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, "");
264 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW,
265 &max_protohdr, 0, "");
266 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, "");
267 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW,
268 &max_datalen, 0, "");
269 SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW,
271 static int do_mbstat(SYSCTL_HANDLER_ARGS);
273 SYSCTL_PROC(_kern_ipc, KIPC_MBSTAT, mbstat, CTLTYPE_STRUCT|CTLFLAG_RD,
274 0, 0, do_mbstat, "S,mbstat", "");
276 static int do_mbtypes(SYSCTL_HANDLER_ARGS);
278 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbtypes, CTLTYPE_ULONG|CTLFLAG_RD,
279 0, 0, do_mbtypes, "LU", "");
282 do_mbstat(SYSCTL_HANDLER_ARGS)
284 struct mbstat mbstat_total;
285 struct mbstat *mbstat_totalp;
288 bzero(&mbstat_total, sizeof(mbstat_total));
289 mbstat_totalp = &mbstat_total;
291 for (i = 0; i < ncpus; i++)
293 mbstat_total.m_mbufs += mbstat[i].m_mbufs;
294 mbstat_total.m_clusters += mbstat[i].m_clusters;
295 mbstat_total.m_spare += mbstat[i].m_spare;
296 mbstat_total.m_clfree += mbstat[i].m_clfree;
297 mbstat_total.m_drops += mbstat[i].m_drops;
298 mbstat_total.m_wait += mbstat[i].m_wait;
299 mbstat_total.m_drain += mbstat[i].m_drain;
300 mbstat_total.m_mcfail += mbstat[i].m_mcfail;
301 mbstat_total.m_mpfail += mbstat[i].m_mpfail;
305 * The following fields are not cumulative fields so just
306 * get their values once.
308 mbstat_total.m_msize = mbstat[0].m_msize;
309 mbstat_total.m_mclbytes = mbstat[0].m_mclbytes;
310 mbstat_total.m_minclsize = mbstat[0].m_minclsize;
311 mbstat_total.m_mlen = mbstat[0].m_mlen;
312 mbstat_total.m_mhlen = mbstat[0].m_mhlen;
314 return(sysctl_handle_opaque(oidp, mbstat_totalp, sizeof(mbstat_total), req));
318 do_mbtypes(SYSCTL_HANDLER_ARGS)
320 u_long totals[MT_NTYPES];
323 for (i = 0; i < MT_NTYPES; i++)
326 for (i = 0; i < ncpus; i++)
328 for (j = 0; j < MT_NTYPES; j++)
329 totals[j] += mbtypes[i][j];
332 return(sysctl_handle_opaque(oidp, totals, sizeof(totals), req));
336 * These are read-only because we do not currently have any code
337 * to adjust the objcache limits after the fact. The variables
338 * may only be set as boot-time tunables.
340 SYSCTL_INT(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLFLAG_RD,
341 &nmbclusters, 0, "Maximum number of mbuf clusters available");
342 SYSCTL_INT(_kern_ipc, OID_AUTO, nmbufs, CTLFLAG_RD, &nmbufs, 0,
343 "Maximum number of mbufs available");
345 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
346 &m_defragpackets, 0, "");
347 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
348 &m_defragbytes, 0, "");
349 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
350 &m_defraguseless, 0, "");
351 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
352 &m_defragfailure, 0, "");
353 #ifdef MBUF_STRESS_TEST
354 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
355 &m_defragrandomfailures, 0, "");
358 static MALLOC_DEFINE(M_MBUF, "mbuf", "mbuf");
359 static MALLOC_DEFINE(M_MBUFCL, "mbufcl", "mbufcl");
360 static MALLOC_DEFINE(M_MCLMETA, "mclmeta", "mclmeta");
362 static void m_reclaim (void);
363 static void m_mclref(void *arg);
364 static void m_mclfree(void *arg);
367 * NOTE: Default NMBUFS must take into account a possible DOS attack
368 * using fd passing on unix domain sockets.
371 #define NMBCLUSTERS (512 + maxusers * 16)
374 #define NMBUFS (nmbclusters * 2 + maxfiles)
378 * Perform sanity checks of tunables declared above.
381 tunable_mbinit(void *dummy)
384 * This has to be done before VM init.
386 nmbclusters = NMBCLUSTERS;
387 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
389 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs);
391 if (nmbufs < nmbclusters * 2)
392 nmbufs = nmbclusters * 2;
394 SYSINIT(tunable_mbinit, SI_BOOT1_TUNABLES, SI_ORDER_ANY,
395 tunable_mbinit, NULL);
397 /* "number of clusters of pages" */
403 * The mbuf object cache only guarantees that m_next and m_nextpkt are
404 * NULL and that m_data points to the beginning of the data area. In
405 * particular, m_len and m_pkthdr.len are uninitialized. It is the
406 * responsibility of the caller to initialize those fields before use.
409 static boolean_t __inline
410 mbuf_ctor(void *obj, void *private, int ocflags)
412 struct mbuf *m = obj;
416 m->m_data = m->m_dat;
423 * Initialize the mbuf and the packet header fields.
