4 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved.
5 * Copyright (c) 2004 The DragonFly Project. All rights reserved.
7 * This code is derived from software contributed to The DragonFly Project
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of The DragonFly Project nor the names of its
19 * contributors may be used to endorse or promote products derived
20 * from this software without specific, prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
25 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
26 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
28 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
29 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
30 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
31 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
32 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * Copyright (c) 1982, 1986, 1988, 1991, 1993
38 * The Regents of the University of California. All rights reserved.
40 * Redistribution and use in source and binary forms, with or without
41 * modification, are permitted provided that the following conditions
43 * 1. Redistributions of source code must retain the above copyright
44 * notice, this list of conditions and the following disclaimer.
45 * 2. Redistributions in binary form must reproduce the above copyright
46 * notice, this list of conditions and the following disclaimer in the
47 * documentation and/or other materials provided with the distribution.
48 * 3. All advertising materials mentioning features or use of this software
49 * must display the following acknowledgement:
50 * This product includes software developed by the University of
51 * California, Berkeley and its contributors.
52 * 4. Neither the name of the University nor the names of its contributors
53 * may be used to endorse or promote products derived from this software
54 * without specific prior written permission.
56 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
57 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
58 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
59 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
60 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
61 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
62 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
63 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
64 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
65 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94
69 * $FreeBSD: src/sys/kern/uipc_mbuf.c,v 1.51.2.24 2003/04/15 06:59:29 silby Exp $
70 * $DragonFly: src/sys/kern/uipc_mbuf.c,v 1.70 2008/11/20 14:21:01 sephe Exp $
73 #include "opt_param.h"
74 #include "opt_mbuf_stress_test.h"
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/malloc.h>
79 #include <sys/kernel.h>
80 #include <sys/sysctl.h>
81 #include <sys/domain.h>
82 #include <sys/objcache.h>
84 #include <sys/protosw.h>
86 #include <sys/thread.h>
87 #include <sys/globaldata.h>
89 #include <sys/thread2.h>
90 #include <sys/spinlock2.h>
92 #include <machine/atomic.h>
93 #include <machine/limits.h>
96 #include <vm/vm_kern.h>
97 #include <vm/vm_extern.h>
100 #include <machine/cpu.h>
104 * mbuf cluster meta-data
112 * mbuf tracking for debugging purposes
116 static MALLOC_DEFINE(M_MTRACK, "mtrack", "mtrack");
119 RB_HEAD(mbuf_rb_tree, mbtrack);
120 RB_PROTOTYPE2(mbuf_rb_tree, mbtrack, rb_node, mbtrack_cmp, struct mbuf *);
123 RB_ENTRY(mbtrack) rb_node;
129 mbtrack_cmp(struct mbtrack *mb1, struct mbtrack *mb2)
138 RB_GENERATE2(mbuf_rb_tree, mbtrack, rb_node, mbtrack_cmp, struct mbuf *, m);
140 struct mbuf_rb_tree mbuf_track_root;
141 static struct spinlock mbuf_track_spin = SPINLOCK_INITIALIZER(mbuf_track_spin);
144 mbuftrack(struct mbuf *m)
148 mbt = kmalloc(sizeof(*mbt), M_MTRACK, M_INTWAIT|M_ZERO);
149 spin_lock(&mbuf_track_spin);
151 if (mbuf_rb_tree_RB_INSERT(&mbuf_track_root, mbt)) {
152 spin_unlock(&mbuf_track_spin);
153 panic("mbuftrack: mbuf %p already being tracked\n", m);
155 spin_unlock(&mbuf_track_spin);
159 mbufuntrack(struct mbuf *m)
163 spin_lock(&mbuf_track_spin);
164 mbt = mbuf_rb_tree_RB_LOOKUP(&mbuf_track_root, m);
166 spin_unlock(&mbuf_track_spin);
167 panic("mbufuntrack: mbuf %p was not tracked\n", m);
169 mbuf_rb_tree_RB_REMOVE(&mbuf_track_root, mbt);
170 spin_unlock(&mbuf_track_spin);
171 kfree(mbt, M_MTRACK);
176 mbuftrackid(struct mbuf *m, int trackid)
181 spin_lock(&mbuf_track_spin);
185 mbt = mbuf_rb_tree_RB_LOOKUP(&mbuf_track_root, m);
187 spin_unlock(&mbuf_track_spin);
188 panic("mbuftrackid: mbuf %p not tracked", m);
190 mbt->trackid = trackid;
195 spin_unlock(&mbuf_track_spin);
199 mbuftrack_callback(struct mbtrack *mbt, void *arg)
201 struct sysctl_req *req = arg;
205 ksnprintf(buf, sizeof(buf), "mbuf %p track %d\n", mbt->m, mbt->trackid);
207 spin_unlock(&mbuf_track_spin);
208 error = SYSCTL_OUT(req, buf, strlen(buf));
209 spin_lock(&mbuf_track_spin);
216 mbuftrack_show(SYSCTL_HANDLER_ARGS)
220 spin_lock(&mbuf_track_spin);
221 error = mbuf_rb_tree_RB_SCAN(&mbuf_track_root, NULL,
222 mbuftrack_callback, req);
223 spin_unlock(&mbuf_track_spin);
226 SYSCTL_PROC(_kern_ipc, OID_AUTO, showmbufs, CTLFLAG_RD|CTLTYPE_STRING,
