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>
78 #include <sys/malloc.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
82 #include <sys/domain.h>
83 #include <sys/objcache.h>
85 #include <sys/protosw.h>
87 #include <sys/thread.h>
88 #include <sys/globaldata.h>
90 #include <sys/thread2.h>
91 #include <sys/spinlock2.h>
93 #include <machine/atomic.h>
94 #include <machine/limits.h>
97 #include <vm/vm_kern.h>
98 #include <vm/vm_extern.h>
101 #include <machine/cpu.h>
105 * mbuf cluster meta-data
113 * mbuf tracking for debugging purposes
117 static MALLOC_DEFINE(M_MTRACK, "mtrack", "mtrack");
120 RB_HEAD(mbuf_rb_tree, mbtrack);
121 RB_PROTOTYPE2(mbuf_rb_tree, mbtrack, rb_node, mbtrack_cmp, struct mbuf *);
124 RB_ENTRY(mbtrack) rb_node;
130 mbtrack_cmp(struct mbtrack *mb1, struct mbtrack *mb2)
139 RB_GENERATE2(mbuf_rb_tree, mbtrack, rb_node, mbtrack_cmp, struct mbuf *, m);
141 struct mbuf_rb_tree mbuf_track_root;
142 static struct spinlock mbuf_track_spin = SPINLOCK_INITIALIZER(mbuf_track_spin);
145 mbuftrack(struct mbuf *m)
149 mbt = kmalloc(sizeof(*mbt), M_MTRACK, M_INTWAIT|M_ZERO);
150 spin_lock(&mbuf_track_spin);
152 if (mbuf_rb_tree_RB_INSERT(&mbuf_track_root, mbt)) {
153 spin_unlock(&mbuf_track_spin);
154 panic("mbuftrack: mbuf %p already being tracked\n", m);
156 spin_unlock(&mbuf_track_spin);
160 mbufuntrack(struct mbuf *m)
164 spin_lock(&mbuf_track_spin);
165 mbt = mbuf_rb_tree_RB_LOOKUP(&mbuf_track_root, m);
167 spin_unlock(&mbuf_track_spin);
168 panic("mbufuntrack: mbuf %p was not tracked\n", m);
170 mbuf_rb_tree_RB_REMOVE(&mbuf_track_root, mbt);
171 spin_unlock(&mbuf_track_spin);
172 kfree(mbt, M_MTRACK);
177 mbuftrackid(struct mbuf *m, int trackid)
182 spin_lock(&mbuf_track_spin);
186 mbt = mbuf_rb_tree_RB_LOOKUP(&mbuf_track_root, m);
188 spin_unlock(&mbuf_track_spin);
189 panic("mbuftrackid: mbuf %p not tracked", m);
191 mbt->trackid = trackid;
196 spin_unlock(&mbuf_track_spin);
200 mbuftrack_callback(struct mbtrack *mbt, void *arg)
202 struct sysctl_req *req = arg;
206 ksnprintf(buf, sizeof(buf), "mbuf %p track %d\n", mbt->m, mbt->trackid);
208 spin_unlock(&mbuf_track_spin);
209 error = SYSCTL_OUT(req, buf, strlen(buf));
210 spin_lock(&mbuf_track_spin);
217 mbuftrack_show(SYSCTL_HANDLER_ARGS)
221 spin_lock(&mbuf_track_spin);
222 error = mbuf_rb_tree_RB_SCAN(&mbuf_track_root, NULL,
223 mbuftrack_callback, req);
224 spin_unlock(&mbuf_track_spin);
227 SYSCTL_PROC(_kern_ipc, OID_AUTO, showmbufs, CTLFLAG_RD|CTLTYPE_STRING,
228 0, 0, mbuftrack_show, "A", "Show all in-use mbufs");
233 #define mbufuntrack(m)
237 static void mbinit(void *);
238 SYSINIT(mbuf, SI_BOOT2_MACHDEP, SI_ORDER_FIRST, mbinit, NULL)
240 static u_long mbtypes[SMP_MAXCPU][MT_NTYPES];
242 static struct mbstat mbstat[SMP_MAXCPU];
251 #ifdef MBUF_STRESS_TEST
252 int m_defragrandomfailures;
255 struct objcache *mbuf_cache, *mbufphdr_cache;
256 struct objcache *mclmeta_cache;
257 struct objcache *mbufcluster_cache, *mbufphdrcluster_cache;
262 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW,
263 &max_linkhdr, 0, "Max size of a link-level header");
264 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW,
265 &max_protohdr, 0, "Max size of a protocol header");
266 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0,
267 "Max size of link+protocol headers");
268 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW,
269 &max_datalen, 0, "Max data payload size without headers");
270 SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW,
271 &mbuf_wait, 0, "Time in ticks to sleep after failed mbuf allocations");
272 static int do_mbstat(SYSCTL_HANDLER_ARGS);
274 SYSCTL_PROC(_kern_ipc, KIPC_MBSTAT, mbstat, CTLTYPE_STRUCT|CTLFLAG_RD,
275 0, 0, do_mbstat, "S,mbstat", "mbuf usage statistics");
277 static int do_mbtypes(SYSCTL_HANDLER_ARGS);
279 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbtypes, CTLTYPE_ULONG|CTLFLAG_RD,
280 0, 0, do_mbtypes, "LU", "");
283 do_mbstat(SYSCTL_HANDLER_ARGS)
285 struct mbstat mbstat_total;
286 struct mbstat *mbstat_totalp;
289 bzero(&mbstat_total, sizeof(mbstat_total));
290 mbstat_totalp = &mbstat_total;
292 for (i = 0; i < ncpus; i++)
294 mbstat_total.m_mbufs += mbstat[i].m_mbufs;
295 mbstat_total.m_clusters += mbstat[i].m_clusters;
296 mbstat_total.m_spare += mbstat[i].m_spare;
297 mbstat_total.m_clfree += mbstat[i].m_clfree;
298 mbstat_total.m_drops += mbstat[i].m_drops;
299 mbstat_total.m_wait += mbstat[i].m_wait;
300 mbstat_total.m_drain += mbstat[i].m_drain;
301 mbstat_total.m_mcfail += mbstat[i].m_mcfail;
302 mbstat_total.m_mpfail += mbstat[i].m_mpfail;
306 * The following fields are not cumulative fields so just
307 * get their values once.
