NULL);
718 if (m->m_nextpkt != NULL) {
720 static int afewtimes = 10;
722 if (afewtimes-- > 0) {
723 printf("mfree: m->m_nextpkt != NULL\n");
724 db_print_backtrace();
730 if (m->m_flags & M_PKTHDR) {
731 m_tag_delete_chain(m); /* eliminate XXX JH */
734 m->m_flags &= (M_EXT | M_EXT_CLUSTER | M_CLCACHE | M_PHCACHE);
737 * Clean the M_PKTHDR state so we can return the mbuf to its original
738 * cache. This is based on the PHCACHE flag which tells us whether
739 * the mbuf was originally allocated out of a packet-header cache
740 * or a non-packet-header cache.
742 if (m->m_flags & M_PHCACHE) {
743 m->m_flags |= M_PKTHDR;
744 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */
745 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */
746 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */
747 SLIST_INIT(&m->m_pkthdr.tags);
751 * Handle remaining flags combinations. M_CLCACHE tells us whether
752 * the mbuf was originally allocated from a cluster cache or not,
753 * and is totally separate from whether the mbuf is currently
754 * associated with a cluster.
757 switch(m->m_flags & (M_CLCACHE | M_EXT | M_EXT_CLUSTER)) {
758 case M_CLCACHE | M_EXT | M_EXT_CLUSTER:
760 * mbuf+cluster cache case. The mbuf was allocated from the
761 * combined mbuf_cluster cache and can be returned to the
762 * cache if the cluster hasn't been shared.
764 if (m_sharecount(m) == 1) {
766 * The cluster has not been shared, we can just
767 * reset the data pointer and return the mbuf
768 * to the cluster cache. Note that the reference
769 * count is left intact (it is still associated with
772 m->m_data = m->m_ext.ext_buf;
773 if (m->m_flags & M_PHCACHE)
774 objcache_put(mbufphdrcluster_cache, m);
776 objcache_put(mbufcluster_cache, m);
780 * Hell. Someone else has a ref on this cluster,
781 * we have to disconnect it which means we can't
782 * put it back into the mbufcluster_cache, we
783 * have to destroy the mbuf.
785 * Other mbuf references to the cluster will typically
786 * be M_EXT | M_EXT_CLUSTER but without M_CLCACHE.
788 * XXX we could try to connect another cluster to
791 m->m_ext.ext_free(m->m_ext.ext_arg);
792 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER);
793 if (m->m_flags & M_PHCACHE)
794 objcache_dtor(mbufphdrcluster_cache, m);
796 objcache_dtor(mbufcluster_cache, m);
799 case M_EXT | M_EXT_CLUSTER:
801 * Normal cluster associated with an mbuf that was allocated
802 * from the normal mbuf pool rather then the cluster pool.
803 * The cluster has to be independantly disassociated from the
806 if (m_sharecount(m) == 1)
811 * Normal cluster association case, disconnect the cluster from
812 * the mbuf. The cluster may or may not be custom.
814 m->m_ext.ext_free(m->m_ext.ext_arg);
815 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER);
819 * return the mbuf to the mbuf cache.
821 if (m->m_flags & M_PHCACHE) {
822 m->m_data = m->m_pktdat;
823 objcache_put(mbufphdr_cache, m);
825 m->m_data = m->m_dat;
826 objcache_put(mbuf_cache, m);
832 panic("bad mbuf flags %p %08x\n", m, m->m_flags);
840 m_freem(struct mbuf *m)
849 * mbuf utility routines
853 * Lesser-used path for M_PREPEND: allocate new mbuf to prepend to chain and
857 m_prepend(struct mbuf *m, int len, int how)
861 if (m->m_flags & M_PKTHDR)
862 mn = m_gethdr(how, m->m_type);
864 mn = m_get(how, m->m_type);
869 if (m->m_flags & M_PKTHDR)
870 M_MOVE_PKTHDR(mn, m);
880 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
881 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
882 * The wait parameter is a choice of MB_WAIT/MB_DONTWAIT from caller.
883 * Note that the copy is read-only, because clusters are not copied,
884 * only their reference counts are incremented.
887 m_copym(const struct mbuf *m, int off0, int len, int wait)
889 struct mbuf *n, **np;
894 KASSERT(off >= 0, ("m_copym, negative off %d", off));
895 KASSERT(len >= 0, ("m_copym, negative len %d", len));
896 if (off == 0 && m->m_flags & M_PKTHDR)
899 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
909 KASSERT(len == M_COPYALL,
910 ("m_copym, length > size of mbuf chain"));
914 * Because we are sharing any cluster attachment below,
915 * be sure to get an mbuf that does not have a cluster
916 * associated with it.