227 0, 0, mbuftrack_show, "A", "Show all in-use mbufs");
232 #define mbufuntrack(m)
236 static void mbinit(void *);
237 SYSINIT(mbuf, SI_BOOT2_MACHDEP, SI_ORDER_FIRST, mbinit, NULL)
239 static u_long mbtypes[SMP_MAXCPU][MT_NTYPES];
241 static struct mbstat mbstat[SMP_MAXCPU];
250 #ifdef MBUF_STRESS_TEST
251 int m_defragrandomfailures;
254 struct objcache *mbuf_cache, *mbufphdr_cache;
255 struct objcache *mclmeta_cache;
256 struct objcache *mbufcluster_cache, *mbufphdrcluster_cache;
261 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW,
262 &max_linkhdr, 0, "");
263 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW,
264 &max_protohdr, 0, "");
265 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, "");
266 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW,
267 &max_datalen, 0, "");
268 SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW,
270 static int do_mbstat(SYSCTL_HANDLER_ARGS);
272 SYSCTL_PROC(_kern_ipc, KIPC_MBSTAT, mbstat, CTLTYPE_STRUCT|CTLFLAG_RD,
273 0, 0, do_mbstat, "S,mbstat", "");
275 static int do_mbtypes(SYSCTL_HANDLER_ARGS);
277 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbtypes, CTLTYPE_ULONG|CTLFLAG_RD,
278 0, 0, do_mbtypes, "LU", "");
281 do_mbstat(SYSCTL_HANDLER_ARGS)
283 struct mbstat mbstat_total;
284 struct mbstat *mbstat_totalp;
287 bzero(&mbstat_total, sizeof(mbstat_total));
288 mbstat_totalp = &mbstat_total;
290 for (i = 0; i < ncpus; i++)
292 mbstat_total.m_mbufs += mbstat[i].m_mbufs;
293 mbstat_total.m_clusters += mbstat[i].m_clusters;
294 mbstat_total.m_spare += mbstat[i].m_spare;
295 mbstat_total.m_clfree += mbstat[i].m_clfree;
296 mbstat_total.m_drops += mbstat[i].m_drops;
297 mbstat_total.m_wait += mbstat[i].m_wait;
298 mbstat_total.m_drain += mbstat[i].m_drain;
299 mbstat_total.m_mcfail += mbstat[i].m_mcfail;
300 mbstat_total.m_mpfail += mbstat[i].m_mpfail;
304 * The following fields are not cumulative fields so just
305 * get their values once.
307 mbstat_total.m_msize = mbstat[0].m_msize;
308 mbstat_total.m_mclbytes = mbstat[0].m_mclbytes;
309 mbstat_total.m_minclsize = mbstat[0].m_minclsize;
310 mbstat_total.m_mlen = mbstat[0].m_mlen;
311 mbstat_total.m_mhlen = mbstat[0].m_mhlen;
313 return(sysctl_handle_opaque(oidp, mbstat_totalp, sizeof(mbstat_total), req));
317 do_mbtypes(SYSCTL_HANDLER_ARGS)
319 u_long totals[MT_NTYPES];
322 for (i = 0; i < MT_NTYPES; i++)
325 for (i = 0; i < ncpus; i++)
327 for (j = 0; j < MT_NTYPES; j++)
328 totals[j] += mbtypes[i][j];
331 return(sysctl_handle_opaque(oidp, totals, sizeof(totals), req));
335 * These are read-only because we do not currently have any code
336 * to adjust the objcache limits after the fact. The variables
337 * may only be set as boot-time tunables.
339 SYSCTL_INT(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLFLAG_RD,
340 &nmbclusters, 0, "Maximum number of mbuf clusters available");
341 SYSCTL_INT(_kern_ipc, OID_AUTO, nmbufs, CTLFLAG_RD, &nmbufs, 0,
342 "Maximum number of mbufs available");
344 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
345 &m_defragpackets, 0, "");
346 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
347 &m_defragbytes, 0, "");
348 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
349 &m_defraguseless, 0, "");
350 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
351 &m_defragfailure, 0, "");
352 #ifdef MBUF_STRESS_TEST
353 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
354 &m_defragrandomfailures, 0, "");
357 static MALLOC_DEFINE(M_MBUF, "mbuf", "mbuf");
358 static MALLOC_DEFINE(M_MBUFCL, "mbufcl", "mbufcl");
359 static MALLOC_DEFINE(M_MCLMETA, "mclmeta", "mclmeta");
361 static void m_reclaim (void);
362 static void m_mclref(void *arg);
363 static void m_mclfree(void *arg);
366 #define NMBCLUSTERS (512 + maxusers * 16)
369 #define NMBUFS (nmbclusters * 2)
373 * Perform sanity checks of tunables declared above.
376 tunable_mbinit(void *dummy)
379 * This has to be done before VM init.
381 nmbclusters = NMBCLUSTERS;
382 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
384 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs);
386 if (nmbufs < nmbclusters * 2)
387 nmbufs = nmbclusters * 2;
389 SYSINIT(tunable_mbinit, SI_BOOT1_TUNABLES, SI_ORDER_ANY,
390 tunable_mbinit, NULL);
392 /* "number of clusters of pages" */
398 * The mbuf object cache only guarantees that m_next and m_nextpkt are
399 * NULL and that m_data points to the beginning of the data area. In
400 * particular, m_len and m_pkthdr.len are uninitialized. It is the
401 * responsibility of the caller to initialize those fields before use.
404 static boolean_t __inline
405 mbuf_ctor(void *obj, void *private, int ocflags)
407 struct mbuf *m = obj;
411 m->m_data = m->m_dat;
418 * Initialize the mbuf and the packet header fields.
421 mbufphdr_ctor(void *obj, void *private, int ocflags)
423 struct mbuf *m = obj;
427 m->m_data = m->m_pktdat;
428 m->m_flags = M_PKTHDR | M_PHCACHE;
430 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */
431 SLIST_INIT(&m->m_pkthdr.tags);
432 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */
433 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */
439 * A mbcluster object consists of 2K (MCLBYTES) cluster and a refcount.