309 mbstat_total.m_msize = mbstat[0].m_msize;
310 mbstat_total.m_mclbytes = mbstat[0].m_mclbytes;
311 mbstat_total.m_minclsize = mbstat[0].m_minclsize;
312 mbstat_total.m_mlen = mbstat[0].m_mlen;
313 mbstat_total.m_mhlen = mbstat[0].m_mhlen;
315 return(sysctl_handle_opaque(oidp, mbstat_totalp, sizeof(mbstat_total), req));
319 do_mbtypes(SYSCTL_HANDLER_ARGS)
321 u_long totals[MT_NTYPES];
324 for (i = 0; i < MT_NTYPES; i++)
327 for (i = 0; i < ncpus; i++)
329 for (j = 0; j < MT_NTYPES; j++)
330 totals[j] += mbtypes[i][j];
333 return(sysctl_handle_opaque(oidp, totals, sizeof(totals), req));
337 * These are read-only because we do not currently have any code
338 * to adjust the objcache limits after the fact. The variables
339 * may only be set as boot-time tunables.
341 SYSCTL_INT(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLFLAG_RD,
342 &nmbclusters, 0, "Maximum number of mbuf clusters available");
343 SYSCTL_INT(_kern_ipc, OID_AUTO, nmbufs, CTLFLAG_RD, &nmbufs, 0,
344 "Maximum number of mbufs available");
346 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
347 &m_defragpackets, 0, "Number of defragment packets");
348 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
349 &m_defragbytes, 0, "Number of defragment bytes");
350 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
351 &m_defraguseless, 0, "Number of useless defragment mbuf chain operations");
352 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
353 &m_defragfailure, 0, "Number of failed defragment mbuf chain operations");
354 #ifdef MBUF_STRESS_TEST
355 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
356 &m_defragrandomfailures, 0, "");
359 static MALLOC_DEFINE(M_MBUF, "mbuf", "mbuf");
360 static MALLOC_DEFINE(M_MBUFCL, "mbufcl", "mbufcl");
361 static MALLOC_DEFINE(M_MCLMETA, "mclmeta", "mclmeta");
363 static void m_reclaim (void);
364 static void m_mclref(void *arg);
365 static void m_mclfree(void *arg);
368 * NOTE: Default NMBUFS must take into account a possible DOS attack
369 * using fd passing on unix domain sockets.
372 #define NMBCLUSTERS (512 + maxusers * 16)
375 #define NMBUFS (nmbclusters * 2 + maxfiles)
379 * Perform sanity checks of tunables declared above.
382 tunable_mbinit(void *dummy)
385 * This has to be done before VM init.
387 nmbclusters = NMBCLUSTERS;
388 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
390 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs);
392 if (nmbufs < nmbclusters * 2)
393 nmbufs = nmbclusters * 2;
395 SYSINIT(tunable_mbinit, SI_BOOT1_TUNABLES, SI_ORDER_ANY,
396 tunable_mbinit, NULL);
398 /* "number of clusters of pages" */
404 * The mbuf object cache only guarantees that m_next and m_nextpkt are
405 * NULL and that m_data points to the beginning of the data area. In
406 * particular, m_len and m_pkthdr.len are uninitialized. It is the
407 * responsibility of the caller to initialize those fields before use.
410 static boolean_t __inline
411 mbuf_ctor(void *obj, void *private, int ocflags)
413 struct mbuf *m = obj;
417 m->m_data = m->m_dat;
424 * Initialize the mbuf and the packet header fields.
427 mbufphdr_ctor(void *obj, void *private, int ocflags)
429 struct mbuf *m = obj;
433 m->m_data = m->m_pktdat;
434 m->m_flags = M_PKTHDR | M_PHCACHE;
436 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */
437 SLIST_INIT(&m->m_pkthdr.tags);
438 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */
439 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */
445 * A mbcluster object consists of 2K (MCLBYTES) cluster and a refcount.
448 mclmeta_ctor(void *obj, void *private, int ocflags)
450 struct mbcluster *cl = obj;
453 if (ocflags & M_NOWAIT)
454 buf = kmalloc(MCLBYTES, M_MBUFCL, M_NOWAIT | M_ZERO);
456 buf = kmalloc(MCLBYTES, M_MBUFCL, M_INTWAIT | M_ZERO);
465 mclmeta_dtor(void *obj, void *private)
467 struct mbcluster *mcl = obj;
469 KKASSERT(mcl->mcl_refs == 0);
470 kfree(mcl->mcl_data, M_MBUFCL);
474 linkcluster(struct mbuf *m, struct mbcluster *cl)
477 * Add the cluster to the mbuf. The caller will detect that the
478 * mbuf now has an attached cluster.