919 n = m_gethdr(wait, m->m_type);
921 n = m_get(wait, m->m_type);
926 if (!m_dup_pkthdr(n, m, wait))
928 if (len == M_COPYALL)
929 n->m_pkthdr.len -= off0;
931 n->m_pkthdr.len = len;
934 n->m_len = min(len, m->m_len - off);
935 if (m->m_flags & M_EXT) {
936 KKASSERT((n->m_flags & M_EXT) == 0);
937 n->m_data = m->m_data + off;
938 m->m_ext.ext_ref(m->m_ext.ext_arg);
940 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
942 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
945 if (len != M_COPYALL)
961 * Copy an entire packet, including header (which must be present).
962 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
963 * Note that the copy is read-only, because clusters are not copied,
964 * only their reference counts are incremented.
965 * Preserve alignment of the first mbuf so if the creator has left
966 * some room at the beginning (e.g. for inserting protocol headers)
967 * the copies also have the room available.
970 m_copypacket(struct mbuf *m, int how)
972 struct mbuf *top, *n, *o;
974 n = m_gethdr(how, m->m_type);
979 if (!m_dup_pkthdr(n, m, how))
982 if (m->m_flags & M_EXT) {
983 KKASSERT((n->m_flags & M_EXT) == 0);
984 n->m_data = m->m_data;
985 m->m_ext.ext_ref(m->m_ext.ext_arg);
987 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
989 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
990 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
995 o = m_get(how, m->m_type);
1002 n->m_len = m->m_len;
1003 if (m->m_flags & M_EXT) {
1004 KKASSERT((n->m_flags & M_EXT) == 0);
1005 n->m_data = m->m_data;
1006 m->m_ext.ext_ref(m->m_ext.ext_arg);
1007 n->m_ext = m->m_ext;
1008 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1010 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
1023 * Copy data from an mbuf chain starting "off" bytes from the beginning,
1024 * continuing for "len" bytes, into the indicated buffer.
1027 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
1031 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
1032 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
1034 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
1041 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
1042 count = min(m->m_len - off, len);
1043 bcopy(mtod(m, caddr_t) + off, cp, count);
1052 * Copy a packet header mbuf chain into a completely new chain, including
1053 * copying any mbuf clusters. Use this instead of m_copypacket() when
1054 * you need a writable copy of an mbuf chain.
1057 m_dup(struct mbuf *m, int how)
1059 struct mbuf **p, *top = NULL;
1060 int remain, moff, nsize;
1065 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __func__));
1067 /* While there's more data, get a new mbuf, tack it on, and fill it */
1068 remain = m->m_pkthdr.len;
1071 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
1074 /* Get the next new mbuf */
1075 n = m_getl(remain, how, m->m_type, top == NULL ? M_PKTHDR : 0,
1080 if (!m_dup_pkthdr(n, m, how))
1083 /* Link it into the new chain */
1087 /* Copy data from original mbuf(s) into new mbuf */
1089 while (n->m_len < nsize && m != NULL) {
1090 int chunk = min(nsize - n->m_len, m->m_len - moff);
1092 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
1096 if (moff == m->m_len) {
1102 /* Check correct total mbuf length */
1103 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
1104 ("%s: bogus m_pkthdr.len", __func__));
1116 * Concatenate mbuf chain n to m.
1117 * Both chains must be of the same type (e.g. MT_DATA).
1118 * Any m_pkthdr is not updated.
1121 m_cat(struct mbuf *m, struct mbuf *n)
1125 if (m->m_flags & M_EXT ||
1126 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
1127 /* just join the two chains */
1131 /* splat the data from one into the other */
1132 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1134 m->m_len += n->m_len;
1140 m_adj(struct mbuf *mp, int req_len)
1146 if ((m = mp) == NULL)
1152 while (m != NULL && len > 0) {
1153 if (m->m_len <= len) {
1164 if (mp->m_flags & M_PKTHDR)
1165 m->m_pkthdr.len -= (req_len - len);
1168 * Trim from tail. Scan the mbuf chain,
1169 * calculating its length and finding the last mbuf.
1170 * If the adjustment only affects this mbuf, then just
1171 * adjust and return. Otherwise, rescan and truncate
1172 * after the remaining size.
1178 if (m->m_next == (struct mbuf *)0)
1182 if (m->m_len >= len) {
1184 if (mp->m_flags & M_PKTHDR)
1185 mp->m_pkthdr.len -= len;
1192 * Correct length for chain is "count".