442 mclmeta_ctor(void *obj, void *private, int ocflags)
444 struct mbcluster *cl = obj;
447 if (ocflags & M_NOWAIT)
448 buf = kmalloc(MCLBYTES, M_MBUFCL, M_NOWAIT | M_ZERO);
450 buf = kmalloc(MCLBYTES, M_MBUFCL, M_INTWAIT | M_ZERO);
459 mclmeta_dtor(void *obj, void *private)
461 struct mbcluster *mcl = obj;
463 KKASSERT(mcl->mcl_refs == 0);
464 kfree(mcl->mcl_data, M_MBUFCL);
468 linkcluster(struct mbuf *m, struct mbcluster *cl)
471 * Add the cluster to the mbuf. The caller will detect that the
472 * mbuf now has an attached cluster.
474 m->m_ext.ext_arg = cl;
475 m->m_ext.ext_buf = cl->mcl_data;
476 m->m_ext.ext_ref = m_mclref;
477 m->m_ext.ext_free = m_mclfree;
478 m->m_ext.ext_size = MCLBYTES;
479 atomic_add_int(&cl->mcl_refs, 1);
481 m->m_data = m->m_ext.ext_buf;
482 m->m_flags |= M_EXT | M_EXT_CLUSTER;
486 mbufphdrcluster_ctor(void *obj, void *private, int ocflags)
488 struct mbuf *m = obj;
489 struct mbcluster *cl;
491 mbufphdr_ctor(obj, private, ocflags);
492 cl = objcache_get(mclmeta_cache, ocflags);
494 ++mbstat[mycpu->gd_cpuid].m_drops;
497 m->m_flags |= M_CLCACHE;
503 mbufcluster_ctor(void *obj, void *private, int ocflags)
505 struct mbuf *m = obj;
506 struct mbcluster *cl;
508 mbuf_ctor(obj, private, ocflags);
509 cl = objcache_get(mclmeta_cache, ocflags);
511 ++mbstat[mycpu->gd_cpuid].m_drops;
514 m->m_flags |= M_CLCACHE;
520 * Used for both the cluster and cluster PHDR caches.
522 * The mbuf may have lost its cluster due to sharing, deal
523 * with the situation by checking M_EXT.
526 mbufcluster_dtor(void *obj, void *private)
528 struct mbuf *m = obj;
529 struct mbcluster *mcl;
531 if (m->m_flags & M_EXT) {
532 KKASSERT((m->m_flags & M_EXT_CLUSTER) != 0);
533 mcl = m->m_ext.ext_arg;
534 KKASSERT(mcl->mcl_refs == 1);
536 objcache_put(mclmeta_cache, mcl);
540 struct objcache_malloc_args mbuf_malloc_args = { MSIZE, M_MBUF };
541 struct objcache_malloc_args mclmeta_malloc_args =
542 { sizeof(struct mbcluster), M_MCLMETA };
548 int mb_limit, cl_limit;
553 * Initialize statistics
555 for (i = 0; i < ncpus; i++) {
556 atomic_set_long_nonlocked(&mbstat[i].m_msize, MSIZE);
557 atomic_set_long_nonlocked(&mbstat[i].m_mclbytes, MCLBYTES);
558 atomic_set_long_nonlocked(&mbstat[i].m_minclsize, MINCLSIZE);
559 atomic_set_long_nonlocked(&mbstat[i].m_mlen, MLEN);
560 atomic_set_long_nonlocked(&mbstat[i].m_mhlen, MHLEN);
564 * Create objtect caches and save cluster limits, which will
565 * be used to adjust backing kmalloc pools' limit later.
568 mb_limit = cl_limit = 0;
571 mbuf_cache = objcache_create("mbuf", &limit, 0,
572 mbuf_ctor, NULL, NULL,
573 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
577 mbufphdr_cache = objcache_create("mbuf pkt hdr", &limit, 64,
578 mbufphdr_ctor, NULL, NULL,
579 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
582 cl_limit = nmbclusters;
583 mclmeta_cache = objcache_create("cluster mbuf", &cl_limit, 0,
584 mclmeta_ctor, mclmeta_dtor, NULL,
585 objcache_malloc_alloc, objcache_malloc_free, &mclmeta_malloc_args);
588 mbufcluster_cache = objcache_create("mbuf + cluster", &limit, 0,
589 mbufcluster_ctor, mbufcluster_dtor, NULL,
590 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
594 mbufphdrcluster_cache = objcache_create("mbuf pkt hdr + cluster",
595 &limit, 64, mbufphdrcluster_ctor, mbufcluster_dtor, NULL,
596 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
600 * Adjust backing kmalloc pools' limit
602 * NOTE: We raise the limit by another 1/8 to take the effect
603 * of loosememuse into account.
605 cl_limit += cl_limit / 8;
606 kmalloc_raise_limit(mclmeta_malloc_args.mtype,
607 mclmeta_malloc_args.objsize * cl_limit);
608 kmalloc_raise_limit(M_MBUFCL, MCLBYTES * cl_limit);
610 mb_limit += mb_limit / 8;
611 kmalloc_raise_limit(mbuf_malloc_args.mtype,
612 mbuf_malloc_args.objsize * mb_limit);
616 * Return the number of references to this mbuf's data. 0 is returned
617 * if the mbuf is not M_EXT, a reference count is returned if it is
618 * M_EXT | M_EXT_CLUSTER, and 99 is returned if it is a special M_EXT.
