2 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 1982, 1986, 1988, 1991, 1993
36 * The Regents of the University of California. All rights reserved.
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94
67 * $FreeBSD: src/sys/kern/uipc_mbuf.c,v 1.51.2.24 2003/04/15 06:59:29 silby Exp $
68 * $DragonFly: src/sys/kern/uipc_mbuf.c,v 1.70 2008/11/20 14:21:01 sephe Exp $
71 #include "opt_param.h"
72 #include "opt_mbuf_stress_test.h"
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/malloc.h>
77 #include <sys/kernel.h>
78 #include <sys/sysctl.h>
79 #include <sys/domain.h>
80 #include <sys/objcache.h>
82 #include <sys/protosw.h>
84 #include <sys/thread.h>
85 #include <sys/globaldata.h>
86 #include <sys/thread2.h>
88 #include <machine/atomic.h>
91 #include <vm/vm_kern.h>
92 #include <vm/vm_extern.h>
95 #include <machine/cpu.h>
99 * mbuf cluster meta-data
107 * mbuf tracking for debugging purposes
111 static MALLOC_DEFINE(M_MTRACK
, "mtrack", "mtrack");
114 RB_HEAD(mbuf_rb_tree
, mbtrack
);
115 RB_PROTOTYPE2(mbuf_rb_tree
, mbtrack
, rb_node
, mbtrack_cmp
, struct mbuf
*);
118 RB_ENTRY(mbtrack
) rb_node
;
124 mbtrack_cmp(struct mbtrack
*mb1
, struct mbtrack
*mb2
)
133 RB_GENERATE2(mbuf_rb_tree
, mbtrack
, rb_node
, mbtrack_cmp
, struct mbuf
*, m
);
135 struct mbuf_rb_tree mbuf_track_root
;
138 mbuftrack(struct mbuf
*m
)
143 mbt
= kmalloc(sizeof(*mbt
), M_MTRACK
, M_INTWAIT
|M_ZERO
);
145 if (mbuf_rb_tree_RB_INSERT(&mbuf_track_root
, mbt
))
146 panic("mbuftrack: mbuf %p already being tracked\n", m
);
151 mbufuntrack(struct mbuf
*m
)
156 mbt
= mbuf_rb_tree_RB_LOOKUP(&mbuf_track_root
, m
);
158 kprintf("mbufuntrack: mbuf %p was not tracked\n", m
);
160 mbuf_rb_tree_RB_REMOVE(&mbuf_track_root
, mbt
);
161 kfree(mbt
, M_MTRACK
);
167 mbuftrackid(struct mbuf
*m
, int trackid
)
176 mbt
= mbuf_rb_tree_RB_LOOKUP(&mbuf_track_root
, m
);
178 mbt
->trackid
= trackid
;
187 mbuftrack_callback(struct mbtrack
*mbt
, void *arg
)
189 struct sysctl_req
*req
= arg
;
193 ksnprintf(buf
, sizeof(buf
), "mbuf %p track %d\n", mbt
->m
, mbt
->trackid
);
195 error
= SYSCTL_OUT(req
, buf
, strlen(buf
));
202 mbuftrack_show(SYSCTL_HANDLER_ARGS
)
207 error
= mbuf_rb_tree_RB_SCAN(&mbuf_track_root
, NULL
,
208 mbuftrack_callback
, req
);
212 SYSCTL_PROC(_kern_ipc
, OID_AUTO
, showmbufs
, CTLFLAG_RD
|CTLTYPE_STRING
,
213 0, 0, mbuftrack_show
, "A", "Show all in-use mbufs");
218 #define mbufuntrack(m)
222 static void mbinit(void *);
223 SYSINIT(mbuf
, SI_BOOT2_MACHDEP
, SI_ORDER_FIRST
, mbinit
, NULL
)
225 static u_long mbtypes
[SMP_MAXCPU
][MT_NTYPES
];
227 static struct mbstat mbstat
[SMP_MAXCPU
];
236 #ifdef MBUF_STRESS_TEST
237 int m_defragrandomfailures
;
240 struct objcache
*mbuf_cache
, *mbufphdr_cache
;
241 struct objcache
*mclmeta_cache
;
242 struct objcache
*mbufcluster_cache
, *mbufphdrcluster_cache
;
247 SYSCTL_INT(_kern_ipc
, KIPC_MAX_LINKHDR
, max_linkhdr
, CTLFLAG_RW
,
248 &max_linkhdr
, 0, "");
249 SYSCTL_INT(_kern_ipc
, KIPC_MAX_PROTOHDR
, max_protohdr
, CTLFLAG_RW
,
250 &max_protohdr
, 0, "");
251 SYSCTL_INT(_kern_ipc
, KIPC_MAX_HDR
, max_hdr
, CTLFLAG_RW
, &max_hdr
, 0, "");
252 SYSCTL_INT(_kern_ipc
, KIPC_MAX_DATALEN
, max_datalen
, CTLFLAG_RW
,
253 &max_datalen