480 m->m_ext.ext_arg = cl;
481 m->m_ext.ext_buf = cl->mcl_data;
482 m->m_ext.ext_ref = m_mclref;
483 m->m_ext.ext_free = m_mclfree;
484 m->m_ext.ext_size = MCLBYTES;
485 atomic_add_int(&cl->mcl_refs, 1);
487 m->m_data = m->m_ext.ext_buf;
488 m->m_flags |= M_EXT | M_EXT_CLUSTER;
492 mbufphdrcluster_ctor(void *obj, void *private, int ocflags)
494 struct mbuf *m = obj;
495 struct mbcluster *cl;
497 mbufphdr_ctor(obj, private, ocflags);
498 cl = objcache_get(mclmeta_cache, ocflags);
500 ++mbstat[mycpu->gd_cpuid].m_drops;
503 m->m_flags |= M_CLCACHE;
509 mbufcluster_ctor(void *obj, void *private, int ocflags)
511 struct mbuf *m = obj;
512 struct mbcluster *cl;
514 mbuf_ctor(obj, private, ocflags);
515 cl = objcache_get(mclmeta_cache, ocflags);
517 ++mbstat[mycpu->gd_cpuid].m_drops;
520 m->m_flags |= M_CLCACHE;
526 * Used for both the cluster and cluster PHDR caches.
528 * The mbuf may have lost its cluster due to sharing, deal
529 * with the situation by checking M_EXT.
532 mbufcluster_dtor(void *obj, void *private)
534 struct mbuf *m = obj;
535 struct mbcluster *mcl;
537 if (m->m_flags & M_EXT) {
538 KKASSERT((m->m_flags & M_EXT_CLUSTER) != 0);
539 mcl = m->m_ext.ext_arg;
540 KKASSERT(mcl->mcl_refs == 1);
542 objcache_put(mclmeta_cache, mcl);
546 struct objcache_malloc_args mbuf_malloc_args = { MSIZE, M_MBUF };
547 struct objcache_malloc_args mclmeta_malloc_args =
548 { sizeof(struct mbcluster), M_MCLMETA };
554 int mb_limit, cl_limit;
559 * Initialize statistics
561 for (i = 0; i < ncpus; i++) {
562 atomic_set_long_nonlocked(&mbstat[i].m_msize, MSIZE);
563 atomic_set_long_nonlocked(&mbstat[i].m_mclbytes, MCLBYTES);
564 atomic_set_long_nonlocked(&mbstat[i].m_minclsize, MINCLSIZE);
565 atomic_set_long_nonlocked(&mbstat[i].m_mlen, MLEN);
566 atomic_set_long_nonlocked(&mbstat[i].m_mhlen, MHLEN);
570 * Create objtect caches and save cluster limits, which will
571 * be used to adjust backing kmalloc pools' limit later.
574 mb_limit = cl_limit = 0;
577 mbuf_cache = objcache_create("mbuf", &limit, 0,
578 mbuf_ctor, NULL, NULL,
579 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
583 mbufphdr_cache = objcache_create("mbuf pkt hdr", &limit, 64,
584 mbufphdr_ctor, NULL, NULL,
585 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
588 cl_limit = nmbclusters;
589 mclmeta_cache = objcache_create("cluster mbuf", &cl_limit, 0,
590 mclmeta_ctor, mclmeta_dtor, NULL,
591 objcache_malloc_alloc, objcache_malloc_free, &mclmeta_malloc_args);
594 mbufcluster_cache = objcache_create("mbuf + cluster", &limit, 0,
595 mbufcluster_ctor, mbufcluster_dtor, NULL,
596 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
600 mbufphdrcluster_cache = objcache_create("mbuf pkt hdr + cluster",
601 &limit, 64, mbufphdrcluster_ctor, mbufcluster_dtor, NULL,
602 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
606 * Adjust backing kmalloc pools' limit
608 * NOTE: We raise the limit by another 1/8 to take the effect
609 * of loosememuse into account.
611 cl_limit += cl_limit / 8;
612 kmalloc_raise_limit(mclmeta_malloc_args.mtype,
613 mclmeta_malloc_args.objsize * cl_limit);
614 kmalloc_raise_limit(M_MBUFCL, MCLBYTES * cl_limit);
616 mb_limit += mb_limit / 8;
617 kmalloc_raise_limit(mbuf_malloc_args.mtype,
618 mbuf_malloc_args.objsize * mb_limit);
622 * Return the number of references to this mbuf's data. 0 is returned
623 * if the mbuf is not M_EXT, a reference count is returned if it is
624 * M_EXT | M_EXT_CLUSTER, and 99 is returned if it is a special M_EXT.
627 m_sharecount(struct mbuf *m)
629 switch (m->m_flags & (M_EXT | M_EXT_CLUSTER)) {
634 case M_EXT | M_EXT_CLUSTER:
635 return (((struct mbcluster *)m->m_ext.ext_arg)->mcl_refs);
638 return (0); /* to shut up compiler */
642 * change mbuf to new type
645 m_chtype(struct mbuf *m, int type)
647 struct globaldata *gd = mycpu;
649 atomic_add_long_nonlocked(&mbtypes[gd->gd_cpuid][type], 1);
650 atomic_subtract_long_nonlocked(&mbtypes[gd->gd_cpuid][m->m_type], 1);
651 atomic_set_short_nonlocked(&m->m_type, type);
660 kprintf("Debug: m_reclaim() called\n");
662 SLIST_FOREACH(dp, &domains, dom_next) {
663 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
668 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_drain, 1);
672 updatestats(struct mbuf *m, int type)
674 struct globaldata *gd = mycpu;
679 KASSERT(m->m_next == NULL, ("mbuf %p: bad m_next in get", m));
680 KASSERT(m->m_nextpkt == NULL, ("mbuf %p: bad m_nextpkt in get", m));
683 atomic_add_long_nonlocked(&mbtypes[gd->gd_cpuid][type], 1);
684 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mbufs, 1);
692 m_get(int how, int type)
696 int ocf = MBTOM(how);
700 m = objcache_get(mbuf_cache, ocf);
703 if ((how & MB_TRYWAIT) && ntries++ == 0) {
704 struct objcache *reclaimlist[] = {
707 mbufphdrcluster_cache
709 const int nreclaims = NELEM(reclaimlist);
711 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf))
715 ++mbstat[mycpu->gd_cpuid].m_drops;
719 KASSERT(m->m_data == m->m_dat, ("mbuf %p: bad m_data in get", m));
723 updatestats(m, type);
728 m_gethdr(int how, int type)
731 int ocf = MBTOM(how);
736 m = objcache_get(mbufphdr_cache, ocf);
739 if ((how & MB_TRYWAIT) && ntries++ == 0) {
740 struct objcache *reclaimlist[] = {
742 mbufcluster_cache, mbufphdrcluster_cache
744 const int nreclaims = NELEM(reclaimlist);
746 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf))
750 ++mbstat[mycpu->gd_cpuid].m_drops;
754 KASSERT(m->m_data == m->m_pktdat, ("mbuf %p: bad m_data in get", m));
759 updatestats(m, type);
764 * Get a mbuf (not a mbuf cluster!) and zero it.