621 m_sharecount(struct mbuf *m)
623 switch (m->m_flags & (M_EXT | M_EXT_CLUSTER)) {
628 case M_EXT | M_EXT_CLUSTER:
629 return (((struct mbcluster *)m->m_ext.ext_arg)->mcl_refs);
632 return (0); /* to shut up compiler */
636 * change mbuf to new type
639 m_chtype(struct mbuf *m, int type)
641 struct globaldata *gd = mycpu;
643 atomic_add_long_nonlocked(&mbtypes[gd->gd_cpuid][type], 1);
644 atomic_subtract_long_nonlocked(&mbtypes[gd->gd_cpuid][m->m_type], 1);
645 atomic_set_short_nonlocked(&m->m_type, type);
654 kprintf("Debug: m_reclaim() called\n");
656 SLIST_FOREACH(dp, &domains, dom_next) {
657 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
662 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_drain, 1);
666 updatestats(struct mbuf *m, int type)
668 struct globaldata *gd = mycpu;
672 KKASSERT(m->m_next == NULL);
673 KKASSERT(m->m_nextpkt == NULL);
675 atomic_add_long_nonlocked(&mbtypes[gd->gd_cpuid][type], 1);
676 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mbufs, 1);
684 m_get(int how, int type)
688 int ocf = MBTOM(how);
692 m = objcache_get(mbuf_cache, ocf);
695 if ((how & MB_TRYWAIT) && ntries++ == 0) {
696 struct objcache *reclaimlist[] = {
699 mbufphdrcluster_cache
701 const int nreclaims = __arysize(reclaimlist);
703 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf))
707 ++mbstat[mycpu->gd_cpuid].m_drops;
711 updatestats(m, type);
716 m_gethdr(int how, int type)
719 int ocf = MBTOM(how);
724 m = objcache_get(mbufphdr_cache, ocf);
727 if ((how & MB_TRYWAIT) && ntries++ == 0) {
728 struct objcache *reclaimlist[] = {
730 mbufcluster_cache, mbufphdrcluster_cache
732 const int nreclaims = __arysize(reclaimlist);
734 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf))
738 ++mbstat[mycpu->gd_cpuid].m_drops;
742 updatestats(m, type);
747 * Get a mbuf (not a mbuf cluster!) and zero it.
751 m_getclr(int how, int type)
755 m = m_get(how, type);
757 bzero(m->m_data, MLEN);
762 * Returns an mbuf with an attached cluster.
763 * Because many network drivers use this kind of buffers a lot, it is
764 * convenient to keep a small pool of free buffers of this kind.
765 * Even a small size such as 10 gives about 10% improvement in the
766 * forwarding rate in a bridge or router.
769 m_getcl(int how, short type, int flags)
772 int ocflags = MBTOM(how);
777 if (flags & M_PKTHDR)
778 m = objcache_get(mbufphdrcluster_cache, ocflags);
780 m = objcache_get(mbufcluster_cache, ocflags);
783 if ((how & MB_TRYWAIT) && ntries++ == 0) {
784 struct objcache *reclaimlist[1];
786 if (flags & M_PKTHDR)
787 reclaimlist[0] = mbufcluster_cache;
789 reclaimlist[0] = mbufphdrcluster_cache;
790 if (!objcache_reclaimlist(reclaimlist, 1, ocflags))
794 ++mbstat[mycpu->gd_cpuid].m_drops;
802 atomic_add_long_nonlocked(&mbtypes[mycpu->gd_cpuid][type], 1);
803 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters, 1);
808 * Allocate chain of requested length.
811 m_getc(int len, int how, int type)
813 struct mbuf *n, *nfirst = NULL, **ntail = &nfirst;
817 n = m_getl(len, how, type, 0, &nsize);
833 * Allocate len-worth of mbufs and/or mbuf clusters (whatever fits best)
834 * and return a pointer to the head of the allocated chain. If m0 is
835 * non-null, then we assume that it is a single mbuf or an mbuf chain to
836 * which we want len bytes worth of mbufs and/or clusters attached, and so
837 * if we succeed in allocating it, we will just return a pointer to m0.
839 * If we happen to fail at any point during the allocation, we will free
840 * up everything we have already allocated and return NULL.
842 * Deprecated. Use m_getc() and m_cat() instead.
845 m_getm(struct mbuf *m0, int len, int type, int how)
849 nfirst = m_getc(len, how, type);
852 m_last(m0)->m_next = nfirst;
860 * Adds a cluster to a normal mbuf, M_EXT is set on success.
861 * Deprecated. Use m_getcl() instead.
864 m_mclget(struct mbuf *m, int how)
866 struct mbcluster *mcl;
868 KKASSERT((m->m_flags & M_EXT) == 0);
869 mcl = objcache_get(mclmeta_cache, MBTOM(how));
872 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters,
875 ++mbstat[mycpu->gd_cpuid].m_drops;
880 * Updates to mbcluster must be MPSAFE. Only an entity which already has
881 * a reference to the cluster can ref it, so we are in no danger of
882 * racing an add with a subtract. But the operation must still be atomic
883 * since multiple entities may have a reference on the cluster.
885 * m_mclfree() is almost the same but it must contend with two entities
886 * freeing the cluster at the same time.
891 struct mbcluster *mcl = arg;
893 atomic_add_int(&mcl->mcl_refs, 1);
897 * When dereferencing a cluster we have to deal with a N->0 race, where
898 * N entities free their references simultaniously. To do this we use
899 * atomic_fetchadd_int().
904 struct mbcluster *mcl = arg;
906 if (atomic_fetchadd_int(&mcl->mcl_refs, -1) == 1)
907 objcache_put(mclmeta_cache, mcl);
911 * Free a single mbuf and any associated external storage. The successor,
912 * if any, is returned.
914 * We do need to check non-first mbuf for m_aux, since some of existing
915 * code does not call M_PREPEND properly.
916 * (example: call to bpf_mtap from drivers)
919 m_free(struct mbuf *m)
922 struct globaldata *gd = mycpu;
924 KASSERT(m->m_type != MT_FREE, ("freeing free mbuf %p", m));
925 atomic_subtract_long_nonlocked(&mbtypes[gd->gd_cpuid][m->m_type], 1);
930 * Make sure the mbuf is in constructed state before returning it
936 KKASSERT(m->m_nextpkt == NULL);
938 if (m->m_nextpkt != NULL) {
939 static int afewtimes = 10;
941 if (afewtimes-- > 0) {
942 kprintf("mfree: m->m_nextpkt != NULL\n");
948 if (m->m_flags & M_PKTHDR) {
949 m_tag_delete_chain(m); /* eliminate XXX JH */
952 m->m_flags &= (M_EXT | M_EXT_CLUSTER | M_CLCACHE | M_PHCACHE);
955 * Clean the M_PKTHDR state so we can return the mbuf to its original
956 * cache. This is based on the PHCACHE flag which tells us whether
957 * the mbuf was originally allocated out of a packet-header cache
958 * or a non-packet-header cache.