426 mbufphdr_ctor(void *obj, void *private, int ocflags)
428 struct mbuf *m = obj;
432 m->m_data = m->m_pktdat;
433 m->m_flags = M_PKTHDR | M_PHCACHE;
435 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */
436 SLIST_INIT(&m->m_pkthdr.tags);
437 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */
438 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */
444 * A mbcluster object consists of 2K (MCLBYTES) cluster and a refcount.
447 mclmeta_ctor(void *obj, void *private, int ocflags)
449 struct mbcluster *cl = obj;
452 if (ocflags & M_NOWAIT)
453 buf = kmalloc(MCLBYTES, M_MBUFCL, M_NOWAIT | M_ZERO);
455 buf = kmalloc(MCLBYTES, M_MBUFCL, M_INTWAIT | M_ZERO);
464 mclmeta_dtor(void *obj, void *private)
466 struct mbcluster *mcl = obj;
468 KKASSERT(mcl->mcl_refs == 0);
469 kfree(mcl->mcl_data, M_MBUFCL);
473 linkcluster(struct mbuf *m, struct mbcluster *cl)
476 * Add the cluster to the mbuf. The caller will detect that the
477 * mbuf now has an attached cluster.
479 m->m_ext.ext_arg = cl;
480 m->m_ext.ext_buf = cl->mcl_data;
481 m->m_ext.ext_ref = m_mclref;
482 m->m_ext.ext_free = m_mclfree;
483 m->m_ext.ext_size = MCLBYTES;
484 atomic_add_int(&cl->mcl_refs, 1);
486 m->m_data = m->m_ext.ext_buf;
487 m->m_flags |= M_EXT | M_EXT_CLUSTER;
491 mbufphdrcluster_ctor(void *obj, void *private, int ocflags)
493 struct mbuf *m = obj;
494 struct mbcluster *cl;
496 mbufphdr_ctor(obj, private, ocflags);
497 cl = objcache_get(mclmeta_cache, ocflags);
499 ++mbstat[mycpu->gd_cpuid].m_drops;
502 m->m_flags |= M_CLCACHE;
508 mbufcluster_ctor(void *obj, void *private, int ocflags)
510 struct mbuf *m = obj;
511 struct mbcluster *cl;
513 mbuf_ctor(obj, private, ocflags);
514 cl = objcache_get(mclmeta_cache, ocflags);
516 ++mbstat[mycpu->gd_cpuid].m_drops;
519 m->m_flags |= M_CLCACHE;
525 * Used for both the cluster and cluster PHDR caches.
527 * The mbuf may have lost its cluster due to sharing, deal
528 * with the situation by checking M_EXT.
531 mbufcluster_dtor(void *obj, void *private)
533 struct mbuf *m = obj;
534 struct mbcluster *mcl;
536 if (m->m_flags & M_EXT) {
537 KKASSERT((m->m_flags & M_EXT_CLUSTER) != 0);
538 mcl = m->m_ext.ext_arg;
539 KKASSERT(mcl->mcl_refs == 1);
541 objcache_put(mclmeta_cache, mcl);
545 struct objcache_malloc_args mbuf_malloc_args = { MSIZE, M_MBUF };
546 struct objcache_malloc_args mclmeta_malloc_args =
547 { sizeof(struct mbcluster), M_MCLMETA };
553 int mb_limit, cl_limit;
558 * Initialize statistics
560 for (i = 0; i < ncpus; i++) {
561 atomic_set_long_nonlocked(&mbstat[i].m_msize, MSIZE);
562 atomic_set_long_nonlocked(&mbstat[i].m_mclbytes, MCLBYTES);
563 atomic_set_long_nonlocked(&mbstat[i].m_minclsize, MINCLSIZE);
564 atomic_set_long_nonlocked(&mbstat[i].m_mlen, MLEN);
565 atomic_set_long_nonlocked(&mbstat[i].m_mhlen, MHLEN);
569 * Create objtect caches and save cluster limits, which will
570 * be used to adjust backing kmalloc pools' limit later.
573 mb_limit = cl_limit = 0;
576 mbuf_cache = objcache_create("mbuf", &limit, 0,
577 mbuf_ctor, NULL, NULL,
578 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
582 mbufphdr_cache = objcache_create("mbuf pkt hdr", &limit, 64,
583 mbufphdr_ctor, NULL, NULL,
584 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
587 cl_limit = nmbclusters;
588 mclmeta_cache = objcache_create("cluster mbuf", &cl_limit, 0,
589 mclmeta_ctor, mclmeta_dtor, NULL,
590 objcache_malloc_alloc, objcache_malloc_free, &mclmeta_malloc_args);
593 mbufcluster_cache = objcache_create("mbuf + cluster", &limit, 0,
594 mbufcluster_ctor, mbufcluster_dtor, NULL,
595 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
599 mbufphdrcluster_cache = objcache_create("mbuf pkt hdr + cluster",
600 &limit, 64, mbufphdrcluster_ctor, mbufcluster_dtor, NULL,
601 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
605 * Adjust backing kmalloc pools' limit
607 * NOTE: We raise the limit by another 1/8 to take the effect
608 * of loosememuse into account.