1193 * Find the mbuf with last data, adjust its length,
1194 * and toss data from remaining mbufs on chain.
1197 if (m->m_flags & M_PKTHDR)
1198 m->m_pkthdr.len = count;
1199 for (; m; m = m->m_next) {
1200 if (m->m_len >= count) {
1207 (m = m->m_next) ->m_len = 0;
1212 * Rearrange an mbuf chain so that len bytes are contiguous
1213 * and in the data area of an mbuf (so that mtod will work for a structure
1214 * of size len). Returns the resulting mbuf chain on success, frees it and
1215 * returns null on failure. If there is room, it will add up to
1216 * max_protohdr-len extra bytes to the contiguous region in an attempt to
1217 * avoid being called next time.
1220 m_pullup(struct mbuf *n, int len)
1227 * If first mbuf has no cluster, and has room for len bytes
1228 * without shifting current data, pullup into it,
1229 * otherwise allocate a new mbuf to prepend to the chain.
1231 if (!(n->m_flags & M_EXT) &&
1232 n->m_data + len < &n->m_dat[MLEN] &&
1234 if (n->m_len >= len)
1242 if (n->m_flags & M_PKTHDR)
1243 m = m_gethdr(MB_DONTWAIT, n->m_type);
1245 m = m_get(MB_DONTWAIT, n->m_type);
1249 if (n->m_flags & M_PKTHDR)
1250 M_MOVE_PKTHDR(m, n);
1252 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1254 count = min(min(max(len, max_protohdr), space), n->m_len);
1255 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1265 } while (len > 0 && n);
1279 * Partition an mbuf chain in two pieces, returning the tail --
1280 * all but the first len0 bytes. In case of failure, it returns NULL and
1281 * attempts to restore the chain to its original state.
1283 * Note that the resulting mbufs might be read-only, because the new
1284 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1285 * the "breaking point" happens to lie within a cluster mbuf. Use the
1286 * M_WRITABLE() macro to check for this case.
1289 m_split(struct mbuf *m0, int len0, int wait)
1292 unsigned len = len0, remain;
1294 for (m = m0; m && len > m->m_len; m = m->m_next)
1298 remain = m->m_len - len;
1299 if (m0->m_flags & M_PKTHDR) {
1300 n = m_gethdr(wait, m0->m_type);
1303 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1304 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1305 m0->m_pkthdr.len = len0;
1306 if (m->m_flags & M_EXT)
1308 if (remain > MHLEN) {
1309 /* m can't be the lead packet */
1311 n->m_next = m_split(m, len, wait);
1312 if (n->m_next == NULL) {
1320 MH_ALIGN(n, remain);
1321 } else if (remain == 0) {
1326 n = m_get(wait, m->m_type);
1332 if (m->m_flags & M_EXT) {
1333 KKASSERT((n->m_flags & M_EXT) == 0);
1334 n->m_data = m->m_data + len;
1335 m->m_ext.ext_ref(m->m_ext.ext_arg);
1336 n->m_ext = m->m_ext;
1337 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1339 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1343 n->m_next = m->m_next;
1349 * Routine to copy from device local memory into mbufs.
1350 * Note: "offset" is ill-defined and always called as 0, so ignore it.
1353 m_devget(char *buf, int len, int offset, struct ifnet *ifp,
1354 void (*copy)(volatile const void *from, volatile void *to, size_t length))
1356 struct mbuf *m, *mfirst = NULL, **mtail;
1365 m = m_getl(len, MB_DONTWAIT, MT_DATA, flags, &nsize);
1370 m->m_len = min(len, nsize);
1372 if (flags & M_PKTHDR) {
1373 if (len + max_linkhdr <= nsize)
1374 m->m_data += max_linkhdr;
1375 m->m_pkthdr.rcvif = ifp;
1376 m->m_pkthdr.len = len;
1380 copy(buf, m->m_data, (unsigned)m->m_len);
1391 * Copy data from a buffer back into the indicated mbuf chain,
1392 * starting "off" bytes from the beginning, extending the mbuf
1393 * chain if necessary.
1396 m_copyback(struct mbuf *m0, int off, int len, caddr_t cp)
1399 struct mbuf *m = m0, *n;
1404 while (off > (mlen = m->m_len)) {
1407 if (m->m_next == NULL) {
1408 n = m_getclr(MB_DONTWAIT, m->m_type);
1411 n->m_len = min(MLEN, len + off);
1417 mlen = min (m->m_len - off, len);
1418 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen);
1426 if (m->m_next == NULL) {
1427 n = m_get(MB_DONTWAIT, m->m_type);
1430 n->m_len = min(MLEN, len);
1435 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1436 m->m_pkthdr.len = totlen;
1440 m_print(const struct mbuf *m)
1443 const struct mbuf *m2;
1445 len = m->m_pkthdr.len;
1448 printf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-");
1456 * "Move" mbuf pkthdr from "from" to "to".