960 if (m->m_flags & M_PHCACHE) {
961 m->m_flags |= M_PKTHDR;
962 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */
963 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */
964 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */
965 SLIST_INIT(&m->m_pkthdr.tags);
969 * Handle remaining flags combinations. M_CLCACHE tells us whether
970 * the mbuf was originally allocated from a cluster cache or not,
971 * and is totally separate from whether the mbuf is currently
972 * associated with a cluster.
974 switch(m->m_flags & (M_CLCACHE | M_EXT | M_EXT_CLUSTER)) {
975 case M_CLCACHE | M_EXT | M_EXT_CLUSTER:
977 * mbuf+cluster cache case. The mbuf was allocated from the
978 * combined mbuf_cluster cache and can be returned to the
979 * cache if the cluster hasn't been shared.
981 if (m_sharecount(m) == 1) {
983 * The cluster has not been shared, we can just
984 * reset the data pointer and return the mbuf
985 * to the cluster cache. Note that the reference
986 * count is left intact (it is still associated with
989 m->m_data = m->m_ext.ext_buf;
990 if (m->m_flags & M_PHCACHE)
991 objcache_put(mbufphdrcluster_cache, m);
993 objcache_put(mbufcluster_cache, m);
994 atomic_subtract_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters, 1);
997 * Hell. Someone else has a ref on this cluster,
998 * we have to disconnect it which means we can't
999 * put it back into the mbufcluster_cache, we
1000 * have to destroy the mbuf.
1002 * Other mbuf references to the cluster will typically
1003 * be M_EXT | M_EXT_CLUSTER but without M_CLCACHE.
1005 * XXX we could try to connect another cluster to
1008 m->m_ext.ext_free(m->m_ext.ext_arg);
1009 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER);
1010 if (m->m_flags & M_PHCACHE)
1011 objcache_dtor(mbufphdrcluster_cache, m);
1013 objcache_dtor(mbufcluster_cache, m);
1016 case M_EXT | M_EXT_CLUSTER:
1018 * Normal cluster associated with an mbuf that was allocated
1019 * from the normal mbuf pool rather then the cluster pool.
1020 * The cluster has to be independantly disassociated from the
1023 if (m_sharecount(m) == 1)
1024 atomic_subtract_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters, 1);
1028 * Normal cluster association case, disconnect the cluster from
1029 * the mbuf. The cluster may or may not be custom.
1031 m->m_ext.ext_free(m->m_ext.ext_arg);
1032 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER);
1036 * return the mbuf to the mbuf cache.
1038 if (m->m_flags & M_PHCACHE) {
1039 m->m_data = m->m_pktdat;
1040 objcache_put(mbufphdr_cache, m);
1042 m->m_data = m->m_dat;
1043 objcache_put(mbuf_cache, m);
1045 atomic_subtract_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mbufs, 1);
1049 panic("bad mbuf flags %p %08x\n", m, m->m_flags);
1056 m_freem(struct mbuf *m)
1063 * mbuf utility routines
1067 * Lesser-used path for M_PREPEND: allocate new mbuf to prepend to chain and
1071 m_prepend(struct mbuf *m, int len, int how)
1075 if (m->m_flags & M_PKTHDR)
1076 mn = m_gethdr(how, m->m_type);
1078 mn = m_get(how, m->m_type);
1083 if (m->m_flags & M_PKTHDR)
1084 M_MOVE_PKTHDR(mn, m);
1094 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
1095 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
1096 * The wait parameter is a choice of MB_WAIT/MB_DONTWAIT from caller.
1097 * Note that the copy is read-only, because clusters are not copied,
1098 * only their reference counts are incremented.
1101 m_copym(const struct mbuf *m, int off0, int len, int wait)
1103 struct mbuf *n, **np;
1108 KASSERT(off >= 0, ("m_copym, negative off %d", off));
1109 KASSERT(len >= 0, ("m_copym, negative len %d", len));
1110 if (off == 0 && (m->m_flags & M_PKTHDR))
1113 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
1123 KASSERT(len == M_COPYALL,
1124 ("m_copym, length > size of mbuf chain"));
1128 * Because we are sharing any cluster attachment below,
1129 * be sure to get an mbuf that does not have a cluster
1130 * associated with it.
1133 n = m_gethdr(wait, m->m_type);
1135 n = m_get(wait, m->m_type);
1140 if (!m_dup_pkthdr(n, m, wait))
1142 if (len == M_COPYALL)
1143 n->m_pkthdr.len -= off0;
1145 n->m_pkthdr.len = len;
1148 n->m_len = min(len, m->m_len - off);
1149 if (m->m_flags & M_EXT) {
1150 KKASSERT((n->m_flags & M_EXT) == 0);
1151 n->m_data = m->m_data + off;
1152 m->m_ext.ext_ref(m->m_ext.ext_arg);
1153 n->m_ext = m->m_ext;
1154 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1156 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
1157 (unsigned)n->m_len);
1159 if (len != M_COPYALL)
1166 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1170 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1175 * Copy an entire packet, including header (which must be present).
1176 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
1177 * Note that the copy is read-only, because clusters are not copied,
1178 * only their reference counts are incremented.
1179 * Preserve alignment of the first mbuf so if the creator has left
1180 * some room at the beginning (e.g. for inserting protocol headers)
1181 * the copies also have the room available.