610 cl_limit += cl_limit / 8;
611 kmalloc_raise_limit(mclmeta_malloc_args.mtype,
612 mclmeta_malloc_args.objsize * cl_limit);
613 kmalloc_raise_limit(M_MBUFCL, MCLBYTES * cl_limit);
615 mb_limit += mb_limit / 8;
616 kmalloc_raise_limit(mbuf_malloc_args.mtype,
617 mbuf_malloc_args.objsize * mb_limit);
621 * Return the number of references to this mbuf's data. 0 is returned
622 * if the mbuf is not M_EXT, a reference count is returned if it is
623 * M_EXT | M_EXT_CLUSTER, and 99 is returned if it is a special M_EXT.
626 m_sharecount(struct mbuf *m)
628 switch (m->m_flags & (M_EXT | M_EXT_CLUSTER)) {
633 case M_EXT | M_EXT_CLUSTER:
634 return (((struct mbcluster *)m->m_ext.ext_arg)->mcl_refs);
637 return (0); /* to shut up compiler */
641 * change mbuf to new type
644 m_chtype(struct mbuf *m, int type)
646 struct globaldata *gd = mycpu;
648 atomic_add_long_nonlocked(&mbtypes[gd->gd_cpuid][type], 1);
649 atomic_subtract_long_nonlocked(&mbtypes[gd->gd_cpuid][m->m_type], 1);
650 atomic_set_short_nonlocked(&m->m_type, type);
659 kprintf("Debug: m_reclaim() called\n");
661 SLIST_FOREACH(dp, &domains, dom_next) {
662 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
667 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_drain, 1);
671 updatestats(struct mbuf *m, int type)
673 struct globaldata *gd = mycpu;
678 KASSERT(m->m_next == NULL, ("mbuf %p: bad m_next in get", m));
679 KASSERT(m->m_nextpkt == NULL, ("mbuf %p: bad m_nextpkt in get", m));
682 atomic_add_long_nonlocked(&mbtypes[gd->gd_cpuid][type], 1);
683 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mbufs, 1);
691 m_get(int how, int type)
695 int ocf = MBTOM(how);
699 m = objcache_get(mbuf_cache, ocf);
702 if ((how & MB_TRYWAIT) && ntries++ == 0) {
703 struct objcache *reclaimlist[] = {
706 mbufphdrcluster_cache
708 const int nreclaims = __arysize(reclaimlist);
710 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf))
714 ++mbstat[mycpu->gd_cpuid].m_drops;
718 KASSERT(m->m_data == m->m_dat, ("mbuf %p: bad m_data in get", m));
722 updatestats(m, type);
727 m_gethdr(int how, int type)
730 int ocf = MBTOM(how);
735 m = objcache_get(mbufphdr_cache, ocf);
738 if ((how & MB_TRYWAIT) && ntries++ == 0) {
739 struct objcache *reclaimlist[] = {
741 mbufcluster_cache, mbufphdrcluster_cache
743 const int nreclaims = __arysize(reclaimlist);
745 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf))
749 ++mbstat[mycpu->gd_cpuid].m_drops;
753 KASSERT(m->m_data == m->m_pktdat, ("mbuf %p: bad m_data in get", m));
758 updatestats(m, type);
763 * Get a mbuf (not a mbuf cluster!) and zero it.
767 m_getclr(int how, int type)
771 m = m_get(how, type);
773 bzero(m->m_data, MLEN);
778 * Returns an mbuf with an attached cluster.
779 * Because many network drivers use this kind of buffers a lot, it is
780 * convenient to keep a small pool of free buffers of this kind.
781 * Even a small size such as 10 gives about 10% improvement in the
782 * forwarding rate in a bridge or router.
785 m_getcl(int how, short type, int flags)
788 int ocflags = MBTOM(how);
793 if (flags & M_PKTHDR)
794 m = objcache_get(mbufphdrcluster_cache, ocflags);
796 m = objcache_get(mbufcluster_cache, ocflags);
799 if ((how & MB_TRYWAIT) && ntries++ == 0) {
800 struct objcache *reclaimlist[1];
802 if (flags & M_PKTHDR)
803 reclaimlist[0] = mbufcluster_cache;
805 reclaimlist[0] = mbufphdrcluster_cache;
806 if (!objcache_reclaimlist(reclaimlist, 1, ocflags))
810 ++mbstat[mycpu->gd_cpuid].m_drops;
815 KASSERT(m->m_data == m->m_ext.ext_buf,
816 ("mbuf %p: bad m_data in get", m));
820 m->m_pkthdr.len = 0; /* just do it unconditonally */
824 atomic_add_long_nonlocked(&mbtypes[mycpu->gd_cpuid][type], 1);
825 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters, 1);
830 * Allocate chain of requested length.
833 m_getc(int len, int how, int type)
835 struct mbuf *n, *nfirst = NULL, **ntail = &nfirst;
839 n = m_getl(len, how, type, 0, &nsize);
855 * Allocate len-worth of mbufs and/or mbuf clusters (whatever fits best)
856 * and return a pointer to the head of the allocated chain. If m0 is
857 * non-null, then we assume that it is a single mbuf or an mbuf chain to
858 * which we want len bytes worth of mbufs and/or clusters attached, and so
859 * if we succeed in allocating it, we will just return a pointer to m0.
861 * If we happen to fail at any point during the allocation, we will free
862 * up everything we have already allocated and return NULL.