1457 * "from" must have M_PKTHDR set, and "to" must be empty.
1460 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
1462 KASSERT((to->m_flags & M_PKTHDR), ("m_move_pkthdr: not packet header"));
1464 to->m_flags |= from->m_flags & M_COPYFLAGS;
1465 to->m_pkthdr = from->m_pkthdr; /* especially tags */
1466 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
1470 * Duplicate "from"'s mbuf pkthdr in "to".
1471 * "from" must have M_PKTHDR set, and "to" must be empty.
1472 * In particular, this does a deep copy of the packet tags.
1475 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how)
1477 KASSERT((to->m_flags & M_PKTHDR), ("m_dup_pkthdr: not packet header"));
1479 to->m_flags = (from->m_flags & M_COPYFLAGS) |
1480 (to->m_flags & ~M_COPYFLAGS);
1481 to->m_pkthdr = from->m_pkthdr;
1482 SLIST_INIT(&to->m_pkthdr.tags);
1483 return (m_tag_copy_chain(to, from, how));
1487 * Defragment a mbuf chain, returning the shortest possible
1488 * chain of mbufs and clusters. If allocation fails and
1489 * this cannot be completed, NULL will be returned, but
1490 * the passed in chain will be unchanged. Upon success,
1491 * the original chain will be freed, and the new chain
1494 * If a non-packet header is passed in, the original
1495 * mbuf (chain?) will be returned unharmed.
1497 * m_defrag_nofree doesn't free the passed in mbuf.
1500 m_defrag(struct mbuf *m0, int how)
1504 if ((m_new = m_defrag_nofree(m0, how)) == NULL)
1512 m_defrag_nofree(struct mbuf *m0, int how)
1514 struct mbuf *m_new = NULL, *m_final = NULL;
1515 int progress = 0, length, nsize;
1517 if (!(m0->m_flags & M_PKTHDR))
1520 #ifdef MBUF_STRESS_TEST
1521 if (m_defragrandomfailures) {
1522 int temp = karc4random() & 0xff;
1528 m_final = m_getl(m0->m_pkthdr.len, how, MT_DATA, M_PKTHDR, &nsize);
1529 if (m_final == NULL)
1531 m_final->m_len = 0; /* in case m0->m_pkthdr.len is zero */
1533 if (m_dup_pkthdr(m_final, m0, how) == NULL)
1538 while (progress < m0->m_pkthdr.len) {
1539 length = m0->m_pkthdr.len - progress;
1540 if (length > MCLBYTES)
1543 if (m_new == NULL) {
1544 m_new = m_getl(length, how, MT_DATA, 0, &nsize);
1549 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1551 m_new->m_len = length;
1552 if (m_new != m_final)
1553 m_cat(m_final, m_new);
1556 if (m0->m_next == NULL)
1559 m_defragbytes += m_final->m_pkthdr.len;
1570 * Move data from uio into mbufs.
1573 m_uiomove(struct uio *uio)
1575 struct mbuf *m; /* current working mbuf */
1576 struct mbuf *head = NULL; /* result mbuf chain */
1577 struct mbuf **mp = &head;
1578 int resid = uio->uio_resid, nsize, flags = M_PKTHDR, error;
1581 m = m_getl(resid, MB_WAIT, MT_DATA, flags, &nsize);
1583 m->m_pkthdr.len = 0;
1584 /* Leave room for protocol headers. */
1589 m->m_len = min(nsize, resid);
1590 error = uiomove(mtod(m, caddr_t), m->m_len, uio);
1597 head->m_pkthdr.len += m->m_len;
1599 } while (resid > 0);
1609 m_last(struct mbuf *m)
1617 * Return the number of bytes in an mbuf chain.
1618 * If lastm is not NULL, also return the last mbuf.
1621 m_lengthm(struct mbuf *m, struct mbuf **lastm)
1624 struct mbuf *prev = m;
1637 * Like m_lengthm(), except also keep track of mbuf usage.
1640 m_countm(struct mbuf *m, struct mbuf **lastm, u_int *pmbcnt)
1642 u_int len = 0, mbcnt = 0;
1643 struct mbuf *prev = m;
1648 if (m->m_flags & M_EXT)
1649 mbcnt += m->m_ext.ext_size;