1184 m_copypacket(struct mbuf *m, int how)
1186 struct mbuf *top, *n, *o;
1188 n = m_gethdr(how, m->m_type);
1193 if (!m_dup_pkthdr(n, m, how))
1195 n->m_len = m->m_len;
1196 if (m->m_flags & M_EXT) {
1197 KKASSERT((n->m_flags & M_EXT) == 0);
1198 n->m_data = m->m_data;
1199 m->m_ext.ext_ref(m->m_ext.ext_arg);
1200 n->m_ext = m->m_ext;
1201 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1203 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
1204 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
1209 o = m_get(how, m->m_type);
1216 n->m_len = m->m_len;
1217 if (m->m_flags & M_EXT) {
1218 KKASSERT((n->m_flags & M_EXT) == 0);
1219 n->m_data = m->m_data;
1220 m->m_ext.ext_ref(m->m_ext.ext_arg);
1221 n->m_ext = m->m_ext;
1222 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1224 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
1232 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1237 * Copy data from an mbuf chain starting "off" bytes from the beginning,
1238 * continuing for "len" bytes, into the indicated buffer.
1241 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
1245 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
1246 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
1248 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
1255 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
1256 count = min(m->m_len - off, len);
1257 bcopy(mtod(m, caddr_t) + off, cp, count);
1266 * Copy a packet header mbuf chain into a completely new chain, including
1267 * copying any mbuf clusters. Use this instead of m_copypacket() when
1268 * you need a writable copy of an mbuf chain.
1271 m_dup(struct mbuf *m, int how)
1273 struct mbuf **p, *top = NULL;
1274 int remain, moff, nsize;
1279 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __func__));
1281 /* While there's more data, get a new mbuf, tack it on, and fill it */
1282 remain = m->m_pkthdr.len;
1285 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
1288 /* Get the next new mbuf */
1289 n = m_getl(remain, how, m->m_type, top == NULL ? M_PKTHDR : 0,
1294 if (!m_dup_pkthdr(n, m, how))
1297 /* Link it into the new chain */
1301 /* Copy data from original mbuf(s) into new mbuf */
1303 while (n->m_len < nsize && m != NULL) {
1304 int chunk = min(nsize - n->m_len, m->m_len - moff);
1306 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
1310 if (moff == m->m_len) {
1316 /* Check correct total mbuf length */
1317 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
1318 ("%s: bogus m_pkthdr.len", __func__));
1325 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1330 * Copy the non-packet mbuf data chain into a new set of mbufs, including
1331 * copying any mbuf clusters. This is typically used to realign a data
1332 * chain by nfs_realign().
1334 * The original chain is left intact. how should be MB_WAIT or MB_DONTWAIT
1335 * and NULL can be returned if MB_DONTWAIT is passed.
1337 * Be careful to use cluster mbufs, a large mbuf chain converted to non
1338 * cluster mbufs can exhaust our supply of mbufs.
1341 m_dup_data(struct mbuf *m, int how)
1343 struct mbuf **p, *n, *top = NULL;
1344 int mlen, moff, chunk, gsize, nsize;
1353 * Optimize the mbuf allocation but do not get too carried away.
1355 if (m->m_next || m->m_len > MLEN)
1366 * Scan the mbuf chain until nothing is left, the new mbuf chain
1367 * will be allocated on the fly as needed.
1374 KKASSERT(m->m_type == MT_DATA);
1376 n = m_getl(gsize, how, MT_DATA, 0, &nsize);
1383 chunk = imin(mlen, nsize);
1384 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
1399 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1404 * Concatenate mbuf chain n to m.
1405 * Both chains must be of the same type (e.g. MT_DATA).
1406 * Any m_pkthdr is not updated.
1409 m_cat(struct mbuf *m, struct mbuf *n)
1413 if (m->m_flags & M_EXT ||
1414 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
1415 /* just join the two chains */
1419 /* splat the data from one into the other */
1420 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1422 m->m_len += n->m_len;
1428 m_adj(struct mbuf *mp, int req_len)
1434 if ((m = mp) == NULL)
1440 while (m != NULL && len > 0) {
1441 if (m->m_len <= len) {
1452 if (mp->m_flags & M_PKTHDR)
1453 m->m_pkthdr.len -= (req_len - len);
1456 * Trim from tail. Scan the mbuf chain,
1457 * calculating its length and finding the last mbuf.
1458 * If the adjustment only affects this mbuf, then just
1459 * adjust and return. Otherwise, rescan and truncate
1460 * after the remaining size.
1466 if (m->m_next == NULL)
1470 if (m->m_len >= len) {
1472 if (mp->m_flags & M_PKTHDR)
1473 mp->m_pkthdr.len -= len;
1480 * Correct length for chain is "count".
1481 * Find the mbuf with last data, adjust its length,
1482 * and toss data from remaining mbufs on chain.
1485 if (m->m_flags & M_PKTHDR)
1486 m->m_pkthdr.len = count;
1487 for (; m; m = m->m_next) {
1488 if (m->m_len >= count) {
1495 (m = m->m_next) ->m_len = 0;
1500 * Set the m_data pointer of a newly-allocated mbuf
1501 * to place an object of the specified size at the
1502 * end of the mbuf, longword aligned.
1505 m_align(struct mbuf *m, int len)
1509 if (m->m_flags & M_EXT)
1510 adjust = m->m_ext.ext_size - len;
1511 else if (m->m_flags & M_PKTHDR)
1512 adjust = MHLEN - len;
1514 adjust = MLEN - len;
1515 m->m_data += adjust &~ (sizeof(long)-1);
1519 * Rearrange an mbuf chain so that len bytes are contiguous
1520 * and in the data area of an mbuf (so that mtod will work for a structure
1521 * of size len). Returns the resulting mbuf chain on success, frees it and
1522 * returns null on failure. If there is room, it will add up to
1523 * max_protohdr-len extra bytes to the contiguous region in an attempt to
1524 * avoid being called next time.