, 0, "");
254 SYSCTL_INT(_kern_ipc
, OID_AUTO
, mbuf_wait
, CTLFLAG_RW
,
256 static int do_mbstat(SYSCTL_HANDLER_ARGS
);
258 SYSCTL_PROC(_kern_ipc
, KIPC_MBSTAT
, mbstat
, CTLTYPE_STRUCT
|CTLFLAG_RD
,
259 0, 0, do_mbstat
, "S,mbstat", "");
261 static int do_mbtypes(SYSCTL_HANDLER_ARGS
);
263 SYSCTL_PROC(_kern_ipc
, OID_AUTO
, mbtypes
, CTLTYPE_ULONG
|CTLFLAG_RD
,
264 0, 0, do_mbtypes
, "LU", "");
267 do_mbstat(SYSCTL_HANDLER_ARGS
)
269 struct mbstat mbstat_total
;
270 struct mbstat
*mbstat_totalp
;
273 bzero(&mbstat_total
, sizeof(mbstat_total
));
274 mbstat_totalp
= &mbstat_total
;
276 for (i
= 0; i
< ncpus
; i
++)
278 mbstat_total
.m_mbufs
+= mbstat
[i
].m_mbufs
;
279 mbstat_total
.m_clusters
+= mbstat
[i
].m_clusters
;
280 mbstat_total
.m_spare
+= mbstat
[i
].m_spare
;
281 mbstat_total
.m_clfree
+= mbstat
[i
].m_clfree
;
282 mbstat_total
.m_drops
+= mbstat
[i
].m_drops
;
283 mbstat_total
.m_wait
+= mbstat
[i
].m_wait
;
284 mbstat_total
.m_drain
+= mbstat
[i
].m_drain
;
285 mbstat_total
.m_mcfail
+= mbstat
[i
].m_mcfail
;
286 mbstat_total
.m_mpfail
+= mbstat
[i
].m_mpfail
;
290 * The following fields are not cumulative fields so just
291 * get their values once.
293 mbstat_total
.m_msize
= mbstat
[0].m_msize
;
294 mbstat_total
.m_mclbytes
= mbstat
[0].m_mclbytes
;
295 mbstat_total
.m_minclsize
= mbstat
[0].m_minclsize
;
296 mbstat_total
.m_mlen
= mbstat
[0].m_mlen
;
297 mbstat_total
.m_mhlen
= mbstat
[0].m_mhlen
;
299 return(sysctl_handle_opaque(oidp
, mbstat_totalp
, sizeof(mbstat_total
), req
));
303 do_mbtypes(SYSCTL_HANDLER_ARGS
)
305 u_long totals
[MT_NTYPES
];
308 for (i
= 0; i
< MT_NTYPES
; i
++)
311 for (i
= 0; i
< ncpus
; i
++)
313 for (j
= 0; j
< MT_NTYPES
; j
++)
314 totals
[j
] += mbtypes
[i
][j
];
317 return(sysctl_handle_opaque(oidp
, totals
, sizeof(totals
), req
));
321 * These are read-only because we do not currently have any code
322 * to adjust the objcache limits after the fact. The variables
323 * may only be set as boot-time tunables.
325 SYSCTL_INT(_kern_ipc
, KIPC_NMBCLUSTERS
, nmbclusters
, CTLFLAG_RD
,
326 &nmbclusters
, 0, "Maximum number of mbuf clusters available");
327 SYSCTL_INT(_kern_ipc
, OID_AUTO
, nmbufs
, CTLFLAG_RD
, &nmbufs
, 0,
328 "Maximum number of mbufs available");
330 SYSCTL_INT(_kern_ipc
, OID_AUTO
, m_defragpackets
, CTLFLAG_RD
,
331 &m_defragpackets
, 0, "");
332 SYSCTL_INT(_kern_ipc
, OID_AUTO
, m_defragbytes
, CTLFLAG_RD
,
333 &m_defragbytes
, 0, "");
334 SYSCTL_INT(_kern_ipc
, OID_AUTO
, m_defraguseless
, CTLFLAG_RD
,
335 &m_defraguseless
, 0, "");
336 SYSCTL_INT(_kern_ipc
, OID_AUTO
, m_defragfailure
, CTLFLAG_RD
,
337 &m_defragfailure
, 0, "");
338 #ifdef MBUF_STRESS_TEST
339 SYSCTL_INT(_kern_ipc
, OID_AUTO
, m_defragrandomfailures
, CTLFLAG_RW
,
340 &m_defragrandomfailures
, 0, "");
343 static MALLOC_DEFINE(M_MBUF
, "mbuf", "mbuf");
344 static MALLOC_DEFINE(M_MBUFCL
, "mbufcl", "mbufcl");
345 static MALLOC_DEFINE(M_MCLMETA
, "mclmeta", "mclmeta");
347 static void m_reclaim (void);
348 static void m_mclref(void *arg
);
349 static void m_mclfree(void *arg
);
352 #define NMBCLUSTERS (512 + maxusers * 16)
355 #define NMBUFS (nmbclusters * 2)
359 * Perform sanity checks of tunables declared above.