768 m_getclr(int how, int type)
772 m = m_get(how, type);
774 bzero(m->m_data, MLEN);
779 * Returns an mbuf with an attached cluster.
780 * Because many network drivers use this kind of buffers a lot, it is
781 * convenient to keep a small pool of free buffers of this kind.
782 * Even a small size such as 10 gives about 10% improvement in the
783 * forwarding rate in a bridge or router.
786 m_getcl(int how, short type, int flags)
789 int ocflags = MBTOM(how);
794 if (flags & M_PKTHDR)
795 m = objcache_get(mbufphdrcluster_cache, ocflags);
797 m = objcache_get(mbufcluster_cache, ocflags);
800 if ((how & MB_TRYWAIT) && ntries++ == 0) {
801 struct objcache *reclaimlist[1];
803 if (flags & M_PKTHDR)
804 reclaimlist[0] = mbufcluster_cache;
806 reclaimlist[0] = mbufphdrcluster_cache;
807 if (!objcache_reclaimlist(reclaimlist, 1, ocflags))
811 ++mbstat[mycpu->gd_cpuid].m_drops;
816 KASSERT(m->m_data == m->m_ext.ext_buf,
817 ("mbuf %p: bad m_data in get", m));
821 m->m_pkthdr.len = 0; /* just do it unconditonally */
825 atomic_add_long_nonlocked(&mbtypes[mycpu->gd_cpuid][type], 1);
826 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters, 1);
831 * Allocate chain of requested length.
834 m_getc(int len, int how, int type)
836 struct mbuf *n, *nfirst = NULL, **ntail = &nfirst;
840 n = m_getl(len, how, type, 0, &nsize);
856 * Allocate len-worth of mbufs and/or mbuf clusters (whatever fits best)
857 * and return a pointer to the head of the allocated chain. If m0 is
858 * non-null, then we assume that it is a single mbuf or an mbuf chain to
859 * which we want len bytes worth of mbufs and/or clusters attached, and so
860 * if we succeed in allocating it, we will just return a pointer to m0.
862 * If we happen to fail at any point during the allocation, we will free
863 * up everything we have already allocated and return NULL.
865 * Deprecated. Use m_getc() and m_cat() instead.
868 m_getm(struct mbuf *m0, int len, int type, int how)
872 nfirst = m_getc(len, how, type);
875 m_last(m0)->m_next = nfirst;
883 * Adds a cluster to a normal mbuf, M_EXT is set on success.
884 * Deprecated. Use m_getcl() instead.
887 m_mclget(struct mbuf *m, int how)
889 struct mbcluster *mcl;
891 KKASSERT((m->m_flags & M_EXT) == 0);
892 mcl = objcache_get(mclmeta_cache, MBTOM(how));
895 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters,
898 ++mbstat[mycpu->gd_cpuid].m_drops;
903 * Updates to mbcluster must be MPSAFE. Only an entity which already has
904 * a reference to the cluster can ref it, so we are in no danger of
905 * racing an add with a subtract. But the operation must still be atomic
906 * since multiple entities may have a reference on the cluster.
908 * m_mclfree() is almost the same but it must contend with two entities
909 * freeing the cluster at the same time.
914 struct mbcluster *mcl = arg;
916 atomic_add_int(&mcl->mcl_refs, 1);
920 * When dereferencing a cluster we have to deal with a N->0 race, where
921 * N entities free their references simultaniously. To do this we use
922 * atomic_fetchadd_int().
927 struct mbcluster *mcl = arg;
929 if (atomic_fetchadd_int(&mcl->mcl_refs, -1) == 1)
930 objcache_put(mclmeta_cache, mcl);
934 * Free a single mbuf and any associated external storage. The successor,
935 * if any, is returned.
937 * We do need to check non-first mbuf for m_aux, since some of existing
938 * code does not call M_PREPEND properly.
939 * (example: call to bpf_mtap from drivers)
945 _m_free(struct mbuf *m, const char *func)
950 m_free(struct mbuf *m)
955 struct globaldata *gd = mycpu;
957 KASSERT(m->m_type != MT_FREE, ("freeing free mbuf %p", m));
958 KASSERT(M_TRAILINGSPACE(m) >= 0, ("overflowed mbuf %p", m));
959 atomic_subtract_long_nonlocked(&mbtypes[gd->gd_cpuid][m->m_type], 1);
964 * Make sure the mbuf is in constructed state before returning it
970 m->m_hdr.mh_lastfunc = func;
973 KKASSERT(m->m_nextpkt == NULL);
975 if (m->m_nextpkt != NULL) {
976 static int afewtimes = 10;
978 if (afewtimes-- > 0) {
979 kprintf("mfree: m->m_nextpkt != NULL\n");
985 if (m->m_flags & M_PKTHDR) {
986 m_tag_delete_chain(m); /* eliminate XXX JH */
989 m->m_flags &= (M_EXT | M_EXT_CLUSTER | M_CLCACHE | M_PHCACHE);
992 * Clean the M_PKTHDR state so we can return the mbuf to its original
993 * cache. This is based on the PHCACHE flag which tells us whether
994 * the mbuf was originally allocated out of a packet-header cache
995 * or a non-packet-header cache.