1527 m_pullup(struct mbuf *n, int len)
1534 * If first mbuf has no cluster, and has room for len bytes
1535 * without shifting current data, pullup into it,
1536 * otherwise allocate a new mbuf to prepend to the chain.
1538 if (!(n->m_flags & M_EXT) &&
1539 n->m_data + len < &n->m_dat[MLEN] &&
1541 if (n->m_len >= len)
1549 if (n->m_flags & M_PKTHDR)
1550 m = m_gethdr(MB_DONTWAIT, n->m_type);
1552 m = m_get(MB_DONTWAIT, n->m_type);
1556 if (n->m_flags & M_PKTHDR)
1557 M_MOVE_PKTHDR(m, n);
1559 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1561 count = min(min(max(len, max_protohdr), space), n->m_len);
1562 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1572 } while (len > 0 && n);
1581 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1586 * Partition an mbuf chain in two pieces, returning the tail --
1587 * all but the first len0 bytes. In case of failure, it returns NULL and
1588 * attempts to restore the chain to its original state.
1590 * Note that the resulting mbufs might be read-only, because the new
1591 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1592 * the "breaking point" happens to lie within a cluster mbuf. Use the
1593 * M_WRITABLE() macro to check for this case.
1596 m_split(struct mbuf *m0, int len0, int wait)
1599 unsigned len = len0, remain;
1601 for (m = m0; m && len > m->m_len; m = m->m_next)
1605 remain = m->m_len - len;
1606 if (m0->m_flags & M_PKTHDR) {
1607 n = m_gethdr(wait, m0->m_type);
1610 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1611 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1612 m0->m_pkthdr.len = len0;
1613 if (m->m_flags & M_EXT)
1615 if (remain > MHLEN) {
1616 /* m can't be the lead packet */
1618 n->m_next = m_split(m, len, wait);
1619 if (n->m_next == NULL) {
1627 MH_ALIGN(n, remain);
1628 } else if (remain == 0) {
1633 n = m_get(wait, m->m_type);
1639 if (m->m_flags & M_EXT) {
1640 KKASSERT((n->m_flags & M_EXT) == 0);
1641 n->m_data = m->m_data + len;
1642 m->m_ext.ext_ref(m->m_ext.ext_arg);
1643 n->m_ext = m->m_ext;
1644 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1646 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1650 n->m_next = m->m_next;
1656 * Routine to copy from device local memory into mbufs.
1657 * Note: "offset" is ill-defined and always called as 0, so ignore it.
1660 m_devget(char *buf, int len, int offset, struct ifnet *ifp,
1661 void (*copy)(volatile const void *from, volatile void *to, size_t length))
1663 struct mbuf *m, *mfirst = NULL, **mtail;
1672 m = m_getl(len, MB_DONTWAIT, MT_DATA, flags, &nsize);
1677 m->m_len = min(len, nsize);
1679 if (flags & M_PKTHDR) {
1680 if (len + max_linkhdr <= nsize)
1681 m->m_data += max_linkhdr;
1682 m->m_pkthdr.rcvif = ifp;
1683 m->m_pkthdr.len = len;
1687 copy(buf, m->m_data, (unsigned)m->m_len);
1698 * Routine to pad mbuf to the specified length 'padto'.
1701 m_devpad(struct mbuf *m, int padto)
1703 struct mbuf *last = NULL;
1706 if (padto <= m->m_pkthdr.len)
1709 padlen = padto - m->m_pkthdr.len;
1711 /* if there's only the packet-header and we can pad there, use it. */
1712 if (m->m_pkthdr.len == m->m_len && M_TRAILINGSPACE(m) >= padlen) {
1716 * Walk packet chain to find last mbuf. We will either
1717 * pad there, or append a new mbuf and pad it
1719 for (last = m; last->m_next != NULL; last = last->m_next)
1722 /* `last' now points to last in chain. */
1723 if (M_TRAILINGSPACE(last) < padlen) {
1726 /* Allocate new empty mbuf, pad it. Compact later. */
1727 MGET(n, MB_DONTWAIT, MT_DATA);
1735 KKASSERT(M_TRAILINGSPACE(last) >= padlen);
1736 KKASSERT(M_WRITABLE(last));
1738 /* Now zero the pad area */
1739 bzero(mtod(last, char *) + last->m_len, padlen);
1740 last->m_len += padlen;
1741 m->m_pkthdr.len += padlen;
1746 * Copy data from a buffer back into the indicated mbuf chain,
1747 * starting "off" bytes from the beginning, extending the mbuf
1748 * chain if necessary.
1751 m_copyback(struct mbuf *m0, int off, int len, caddr_t cp)
1754 struct mbuf *m = m0, *n;
1759 while (off > (mlen = m->m_len)) {
1762 if (m->m_next == NULL) {
1763 n = m_getclr(MB_DONTWAIT, m->m_type);
1766 n->m_len = min(MLEN, len + off);
1772 mlen = min (m->m_len - off, len);
1773 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen);
1781 if (m->m_next == NULL) {
1782 n = m_get(MB_DONTWAIT, m->m_type);
1785 n->m_len = min(MLEN, len);
1790 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1791 m->m_pkthdr.len = totlen;
1795 * Append the specified data to the indicated mbuf chain,
1796 * Extend the mbuf chain if the new data does not fit in
1799 * Return 1 if able to complete the job; otherwise 0.
1802 m_append(struct mbuf *m0, int len, c_caddr_t cp)
1805 int remainder, space;
1807 for (m = m0; m->m_next != NULL; m = m->m_next)
1810 space = M_TRAILINGSPACE(m);
1813 * Copy into available space.