864 * Deprecated. Use m_getc() and m_cat() instead.
867 m_getm(struct mbuf *m0, int len, int type, int how)
871 nfirst = m_getc(len, how, type);
874 m_last(m0)->m_next = nfirst;
882 * Adds a cluster to a normal mbuf, M_EXT is set on success.
883 * Deprecated. Use m_getcl() instead.
886 m_mclget(struct mbuf *m, int how)
888 struct mbcluster *mcl;
890 KKASSERT((m->m_flags & M_EXT) == 0);
891 mcl = objcache_get(mclmeta_cache, MBTOM(how));
894 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters,
897 ++mbstat[mycpu->gd_cpuid].m_drops;
902 * Updates to mbcluster must be MPSAFE. Only an entity which already has
903 * a reference to the cluster can ref it, so we are in no danger of
904 * racing an add with a subtract. But the operation must still be atomic
905 * since multiple entities may have a reference on the cluster.
907 * m_mclfree() is almost the same but it must contend with two entities
908 * freeing the cluster at the same time.
913 struct mbcluster *mcl = arg;
915 atomic_add_int(&mcl->mcl_refs, 1);
919 * When dereferencing a cluster we have to deal with a N->0 race, where
920 * N entities free their references simultaniously. To do this we use
921 * atomic_fetchadd_int().
926 struct mbcluster *mcl = arg;
928 if (atomic_fetchadd_int(&mcl->mcl_refs, -1) == 1)
929 objcache_put(mclmeta_cache, mcl);
933 * Free a single mbuf and any associated external storage. The successor,
934 * if any, is returned.
936 * We do need to check non-first mbuf for m_aux, since some of existing
937 * code does not call M_PREPEND properly.
938 * (example: call to bpf_mtap from drivers)
944 _m_free(struct mbuf *m, const char *func)
949 m_free(struct mbuf *m)
954 struct globaldata *gd = mycpu;
956 KASSERT(m->m_type != MT_FREE, ("freeing free mbuf %p", m));
957 KASSERT(M_TRAILINGSPACE(m) >= 0, ("overflowed mbuf %p", m));
958 atomic_subtract_long_nonlocked(&mbtypes[gd->gd_cpuid][m->m_type], 1);
963 * Make sure the mbuf is in constructed state before returning it
969 m->m_hdr.mh_lastfunc = func;
972 KKASSERT(m->m_nextpkt == NULL);
974 if (m->m_nextpkt != NULL) {
975 static int afewtimes = 10;
977 if (afewtimes-- > 0) {
978 kprintf("mfree: m->m_nextpkt != NULL\n");
984 if (m->m_flags & M_PKTHDR) {
985 m_tag_delete_chain(m); /* eliminate XXX JH */
988 m->m_flags &= (M_EXT | M_EXT_CLUSTER | M_CLCACHE | M_PHCACHE);
991 * Clean the M_PKTHDR state so we can return the mbuf to its original
992 * cache. This is based on the PHCACHE flag which tells us whether
993 * the mbuf was originally allocated out of a packet-header cache
994 * or a non-packet-header cache.
996 if (m->m_flags & M_PHCACHE) {
997 m->m_flags |= M_PKTHDR;
998 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */
999 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */
1000 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */
1001 SLIST_INIT(&m->m_pkthdr.tags);
1005 * Handle remaining flags combinations. M_CLCACHE tells us whether
1006 * the mbuf was originally allocated from a cluster cache or not,
1007 * and is totally separate from whether the mbuf is currently
1008 * associated with a cluster.
1010 switch(m->m_flags & (M_CLCACHE | M_EXT | M_EXT_CLUSTER)) {
1011 case M_CLCACHE | M_EXT | M_EXT_CLUSTER:
1013 * mbuf+cluster cache case. The mbuf was allocated from the
1014 * combined mbuf_cluster cache and can be returned to the
1015 * cache if the cluster hasn't been shared.
1017 if (m_sharecount(m) == 1) {
1019 * The cluster has not been shared, we can just
1020 * reset the data pointer and return the mbuf
1021 * to the cluster cache. Note that the reference
1022 * count is left intact (it is still associated with
1025 m->m_data = m->m_ext.ext_buf;
1026 if (m->m_flags & M_PHCACHE)
1027 objcache_put(mbufphdrcluster_cache, m);
1029 objcache_put(mbufcluster_cache, m);
1030 atomic_subtract_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters, 1);
1033 * Hell. Someone else has a ref on this cluster,
1034 * we have to disconnect it which means we can't
1035 * put it back into the mbufcluster_cache, we
1036 * have to destroy the mbuf.
1038 * Other mbuf references to the cluster will typically
1039 * be M_EXT | M_EXT_CLUSTER but without M_CLCACHE.
1041 * XXX we could try to connect another cluster to
1044 m->m_ext.ext_free(m->m_ext.ext_arg);
1045 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER);
1046 if (m->m_flags & M_PHCACHE)
1047 objcache_dtor(mbufphdrcluster_cache, m);
1049 objcache_dtor(mbufcluster_cache, m);
1052 case M_EXT | M_EXT_CLUSTER:
1054 * Normal cluster associated with an mbuf that was allocated
1055 * from the normal mbuf pool rather then the cluster pool.