997 if (m->m_flags & M_PHCACHE) {
998 m->m_flags |= M_PKTHDR;
999 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */
1000 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */
1001 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */
1002 SLIST_INIT(&m->m_pkthdr.tags);
1006 * Handle remaining flags combinations. M_CLCACHE tells us whether
1007 * the mbuf was originally allocated from a cluster cache or not,
1008 * and is totally separate from whether the mbuf is currently
1009 * associated with a cluster.
1011 switch(m->m_flags & (M_CLCACHE | M_EXT | M_EXT_CLUSTER)) {
1012 case M_CLCACHE | M_EXT | M_EXT_CLUSTER:
1014 * mbuf+cluster cache case. The mbuf was allocated from the
1015 * combined mbuf_cluster cache and can be returned to the
1016 * cache if the cluster hasn't been shared.
1018 if (m_sharecount(m) == 1) {
1020 * The cluster has not been shared, we can just
1021 * reset the data pointer and return the mbuf
1022 * to the cluster cache. Note that the reference
1023 * count is left intact (it is still associated with
1026 m->m_data = m->m_ext.ext_buf;
1027 if (m->m_flags & M_PHCACHE)
1028 objcache_put(mbufphdrcluster_cache, m);
1030 objcache_put(mbufcluster_cache, m);
1031 atomic_subtract_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters, 1);
1034 * Hell. Someone else has a ref on this cluster,
1035 * we have to disconnect it which means we can't
1036 * put it back into the mbufcluster_cache, we
1037 * have to destroy the mbuf.
1039 * Other mbuf references to the cluster will typically
1040 * be M_EXT | M_EXT_CLUSTER but without M_CLCACHE.
1042 * XXX we could try to connect another cluster to
1045 m->m_ext.ext_free(m->m_ext.ext_arg);
1046 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER);
1047 if (m->m_flags & M_PHCACHE)
1048 objcache_dtor(mbufphdrcluster_cache, m);
1050 objcache_dtor(mbufcluster_cache, m);
1053 case M_EXT | M_EXT_CLUSTER:
1055 * Normal cluster associated with an mbuf that was allocated
1056 * from the normal mbuf pool rather then the cluster pool.
1057 * The cluster has to be independantly disassociated from the
1060 if (m_sharecount(m) == 1)
1061 atomic_subtract_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters, 1);
1065 * Normal cluster association case, disconnect the cluster from
1066 * the mbuf. The cluster may or may not be custom.
1068 m->m_ext.ext_free(m->m_ext.ext_arg);
1069 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER);
1073 * return the mbuf to the mbuf cache.
1075 if (m->m_flags & M_PHCACHE) {
1076 m->m_data = m->m_pktdat;
1077 objcache_put(mbufphdr_cache, m);
1079 m->m_data = m->m_dat;
1080 objcache_put(mbuf_cache, m);
1082 atomic_subtract_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mbufs, 1);
1086 panic("bad mbuf flags %p %08x\n", m, m->m_flags);
1095 _m_freem(struct mbuf *m, const char *func)
1098 m = _m_free(m, func);
1104 m_freem(struct mbuf *m)
1113 * mbuf utility routines
1117 * Lesser-used path for M_PREPEND: allocate new mbuf to prepend to chain and
1121 m_prepend(struct mbuf *m, int len, int how)
1125 if (m->m_flags & M_PKTHDR)
1126 mn = m_gethdr(how, m->m_type);
1128 mn = m_get(how, m->m_type);
1133 if (m->m_flags & M_PKTHDR)
1134 M_MOVE_PKTHDR(mn, m);
1144 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
1145 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
1146 * The wait parameter is a choice of MB_WAIT/MB_DONTWAIT from caller.
1147 * Note that the copy is read-only, because clusters are not copied,
1148 * only their reference counts are incremented.
1151 m_copym(const struct mbuf *m, int off0, int len, int wait)
1153 struct mbuf *n, **np;
1158 KASSERT(off >= 0, ("m_copym, negative off %d", off));
1159 KASSERT(len >= 0, ("m_copym, negative len %d", len));
1160 if (off == 0 && (m->m_flags & M_PKTHDR))
1163 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
1173 KASSERT(len == M_COPYALL,
1174 ("m_copym, length > size of mbuf chain"));
1178 * Because we are sharing any cluster attachment below,
1179 * be sure to get an mbuf that does not have a cluster
1180 * associated with it.
1183 n = m_gethdr(wait, m->m_type);
1185 n = m_get(wait, m->m_type);
1190 if (!m_dup_pkthdr(n, m, wait))
1192 if (len == M_COPYALL)
1193 n->m_pkthdr.len -= off0;
1195 n->m_pkthdr.len = len;
1198 n->m_len = min(len, m->m_len - off);
1199 if (m->m_flags & M_EXT) {
1200 KKASSERT((n->m_flags & M_EXT) == 0);
1201 n->m_data = m->m_data + off;
1202 m->m_ext.ext_ref(m->m_ext.ext_arg);
1203 n->m_ext = m->m_ext;
1204 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1206 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
1207 (unsigned)n->m_len);
1209 if (len != M_COPYALL)
1216 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1220 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1225 * Copy an entire packet, including header (which must be present).
1226 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
1227 * Note that the copy is read-only, because clusters are not copied,
1228 * only their reference counts are incremented.
1229 * Preserve alignment of the first mbuf so if the creator has left
1230 * some room at the beginning (e.g. for inserting protocol headers)
1231 * the copies also have the room available.