1815 if (space > remainder)
1817 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1819 cp += space, remainder -= space;
1821 while (remainder > 0) {
1823 * Allocate a new mbuf; could check space
1824 * and allocate a cluster instead.
1826 n = m_get(MB_DONTWAIT, m->m_type);
1829 n->m_len = min(MLEN, remainder);
1830 bcopy(cp, mtod(n, caddr_t), n->m_len);
1831 cp += n->m_len, remainder -= n->m_len;
1835 if (m0->m_flags & M_PKTHDR)
1836 m0->m_pkthdr.len += len - remainder;
1837 return (remainder == 0);
1841 * Apply function f to the data in an mbuf chain starting "off" bytes from
1842 * the beginning, continuing for "len" bytes.
1845 m_apply(struct mbuf *m, int off, int len,
1846 int (*f)(void *, void *, u_int), void *arg)
1851 KASSERT(off >= 0, ("m_apply, negative off %d", off));
1852 KASSERT(len >= 0, ("m_apply, negative len %d", len));
1854 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1861 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1862 count = min(m->m_len - off, len);
1863 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1874 * Return a pointer to mbuf/offset of location in mbuf chain.
1877 m_getptr(struct mbuf *m, int loc, int *off)
1881 /* Normal end of search. */
1882 if (m->m_len > loc) {
1887 if (m->m_next == NULL) {
1889 /* Point at the end of valid data. */
1902 m_print(const struct mbuf *m)
1905 const struct mbuf *m2;
1907 len = m->m_pkthdr.len;
1910 kprintf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-");
1918 * "Move" mbuf pkthdr from "from" to "to".
1919 * "from" must have M_PKTHDR set, and "to" must be empty.
1922 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
1924 KASSERT((to->m_flags & M_PKTHDR), ("m_move_pkthdr: not packet header"));
1926 to->m_flags |= from->m_flags & M_COPYFLAGS;
1927 to->m_pkthdr = from->m_pkthdr; /* especially tags */
1928 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
1932 * Duplicate "from"'s mbuf pkthdr in "to".
1933 * "from" must have M_PKTHDR set, and "to" must be empty.
1934 * In particular, this does a deep copy of the packet tags.
1937 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how)
1939 KASSERT((to->m_flags & M_PKTHDR), ("m_dup_pkthdr: not packet header"));
1941 to->m_flags = (from->m_flags & M_COPYFLAGS) |
1942 (to->m_flags & ~M_COPYFLAGS);
1943 to->m_pkthdr = from->m_pkthdr;
1944 SLIST_INIT(&to->m_pkthdr.tags);
1945 return (m_tag_copy_chain(to, from, how));
1949 * Defragment a mbuf chain, returning the shortest possible
1950 * chain of mbufs and clusters. If allocation fails and
1951 * this cannot be completed, NULL will be returned, but
1952 * the passed in chain will be unchanged. Upon success,
1953 * the original chain will be freed, and the new chain
1956 * If a non-packet header is passed in, the original
1957 * mbuf (chain?) will be returned unharmed.
1959 * m_defrag_nofree doesn't free the passed in mbuf.
1962 m_defrag(struct mbuf *m0, int how)
1966 if ((m_new = m_defrag_nofree(m0, how)) == NULL)
1974 m_defrag_nofree(struct mbuf *m0, int how)
1976 struct mbuf *m_new = NULL, *m_final = NULL;
1977 int progress = 0, length, nsize;
1979 if (!(m0->m_flags & M_PKTHDR))
1982 #ifdef MBUF_STRESS_TEST
1983 if (m_defragrandomfailures) {
1984 int temp = karc4random() & 0xff;
1990 m_final = m_getl(m0->m_pkthdr.len, how, MT_DATA, M_PKTHDR, &nsize);
1991 if (m_final == NULL)
1993 m_final->m_len = 0; /* in case m0->m_pkthdr.len is zero */
1995 if (m_dup_pkthdr(m_final, m0, how) == 0)
2000 while (progress < m0->m_pkthdr.len) {
2001 length = m0->m_pkthdr.len - progress;
2002 if (length > MCLBYTES)
2005 if (m_new == NULL) {
2006 m_new = m_getl(length, how, MT_DATA, 0, &nsize);
2011 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
2013 m_new->m_len = length;
2014 if (m_new != m_final)
2015 m_cat(m_final, m_new);
2018 if (m0->m_next == NULL)
2021 m_defragbytes += m_final->m_pkthdr.len;
2032 * Move data from uio into mbufs.
2035 m_uiomove(struct uio *uio)
2037 struct mbuf *m; /* current working mbuf */
2038 struct mbuf *head = NULL; /* result mbuf chain */
2039 struct mbuf **mp = &head;
2040 int flags = M_PKTHDR;
2046 if (uio->uio_resid > INT_MAX)
2049 resid = (int)uio->uio_resid;
2050 m = m_getl(resid, MB_WAIT, MT_DATA, flags, &nsize);
2052 m->m_pkthdr.len = 0;
2053 /* Leave room for protocol headers. */
2058 m->m_len = imin(nsize, resid);
2059 error = uiomove(mtod(m, caddr_t), m->m_len, uio);
2066 head->m_pkthdr.len += m->m_len;
2067 } while (uio->uio_resid > 0);
2077 m_last(struct mbuf *m)
2085 * Return the number of bytes in an mbuf chain.
2086 * If lastm is not NULL, also return the last mbuf.
2089 m_lengthm(struct mbuf *m, struct mbuf **lastm)
2092 struct mbuf *prev = m;
2105 * Like m_lengthm(), except also keep track of mbuf usage.
2108 m_countm(struct mbuf *m, struct mbuf **lastm, u_int *pmbcnt)
2110 u_int len = 0, mbcnt = 0;
2111 struct mbuf *prev = m;
2116 if (m->m_flags & M_EXT)
2117 mbcnt += m->m_ext.ext_size;