1056 * The cluster has to be independantly disassociated from the
1059 if (m_sharecount(m) == 1)
1060 atomic_subtract_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters, 1);
1064 * Normal cluster association case, disconnect the cluster from
1065 * the mbuf. The cluster may or may not be custom.
1067 m->m_ext.ext_free(m->m_ext.ext_arg);
1068 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER);
1072 * return the mbuf to the mbuf cache.
1074 if (m->m_flags & M_PHCACHE) {
1075 m->m_data = m->m_pktdat;
1076 objcache_put(mbufphdr_cache, m);
1078 m->m_data = m->m_dat;
1079 objcache_put(mbuf_cache, m);
1081 atomic_subtract_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mbufs, 1);
1085 panic("bad mbuf flags %p %08x\n", m, m->m_flags);
1094 _m_freem(struct mbuf *m, const char *func)
1097 m = _m_free(m, func);
1103 m_freem(struct mbuf *m)
1112 * mbuf utility routines
1116 * Lesser-used path for M_PREPEND: allocate new mbuf to prepend to chain and
1120 m_prepend(struct mbuf *m, int len, int how)
1124 if (m->m_flags & M_PKTHDR)
1125 mn = m_gethdr(how, m->m_type);
1127 mn = m_get(how, m->m_type);
1132 if (m->m_flags & M_PKTHDR)
1133 M_MOVE_PKTHDR(mn, m);
1143 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
1144 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
1145 * The wait parameter is a choice of MB_WAIT/MB_DONTWAIT from caller.
1146 * Note that the copy is read-only, because clusters are not copied,
1147 * only their reference counts are incremented.
1150 m_copym(const struct mbuf *m, int off0, int len, int wait)
1152 struct mbuf *n, **np;
1157 KASSERT(off >= 0, ("m_copym, negative off %d", off));
1158 KASSERT(len >= 0, ("m_copym, negative len %d", len));
1159 if (off == 0 && (m->m_flags & M_PKTHDR))
1162 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
1172 KASSERT(len == M_COPYALL,
1173 ("m_copym, length > size of mbuf chain"));
1177 * Because we are sharing any cluster attachment below,
1178 * be sure to get an mbuf that does not have a cluster
1179 * associated with it.
1182 n = m_gethdr(wait, m->m_type);
1184 n = m_get(wait, m->m_type);
1189 if (!m_dup_pkthdr(n, m, wait))
1191 if (len == M_COPYALL)
1192 n->m_pkthdr.len -= off0;
1194 n->m_pkthdr.len = len;
1197 n->m_len = min(len, m->m_len - off);
1198 if (m->m_flags & M_EXT) {
1199 KKASSERT((n->m_flags & M_EXT) == 0);
1200 n->m_data = m->m_data + off;
1201 m->m_ext.ext_ref(m->m_ext.ext_arg);
1202 n->m_ext = m->m_ext;
1203 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1205 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
1206 (unsigned)n->m_len);
1208 if (len != M_COPYALL)
1215 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1219 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1224 * Copy an entire packet, including header (which must be present).
1225 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
1226 * Note that the copy is read-only, because clusters are not copied,
1227 * only their reference counts are incremented.
1228 * Preserve alignment of the first mbuf so if the creator has left
1229 * some room at the beginning (e.g. for inserting protocol headers)
1230 * the copies also have the room available.
1233 m_copypacket(struct mbuf *m, int how)
1235 struct mbuf *top, *n, *o;
1237 n = m_gethdr(how, m->m_type);
1242 if (!m_dup_pkthdr(n, m, how))
1244 n->m_len = m->m_len;
1245 if (m->m_flags & M_EXT) {
1246 KKASSERT((n->m_flags & M_EXT) == 0);
1247 n->m_data = m->m_data;
1248 m->m_ext.ext_ref(m->m_ext.ext_arg);
1249 n->m_ext = m->m_ext;
1250 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1252 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
1253 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
1258 o = m_get(how, m->m_type);
1265 n->m_len = m->m_len;
1266 if (m->m_flags & M_EXT) {
1267 KKASSERT((n->m_flags & M_EXT) == 0);
1268 n->m_data = m->m_data;
1269 m->m_ext.ext_ref(m->m_ext.ext_arg);
1270 n->m_ext = m->m_ext;
1271 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1273 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
1281 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1286 * Copy data from an mbuf chain starting "off" bytes from the beginning,
1287 * continuing for "len" bytes, into the indicated buffer.
1290 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
1294 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
1295 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
1297 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
1304 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
1305 count = min(m->m_len - off, len);
1306 bcopy(mtod(m, caddr_t) + off, cp, count);
1315 * Copy a packet header mbuf chain into a completely new chain, including
1316 * copying any mbuf clusters. Use this instead of m_copypacket() when
1317 * you need a writable copy of an mbuf chain.