1234 m_copypacket(struct mbuf *m, int how)
1236 struct mbuf *top, *n, *o;
1238 n = m_gethdr(how, m->m_type);
1243 if (!m_dup_pkthdr(n, m, how))
1245 n->m_len = m->m_len;
1246 if (m->m_flags & M_EXT) {
1247 KKASSERT((n->m_flags & M_EXT) == 0);
1248 n->m_data = m->m_data;
1249 m->m_ext.ext_ref(m->m_ext.ext_arg);
1250 n->m_ext = m->m_ext;
1251 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1253 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
1254 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
1259 o = m_get(how, m->m_type);
1266 n->m_len = m->m_len;
1267 if (m->m_flags & M_EXT) {
1268 KKASSERT((n->m_flags & M_EXT) == 0);
1269 n->m_data = m->m_data;
1270 m->m_ext.ext_ref(m->m_ext.ext_arg);
1271 n->m_ext = m->m_ext;
1272 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1274 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
1282 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1287 * Copy data from an mbuf chain starting "off" bytes from the beginning,
1288 * continuing for "len" bytes, into the indicated buffer.
1291 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
1295 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
1296 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
1298 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
1305 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
1306 count = min(m->m_len - off, len);
1307 bcopy(mtod(m, caddr_t) + off, cp, count);
1316 * Copy a packet header mbuf chain into a completely new chain, including
1317 * copying any mbuf clusters. Use this instead of m_copypacket() when
1318 * you need a writable copy of an mbuf chain.
1321 m_dup(struct mbuf *m, int how)
1323 struct mbuf **p, *top = NULL;
1324 int remain, moff, nsize;
1329 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __func__));
1331 /* While there's more data, get a new mbuf, tack it on, and fill it */
1332 remain = m->m_pkthdr.len;
1335 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
1338 /* Get the next new mbuf */
1339 n = m_getl(remain, how, m->m_type, top == NULL ? M_PKTHDR : 0,
1344 if (!m_dup_pkthdr(n, m, how))
1347 /* Link it into the new chain */
1351 /* Copy data from original mbuf(s) into new mbuf */
1353 while (n->m_len < nsize && m != NULL) {
1354 int chunk = min(nsize - n->m_len, m->m_len - moff);
1356 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
1360 if (moff == m->m_len) {
1366 /* Check correct total mbuf length */
1367 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
1368 ("%s: bogus m_pkthdr.len", __func__));
1375 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1380 * Copy the non-packet mbuf data chain into a new set of mbufs, including
1381 * copying any mbuf clusters. This is typically used to realign a data
1382 * chain by nfs_realign().
1384 * The original chain is left intact. how should be MB_WAIT or MB_DONTWAIT
1385 * and NULL can be returned if MB_DONTWAIT is passed.
1387 * Be careful to use cluster mbufs, a large mbuf chain converted to non
1388 * cluster mbufs can exhaust our supply of mbufs.
1391 m_dup_data(struct mbuf *m, int how)
1393 struct mbuf **p, *n, *top = NULL;
1394 int mlen, moff, chunk, gsize, nsize;
1403 * Optimize the mbuf allocation but do not get too carried away.
1405 if (m->m_next || m->m_len > MLEN)
1416 * Scan the mbuf chain until nothing is left, the new mbuf chain
1417 * will be allocated on the fly as needed.
1424 KKASSERT(m->m_type == MT_DATA);
1426 n = m_getl(gsize, how, MT_DATA, 0, &nsize);
1433 chunk = imin(mlen, nsize);
1434 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
1449 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1454 * Concatenate mbuf chain n to m.
1455 * Both chains must be of the same type (e.g. MT_DATA).
1456 * Any m_pkthdr is not updated.
1459 m_cat(struct mbuf *m, struct mbuf *n)
1463 if (m->m_flags & M_EXT ||
1464 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
1465 /* just join the two chains */
1469 /* splat the data from one into the other */
1470 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1472 m->m_len += n->m_len;
1478 m_adj(struct mbuf *mp, int req_len)
1484 if ((m = mp) == NULL)
1490 while (m != NULL && len > 0) {
1491 if (m->m_len <= len) {
1502 if (mp->m_flags & M_PKTHDR)
1503 m->m_pkthdr.len -= (req_len - len);
1506 * Trim from tail. Scan the mbuf chain,
1507 * calculating its length and finding the last mbuf.
1508 * If the adjustment only affects this mbuf, then just
1509 * adjust and return. Otherwise, rescan and truncate
1510 * after the remaining size.
1516 if (m->m_next == NULL)
1520 if (m->m_len >= len) {
1522 if (mp->m_flags & M_PKTHDR)
1523 mp->m_pkthdr.len -= len;
1530 * Correct length for chain is "count".
1531 * Find the mbuf with last data, adjust its length,
1532 * and toss data from remaining mbufs on chain.
1535 if (m->m_flags & M_PKTHDR)
1536 m->m_pkthdr.len = count;
1537 for (; m; m = m->m_next) {
1538 if (m->m_len >= count) {
1545 (m = m->m_next) ->m_len = 0;
1550 * Set the m_data pointer of a newly-allocated mbuf
1551 * to place an object of the specified size at the
1552 * end of the mbuf, longword aligned.
1555 m_align(struct mbuf *m, int len)
1559 if (m->m_flags & M_EXT)
1560 adjust = m->m_ext.ext_size - len;
1561 else if (m->m_flags & M_PKTHDR)
1562 adjust = MHLEN - len;
1564 adjust = MLEN - len;
1565 m->m_data += adjust &~ (sizeof(long)-1);
1569 * Rearrange an mbuf chain so that len bytes are contiguous
1570 * and in the data area of an mbuf (so that mtod will work for a structure
1571 * of size len). Returns the resulting mbuf chain on success, frees it and
1572 * returns null on failure. If there is room, it will add up to
1573 * max_protohdr-len extra bytes to the contiguous region in an attempt to
1574 * avoid being called next time.