362 tunable_mbinit(void *dummy
)
365 * This has to be done before VM init.
367 nmbclusters
= NMBCLUSTERS
;
368 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters
);
370 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs
);
372 if (nmbufs
< nmbclusters
* 2)
373 nmbufs
= nmbclusters
* 2;
375 SYSINIT(tunable_mbinit
, SI_BOOT1_TUNABLES
, SI_ORDER_ANY
,
376 tunable_mbinit
, NULL
);
378 /* "number of clusters of pages" */
384 * The mbuf object cache only guarantees that m_next and m_nextpkt are
385 * NULL and that m_data points to the beginning of the data area. In
386 * particular, m_len and m_pkthdr.len are uninitialized. It is the
387 * responsibility of the caller to initialize those fields before use.
390 static boolean_t __inline
391 mbuf_ctor(void *obj
, void *private, int ocflags
)
393 struct mbuf
*m
= obj
;
397 m
->m_data
= m
->m_dat
;
404 * Initialize the mbuf and the packet header fields.
407 mbufphdr_ctor(void *obj
, void *private, int ocflags
)
409 struct mbuf
*m
= obj
;
413 m
->m_data
= m
->m_pktdat
;
414 m
->m_flags
= M_PKTHDR
| M_PHCACHE
;
416 m
->m_pkthdr
.rcvif
= NULL
; /* eliminate XXX JH */
417 SLIST_INIT(&m
->m_pkthdr
.tags
);
418 m
->m_pkthdr
.csum_flags
= 0; /* eliminate XXX JH */
419 m
->m_pkthdr
.fw_flags
= 0; /* eliminate XXX JH */
425 * A mbcluster object consists of 2K (MCLBYTES) cluster and a refcount.
428 mclmeta_ctor(void *obj
, void *private, int ocflags
)
430 struct mbcluster
*cl
= obj
;
433 if (ocflags
& M_NOWAIT
)
434 buf
= kmalloc(MCLBYTES
, M_MBUFCL
, M_NOWAIT
| M_ZERO
);
436 buf
= kmalloc(MCLBYTES
, M_MBUFCL
, M_INTWAIT
| M_ZERO
);
445 mclmeta_dtor(void *obj
, void *private)
447 struct mbcluster
*mcl
= obj
;
449 KKASSERT(mcl
->mcl_refs
== 0);
450 kfree(mcl
->mcl_data
, M_MBUFCL
);
454 linkcluster(struct mbuf
*m
, struct mbcluster
*cl
)
457 * Add the cluster to the mbuf. The caller will detect that the
458 * mbuf now has an attached cluster.
460 m
->m_ext
.ext_arg
= cl
;
461 m
->m_ext
.ext_buf
= cl
->mcl_data
;
462 m
->m_ext
.ext_ref
= m_mclref
;
463 m
->m_ext
.ext_free
= m_mclfree
;
464 m
->m_ext
.ext_size
= MCLBYTES
;
465 atomic_add_int(&cl
->mcl_refs
, 1);
467 m
->m_data
= m
->m_ext
.ext_buf
;
468 m
->m_flags
|= M_EXT
| M_EXT_CLUSTER
;
472 mbufphdrcluster_ctor(void *obj
, void *private, int ocflags
)
474 struct mbuf
*m
= obj
;
475 struct mbcluster
*cl
;
477 mbufphdr_ctor(obj
, private, ocflags
);
478 cl
= objcache_get(mclmeta_cache
, ocflags
);
481 m
->m_flags
|= M_CLCACHE
;
487 mbufcluster_ctor(void *obj
, void *private, int ocflags
)
489 struct mbuf
*m
= obj
;
490 struct mbcluster
*cl
;
492 mbuf_ctor(obj
, private, ocflags
);
493 cl
= objcache_get(mclmeta_cache
, ocflags
);
496 m
->m_flags
|= M_CLCACHE
;
502 * Used for both the cluster and cluster PHDR caches.