1320 m_dup(struct mbuf *m, int how)
1322 struct mbuf **p, *top = NULL;
1323 int remain, moff, nsize;
1328 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __func__));
1330 /* While there's more data, get a new mbuf, tack it on, and fill it */
1331 remain = m->m_pkthdr.len;
1334 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
1337 /* Get the next new mbuf */
1338 n = m_getl(remain, how, m->m_type, top == NULL ? M_PKTHDR : 0,
1343 if (!m_dup_pkthdr(n, m, how))
1346 /* Link it into the new chain */
1350 /* Copy data from original mbuf(s) into new mbuf */
1352 while (n->m_len < nsize && m != NULL) {
1353 int chunk = min(nsize - n->m_len, m->m_len - moff);
1355 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
1359 if (moff == m->m_len) {
1365 /* Check correct total mbuf length */
1366 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
1367 ("%s: bogus m_pkthdr.len", __func__));
1374 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1379 * Copy the non-packet mbuf data chain into a new set of mbufs, including
1380 * copying any mbuf clusters. This is typically used to realign a data
1381 * chain by nfs_realign().
1383 * The original chain is left intact. how should be MB_WAIT or MB_DONTWAIT
1384 * and NULL can be returned if MB_DONTWAIT is passed.
1386 * Be careful to use cluster mbufs, a large mbuf chain converted to non
1387 * cluster mbufs can exhaust our supply of mbufs.
1390 m_dup_data(struct mbuf *m, int how)
1392 struct mbuf **p, *n, *top = NULL;
1393 int mlen, moff, chunk, gsize, nsize;
1402 * Optimize the mbuf allocation but do not get too carried away.
1404 if (m->m_next || m->m_len > MLEN)
1415 * Scan the mbuf chain until nothing is left, the new mbuf chain
1416 * will be allocated on the fly as needed.
1423 KKASSERT(m->m_type == MT_DATA);
1425 n = m_getl(gsize, how, MT_DATA, 0, &nsize);
1432 chunk = imin(mlen, nsize);
1433 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
1448 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1453 * Concatenate mbuf chain n to m.
1454 * Both chains must be of the same type (e.g. MT_DATA).
1455 * Any m_pkthdr is not updated.
1458 m_cat(struct mbuf *m, struct mbuf *n)
1462 if (m->m_flags & M_EXT ||
1463 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
1464 /* just join the two chains */
1468 /* splat the data from one into the other */
1469 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1471 m->m_len += n->m_len;
1477 m_adj(struct mbuf *mp, int req_len)
1483 if ((m = mp) == NULL)
1489 while (m != NULL && len > 0) {
1490 if (m->m_len <= len) {
1501 if (mp->m_flags & M_PKTHDR)
1502 m->m_pkthdr.len -= (req_len - len);
1505 * Trim from tail. Scan the mbuf chain,
1506 * calculating its length and finding the last mbuf.
1507 * If the adjustment only affects this mbuf, then just
1508 * adjust and return. Otherwise, rescan and truncate
1509 * after the remaining size.
1515 if (m->m_next == NULL)
1519 if (m->m_len >= len) {
1521 if (mp->m_flags & M_PKTHDR)
1522 mp->m_pkthdr.len -= len;
1529 * Correct length for chain is "count".
1530 * Find the mbuf with last data, adjust its length,
1531 * and toss data from remaining mbufs on chain.
1534 if (m->m_flags & M_PKTHDR)
1535 m->m_pkthdr.len = count;
1536 for (; m; m = m->m_next) {
1537 if (m->m_len >= count) {
1544 (m = m->m_next) ->m_len = 0;
1549 * Set the m_data pointer of a newly-allocated mbuf
1550 * to place an object of the specified size at the
1551 * end of the mbuf, longword aligned.
1554 m_align(struct mbuf *m, int len)
1558 if (m->m_flags & M_EXT)
1559 adjust = m->m_ext.ext_size - len;
1560 else if (m->m_flags & M_PKTHDR)
1561 adjust = MHLEN - len;
1563 adjust = MLEN - len;
1564 m->m_data += adjust &~ (sizeof(long)-1);
1568 * Rearrange an mbuf chain so that len bytes are contiguous
1569 * and in the data area of an mbuf (so that mtod will work for a structure
1570 * of size len). Returns the resulting mbuf chain on success, frees it and
1571 * returns null on failure. If there is room, it will add up to
1572 * max_protohdr-len extra bytes to the contiguous region in an attempt to
1573 * avoid being called next time.
1576 m_pullup(struct mbuf *n, int len)
1583 * If first mbuf has no cluster, and has room for len bytes
1584 * without shifting current data, pullup into it,
1585 * otherwise allocate a new mbuf to prepend to the chain.
1587 if (!(n->m_flags & M_EXT) &&
1588 n->m_data + len < &n->m_dat[MLEN] &&
1590 if (n->m_len >= len)
1598 if (n->m_flags & M_PKTHDR)
1599 m = m_gethdr(MB_DONTWAIT, n->m_type);
1601 m = m_get(MB_DONTWAIT, n->m_type);
1605 if (n->m_flags & M_PKTHDR)
1606 M_MOVE_PKTHDR(m, n);
1608 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1610 count = min(min(max(len, max_protohdr), space), n->m_len);
1611 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1621 } while (len > 0 && n);
1630 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1635 * Partition an mbuf chain in two pieces, returning the tail --
1636 * all but the first len0 bytes. In case of failure, it returns NULL and
1637 * attempts to restore the chain to its original state.