1577 m_pullup(struct mbuf *n, int len)
1584 * If first mbuf has no cluster, and has room for len bytes
1585 * without shifting current data, pullup into it,
1586 * otherwise allocate a new mbuf to prepend to the chain.
1588 if (!(n->m_flags & M_EXT) &&
1589 n->m_data + len < &n->m_dat[MLEN] &&
1591 if (n->m_len >= len)
1599 if (n->m_flags & M_PKTHDR)
1600 m = m_gethdr(MB_DONTWAIT, n->m_type);
1602 m = m_get(MB_DONTWAIT, n->m_type);
1606 if (n->m_flags & M_PKTHDR)
1607 M_MOVE_PKTHDR(m, n);
1609 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1611 count = min(min(max(len, max_protohdr), space), n->m_len);
1612 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1622 } while (len > 0 && n);
1631 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1636 * Partition an mbuf chain in two pieces, returning the tail --
1637 * all but the first len0 bytes. In case of failure, it returns NULL and
1638 * attempts to restore the chain to its original state.
1640 * Note that the resulting mbufs might be read-only, because the new
1641 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1642 * the "breaking point" happens to lie within a cluster mbuf. Use the
1643 * M_WRITABLE() macro to check for this case.
1646 m_split(struct mbuf *m0, int len0, int wait)
1649 unsigned len = len0, remain;
1651 for (m = m0; m && len > m->m_len; m = m->m_next)
1655 remain = m->m_len - len;
1656 if (m0->m_flags & M_PKTHDR) {
1657 n = m_gethdr(wait, m0->m_type);
1660 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1661 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1662 m0->m_pkthdr.len = len0;
1663 if (m->m_flags & M_EXT)
1665 if (remain > MHLEN) {
1666 /* m can't be the lead packet */
1668 n->m_next = m_split(m, len, wait);
1669 if (n->m_next == NULL) {
1677 MH_ALIGN(n, remain);
1678 } else if (remain == 0) {
1683 n = m_get(wait, m->m_type);
1689 if (m->m_flags & M_EXT) {
1690 KKASSERT((n->m_flags & M_EXT) == 0);
1691 n->m_data = m->m_data + len;
1692 m->m_ext.ext_ref(m->m_ext.ext_arg);
1693 n->m_ext = m->m_ext;
1694 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1696 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1700 n->m_next = m->m_next;
1706 * Routine to copy from device local memory into mbufs.
1707 * Note: "offset" is ill-defined and always called as 0, so ignore it.
1710 m_devget(char *buf, int len, int offset, struct ifnet *ifp,
1711 void (*copy)(volatile const void *from, volatile void *to, size_t length))
1713 struct mbuf *m, *mfirst = NULL, **mtail;
1722 m = m_getl(len, MB_DONTWAIT, MT_DATA, flags, &nsize);
1727 m->m_len = min(len, nsize);
1729 if (flags & M_PKTHDR) {
1730 if (len + max_linkhdr <= nsize)
1731 m->m_data += max_linkhdr;
1732 m->m_pkthdr.rcvif = ifp;
1733 m->m_pkthdr.len = len;
1737 copy(buf, m->m_data, (unsigned)m->m_len);
1748 * Routine to pad mbuf to the specified length 'padto'.
1751 m_devpad(struct mbuf *m, int padto)
1753 struct mbuf *last = NULL;
1756 if (padto <= m->m_pkthdr.len)
1759 padlen = padto - m->m_pkthdr.len;
1761 /* if there's only the packet-header and we can pad there, use it. */
1762 if (m->m_pkthdr.len == m->m_len && M_TRAILINGSPACE(m) >= padlen) {
1766 * Walk packet chain to find last mbuf. We will either
1767 * pad there, or append a new mbuf and pad it
1769 for (last = m; last->m_next != NULL; last = last->m_next)
1772 /* `last' now points to last in chain. */
1773 if (M_TRAILINGSPACE(last) < padlen) {
1776 /* Allocate new empty mbuf, pad it. Compact later. */
1777 MGET(n, MB_DONTWAIT, MT_DATA);
1785 KKASSERT(M_TRAILINGSPACE(last) >= padlen);
1786 KKASSERT(M_WRITABLE(last));
1788 /* Now zero the pad area */
1789 bzero(mtod(last, char *) + last->m_len, padlen);
1790 last->m_len += padlen;
1791 m->m_pkthdr.len += padlen;
1796 * Copy data from a buffer back into the indicated mbuf chain,
1797 * starting "off" bytes from the beginning, extending the mbuf
1798 * chain if necessary.
1801 m_copyback(struct mbuf *m0, int off, int len, caddr_t cp)
1804 struct mbuf *m = m0, *n;
1809 while (off > (mlen = m->m_len)) {
1812 if (m->m_next == NULL) {
1813 n = m_getclr(MB_DONTWAIT, m->m_type);
1816 n->m_len = min(MLEN, len + off);
1822 mlen = min (m->m_len - off, len);
1823 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen);
1831 if (m->m_next == NULL) {
1832 n = m_get(MB_DONTWAIT, m->m_type);
1835 n->m_len = min(MLEN, len);
1840 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1841 m->m_pkthdr.len = totlen;
1845 * Append the specified data to the indicated mbuf chain,
1846 * Extend the mbuf chain if the new data does not fit in
1849 * Return 1 if able to complete the job; otherwise 0.
1852 m_append(struct mbuf *m0, int len, c_caddr_t cp)
1855 int remainder, space;
1857 for (m = m0; m->m_next != NULL; m = m->m_next)
1860 space = M_TRAILINGSPACE(m);
1863 * Copy into available space.