1639 * Note that the resulting mbufs might be read-only, because the new
1640 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1641 * the "breaking point" happens to lie within a cluster mbuf. Use the
1642 * M_WRITABLE() macro to check for this case.
1645 m_split(struct mbuf *m0, int len0, int wait)
1648 unsigned len = len0, remain;
1650 for (m = m0; m && len > m->m_len; m = m->m_next)
1654 remain = m->m_len - len;
1655 if (m0->m_flags & M_PKTHDR) {
1656 n = m_gethdr(wait, m0->m_type);
1659 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1660 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1661 m0->m_pkthdr.len = len0;
1662 if (m->m_flags & M_EXT)
1664 if (remain > MHLEN) {
1665 /* m can't be the lead packet */
1667 n->m_next = m_split(m, len, wait);
1668 if (n->m_next == NULL) {
1676 MH_ALIGN(n, remain);
1677 } else if (remain == 0) {
1682 n = m_get(wait, m->m_type);
1688 if (m->m_flags & M_EXT) {
1689 KKASSERT((n->m_flags & M_EXT) == 0);
1690 n->m_data = m->m_data + len;
1691 m->m_ext.ext_ref(m->m_ext.ext_arg);
1692 n->m_ext = m->m_ext;
1693 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
1695 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1699 n->m_next = m->m_next;
1705 * Routine to copy from device local memory into mbufs.
1706 * Note: "offset" is ill-defined and always called as 0, so ignore it.
1709 m_devget(char *buf, int len, int offset, struct ifnet *ifp,
1710 void (*copy)(volatile const void *from, volatile void *to, size_t length))
1712 struct mbuf *m, *mfirst = NULL, **mtail;
1721 m = m_getl(len, MB_DONTWAIT, MT_DATA, flags, &nsize);
1726 m->m_len = min(len, nsize);
1728 if (flags & M_PKTHDR) {
1729 if (len + max_linkhdr <= nsize)
1730 m->m_data += max_linkhdr;
1731 m->m_pkthdr.rcvif = ifp;
1732 m->m_pkthdr.len = len;
1736 copy(buf, m->m_data, (unsigned)m->m_len);
1747 * Routine to pad mbuf to the specified length 'padto'.
1750 m_devpad(struct mbuf *m, int padto)
1752 struct mbuf *last = NULL;
1755 if (padto <= m->m_pkthdr.len)
1758 padlen = padto - m->m_pkthdr.len;
1760 /* if there's only the packet-header and we can pad there, use it. */
1761 if (m->m_pkthdr.len == m->m_len && M_TRAILINGSPACE(m) >= padlen) {
1765 * Walk packet chain to find last mbuf. We will either
1766 * pad there, or append a new mbuf and pad it
1768 for (last = m; last->m_next != NULL; last = last->m_next)
1771 /* `last' now points to last in chain. */
1772 if (M_TRAILINGSPACE(last) < padlen) {
1775 /* Allocate new empty mbuf, pad it. Compact later. */
1776 MGET(n, MB_DONTWAIT, MT_DATA);
1784 KKASSERT(M_TRAILINGSPACE(last) >= padlen);
1785 KKASSERT(M_WRITABLE(last));
1787 /* Now zero the pad area */
1788 bzero(mtod(last, char *) + last->m_len, padlen);
1789 last->m_len += padlen;
1790 m->m_pkthdr.len += padlen;
1795 * Copy data from a buffer back into the indicated mbuf chain,
1796 * starting "off" bytes from the beginning, extending the mbuf
1797 * chain if necessary.
1800 m_copyback(struct mbuf *m0, int off, int len, caddr_t cp)
1803 struct mbuf *m = m0, *n;
1808 while (off > (mlen = m->m_len)) {
1811 if (m->m_next == NULL) {
1812 n = m_getclr(MB_DONTWAIT, m->m_type);
1815 n->m_len = min(MLEN, len + off);
1821 mlen = min (m->m_len - off, len);
1822 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen);
1830 if (m->m_next == NULL) {
1831 n = m_get(MB_DONTWAIT, m->m_type);
1834 n->m_len = min(MLEN, len);
1839 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1840 m->m_pkthdr.len = totlen;
1844 * Append the specified data to the indicated mbuf chain,
1845 * Extend the mbuf chain if the new data does not fit in
1848 * Return 1 if able to complete the job; otherwise 0.
1851 m_append(struct mbuf *m0, int len, c_caddr_t cp)
1854 int remainder, space;
1856 for (m = m0; m->m_next != NULL; m = m->m_next)
1859 space = M_TRAILINGSPACE(m);
1862 * Copy into available space.
1864 if (space > remainder)
1866 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1868 cp += space, remainder -= space;
1870 while (remainder > 0) {
1872 * Allocate a new mbuf; could check space
1873 * and allocate a cluster instead.
1875 n = m_get(MB_DONTWAIT, m->m_type);
1878 n->m_len = min(MLEN, remainder);
1879 bcopy(cp, mtod(n, caddr_t), n->m_len);
1880 cp += n->m_len, remainder -= n->m_len;
1884 if (m0->m_flags & M_PKTHDR)
1885 m0->m_pkthdr.len += len - remainder;
1886 return (remainder == 0);
1890 * Apply function f to the data in an mbuf chain starting "off" bytes from
1891 * the beginning, continuing for "len" bytes.