1865 if (space > remainder)
1867 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1869 cp += space, remainder -= space;
1871 while (remainder > 0) {
1873 * Allocate a new mbuf; could check space
1874 * and allocate a cluster instead.
1876 n = m_get(MB_DONTWAIT, m->m_type);
1879 n->m_len = min(MLEN, remainder);
1880 bcopy(cp, mtod(n, caddr_t), n->m_len);
1881 cp += n->m_len, remainder -= n->m_len;
1885 if (m0->m_flags & M_PKTHDR)
1886 m0->m_pkthdr.len += len - remainder;
1887 return (remainder == 0);
1891 * Apply function f to the data in an mbuf chain starting "off" bytes from
1892 * the beginning, continuing for "len" bytes.
1895 m_apply(struct mbuf *m, int off, int len,
1896 int (*f)(void *, void *, u_int), void *arg)
1901 KASSERT(off >= 0, ("m_apply, negative off %d", off));
1902 KASSERT(len >= 0, ("m_apply, negative len %d", len));
1904 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1911 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1912 count = min(m->m_len - off, len);
1913 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1924 * Return a pointer to mbuf/offset of location in mbuf chain.
1927 m_getptr(struct mbuf *m, int loc, int *off)
1931 /* Normal end of search. */
1932 if (m->m_len > loc) {
1937 if (m->m_next == NULL) {
1939 /* Point at the end of valid data. */
1952 m_print(const struct mbuf *m)
1955 const struct mbuf *m2;
1957 len = m->m_pkthdr.len;
1960 kprintf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-");
1968 * "Move" mbuf pkthdr from "from" to "to".
1969 * "from" must have M_PKTHDR set, and "to" must be empty.
1972 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
1974 KASSERT((to->m_flags & M_PKTHDR), ("m_move_pkthdr: not packet header"));
1976 to->m_flags |= from->m_flags & M_COPYFLAGS;
1977 to->m_pkthdr = from->m_pkthdr; /* especially tags */
1978 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
1982 * Duplicate "from"'s mbuf pkthdr in "to".
1983 * "from" must have M_PKTHDR set, and "to" must be empty.
1984 * In particular, this does a deep copy of the packet tags.
1987 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how)
1989 KASSERT((to->m_flags & M_PKTHDR), ("m_dup_pkthdr: not packet header"));
1991 to->m_flags = (from->m_flags & M_COPYFLAGS) |
1992 (to->m_flags & ~M_COPYFLAGS);
1993 to->m_pkthdr = from->m_pkthdr;
1994 SLIST_INIT(&to->m_pkthdr.tags);
1995 return (m_tag_copy_chain(to, from, how));
1999 * Defragment a mbuf chain, returning the shortest possible
2000 * chain of mbufs and clusters. If allocation fails and
2001 * this cannot be completed, NULL will be returned, but
2002 * the passed in chain will be unchanged. Upon success,
2003 * the original chain will be freed, and the new chain
2006 * If a non-packet header is passed in, the original
2007 * mbuf (chain?) will be returned unharmed.
2009 * m_defrag_nofree doesn't free the passed in mbuf.
2012 m_defrag(struct mbuf *m0, int how)
2016 if ((m_new = m_defrag_nofree(m0, how)) == NULL)
2024 m_defrag_nofree(struct mbuf *m0, int how)
2026 struct mbuf *m_new = NULL, *m_final = NULL;
2027 int progress = 0, length, nsize;
2029 if (!(m0->m_flags & M_PKTHDR))
2032 #ifdef MBUF_STRESS_TEST
2033 if (m_defragrandomfailures) {
2034 int temp = karc4random() & 0xff;
2040 m_final = m_getl(m0->m_pkthdr.len, how, MT_DATA, M_PKTHDR, &nsize);
2041 if (m_final == NULL)
2043 m_final->m_len = 0; /* in case m0->m_pkthdr.len is zero */
2045 if (m_dup_pkthdr(m_final, m0, how) == 0)
2050 while (progress < m0->m_pkthdr.len) {
2051 length = m0->m_pkthdr.len - progress;
2052 if (length > MCLBYTES)
2055 if (m_new == NULL) {
2056 m_new = m_getl(length, how, MT_DATA, 0, &nsize);
2061 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
2063 m_new->m_len = length;
2064 if (m_new != m_final)
2065 m_cat(m_final, m_new);
2068 if (m0->m_next == NULL)
2071 m_defragbytes += m_final->m_pkthdr.len;
2082 * Move data from uio into mbufs.
2085 m_uiomove(struct uio *uio)
2087 struct mbuf *m; /* current working mbuf */
2088 struct mbuf *head = NULL; /* result mbuf chain */
2089 struct mbuf **mp = &head;
2090 int flags = M_PKTHDR;
2096 if (uio->uio_resid > INT_MAX)
2099 resid = (int)uio->uio_resid;
2100 m = m_getl(resid, MB_WAIT, MT_DATA, flags, &nsize);
2102 m->m_pkthdr.len = 0;
2103 /* Leave room for protocol headers. */
2108 m->m_len = imin(nsize, resid);
2109 error = uiomove(mtod(m, caddr_t), m->m_len, uio);
2116 head->m_pkthdr.len += m->m_len;
2117 } while (uio->uio_resid > 0);
2127 m_last(struct mbuf *m)
2135 * Return the number of bytes in an mbuf chain.
2136 * If lastm is not NULL, also return the last mbuf.
2139 m_lengthm(struct mbuf *m, struct mbuf **lastm)
2142 struct mbuf *prev = m;
2155 * Like m_lengthm(), except also keep track of mbuf usage.
2158 m_countm(struct mbuf *m, struct mbuf **lastm, u_int *pmbcnt)
2160 u_int len = 0, mbcnt = 0;
2161 struct mbuf *prev = m;
2166 if (m->m_flags & M_EXT)
2167 mbcnt += m->m_ext.ext_size;