1894 m_apply(struct mbuf *m, int off, int len,
1895 int (*f)(void *, void *, u_int), void *arg)
1900 KASSERT(off >= 0, ("m_apply, negative off %d", off));
1901 KASSERT(len >= 0, ("m_apply, negative len %d", len));
1903 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1910 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1911 count = min(m->m_len - off, len);
1912 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1923 * Return a pointer to mbuf/offset of location in mbuf chain.
1926 m_getptr(struct mbuf *m, int loc, int *off)
1930 /* Normal end of search. */
1931 if (m->m_len > loc) {
1936 if (m->m_next == NULL) {
1938 /* Point at the end of valid data. */
1951 m_print(const struct mbuf *m)
1954 const struct mbuf *m2;
1956 len = m->m_pkthdr.len;
1959 kprintf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-");
1967 * "Move" mbuf pkthdr from "from" to "to".
1968 * "from" must have M_PKTHDR set, and "to" must be empty.
1971 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
1973 KASSERT((to->m_flags & M_PKTHDR), ("m_move_pkthdr: not packet header"));
1975 to->m_flags |= from->m_flags & M_COPYFLAGS;
1976 to->m_pkthdr = from->m_pkthdr; /* especially tags */
1977 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
1981 * Duplicate "from"'s mbuf pkthdr in "to".
1982 * "from" must have M_PKTHDR set, and "to" must be empty.
1983 * In particular, this does a deep copy of the packet tags.
1986 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how)
1988 KASSERT((to->m_flags & M_PKTHDR), ("m_dup_pkthdr: not packet header"));
1990 to->m_flags = (from->m_flags & M_COPYFLAGS) |
1991 (to->m_flags & ~M_COPYFLAGS);
1992 to->m_pkthdr = from->m_pkthdr;
1993 SLIST_INIT(&to->m_pkthdr.tags);
1994 return (m_tag_copy_chain(to, from, how));
1998 * Defragment a mbuf chain, returning the shortest possible
1999 * chain of mbufs and clusters. If allocation fails and
2000 * this cannot be completed, NULL will be returned, but
2001 * the passed in chain will be unchanged. Upon success,
2002 * the original chain will be freed, and the new chain
2005 * If a non-packet header is passed in, the original
2006 * mbuf (chain?) will be returned unharmed.
2008 * m_defrag_nofree doesn't free the passed in mbuf.
2011 m_defrag(struct mbuf *m0, int how)
2015 if ((m_new = m_defrag_nofree(m0, how)) == NULL)
2023 m_defrag_nofree(struct mbuf *m0, int how)
2025 struct mbuf *m_new = NULL, *m_final = NULL;
2026 int progress = 0, length, nsize;
2028 if (!(m0->m_flags & M_PKTHDR))
2031 #ifdef MBUF_STRESS_TEST
2032 if (m_defragrandomfailures) {
2033 int temp = karc4random() & 0xff;
2039 m_final = m_getl(m0->m_pkthdr.len, how, MT_DATA, M_PKTHDR, &nsize);
2040 if (m_final == NULL)
2042 m_final->m_len = 0; /* in case m0->m_pkthdr.len is zero */
2044 if (m_dup_pkthdr(m_final, m0, how) == 0)
2049 while (progress < m0->m_pkthdr.len) {
2050 length = m0->m_pkthdr.len - progress;
2051 if (length > MCLBYTES)
2054 if (m_new == NULL) {
2055 m_new = m_getl(length, how, MT_DATA, 0, &nsize);
2060 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
2062 m_new->m_len = length;
2063 if (m_new != m_final)
2064 m_cat(m_final, m_new);
2067 if (m0->m_next == NULL)
2070 m_defragbytes += m_final->m_pkthdr.len;
2081 * Move data from uio into mbufs.
2084 m_uiomove(struct uio *uio)
2086 struct mbuf *m; /* current working mbuf */
2087 struct mbuf *head = NULL; /* result mbuf chain */
2088 struct mbuf **mp = &head;
2089 int flags = M_PKTHDR;
2095 if (uio->uio_resid > INT_MAX)
2098 resid = (int)uio->uio_resid;
2099 m = m_getl(resid, MB_WAIT, MT_DATA, flags, &nsize);
2101 m->m_pkthdr.len = 0;
2102 /* Leave room for protocol headers. */
2107 m->m_len = imin(nsize, resid);
2108 error = uiomove(mtod(m, caddr_t), m->m_len, uio);
2115 head->m_pkthdr.len += m->m_len;
2116 } while (uio->uio_resid > 0);
2126 m_last(struct mbuf *m)
2134 * Return the number of bytes in an mbuf chain.
2135 * If lastm is not NULL, also return the last mbuf.
2138 m_lengthm(struct mbuf *m, struct mbuf **lastm)
2141 struct mbuf *prev = m;
2154 * Like m_lengthm(), except also keep track of mbuf usage.
2157 m_countm(struct mbuf *m, struct mbuf **lastm, u_int *pmbcnt)
2159 u_int len = 0, mbcnt = 0;
2160 struct mbuf *prev = m;
2165 if (m->m_flags & M_EXT)
2166 mbcnt += m->m_ext.ext_size;