504 * The mbuf may have lost its cluster due to sharing, deal
505 * with the situation by checking M_EXT.
508 mbufcluster_dtor(void *obj
, void *private)
510 struct mbuf
*m
= obj
;
511 struct mbcluster
*mcl
;
513 if (m
->m_flags
& M_EXT
) {
514 KKASSERT((m
->m_flags
& M_EXT_CLUSTER
) != 0);
515 mcl
= m
->m_ext
.ext_arg
;
516 KKASSERT(mcl
->mcl_refs
== 1);
518 objcache_put(mclmeta_cache
, mcl
);
522 struct objcache_malloc_args mbuf_malloc_args
= { MSIZE
, M_MBUF
};
523 struct objcache_malloc_args mclmeta_malloc_args
=
524 { sizeof(struct mbcluster
), M_MCLMETA
};
530 int mb_limit
, cl_limit
, mbcl_limit
;
535 * Initialize statistics
537 for (i
= 0; i
< ncpus
; i
++) {
538 atomic_set_long_nonlocked(&mbstat
[i
].m_msize
, MSIZE
);
539 atomic_set_long_nonlocked(&mbstat
[i
].m_mclbytes
, MCLBYTES
);
540 atomic_set_long_nonlocked(&mbstat
[i
].m_minclsize
, MINCLSIZE
);
541 atomic_set_long_nonlocked(&mbstat
[i
].m_mlen
, MLEN
);
542 atomic_set_long_nonlocked(&mbstat
[i
].m_mhlen
, MHLEN
);
546 * Create objtect caches and save cluster limits, which will
547 * be used to adjust backing kmalloc pools' limit later.
550 mb_limit
= cl_limit
= mbcl_limit
= 0;
553 mbuf_cache
= objcache_create("mbuf", &limit
, 0,
554 mbuf_ctor
, NULL
, NULL
,
555 objcache_malloc_alloc
, objcache_malloc_free
, &mbuf_malloc_args
);
556 if (limit
> mb_limit
)
560 mbufphdr_cache
= objcache_create("mbuf pkt hdr", &limit
, 64,
561 mbufphdr_ctor
, NULL
, NULL
,
562 objcache_malloc_alloc
, objcache_malloc_free
, &mbuf_malloc_args
);
563 if (limit
> mb_limit
)
566 cl_limit
= nmbclusters
;
567 mclmeta_cache
= objcache_create("cluster mbuf", &cl_limit
, 0,
568 mclmeta_ctor
, mclmeta_dtor
, NULL
,
569 objcache_malloc_alloc
, objcache_malloc_free
, &mclmeta_malloc_args
);
572 mbufcluster_cache
= objcache_create("mbuf + cluster", &limit
, 0,
573 mbufcluster_ctor
, mbufcluster_dtor
, NULL
,
574 objcache_malloc_alloc
, objcache_malloc_free
, &mbuf_malloc_args
);
575 if (limit
> mbcl_limit
)
579 mbufphdrcluster_cache
= objcache_create("mbuf pkt hdr + cluster",
580 &limit
, 64, mbufphdrcluster_ctor
, mbufcluster_dtor
, NULL
,
581 objcache_malloc_alloc
, objcache_malloc_free
, &mbuf_malloc_args
);
582 if (limit
> mbcl_limit
)
586 * Adjust backing kmalloc pools' limit
588 * NOTE: We raise the limit by another 1/8 to take the effect
589 * of loosememuse into account.
591 cl_limit
+= cl_limit
/ 8;
592 kmalloc_raise_limit(mclmeta_malloc_args
.mtype
,
593 mclmeta_malloc_args
.objsize
* cl_limit
);
594 kmalloc_raise_limit(M_MBUFCL
, MCLBYTES
* cl_limit
);
596 mb_limit
+= mbcl_limit
;
597 mb_limit
+= mb_limit
/ 4; /* save some space for non-pkthdr mbufs */
598 mb_limit
+= mb_limit
/ 8;
599 kmalloc_raise_limit(mbuf_malloc_args
.mtype
,
600 mbuf_malloc_args
.objsize
* mb_limit
);
604 * Return the number of references to this mbuf's data. 0 is returned
605 * if the mbuf is not M_EXT, a reference count is returned if it is
606 * M_EXT | M_EXT_CLUSTER, and 99 is returned if it is a special M_EXT.
609 m_sharecount(struct mbuf
*m
)
611 switch (m
->m_flags
& (M_EXT
| M_EXT_CLUSTER
)) {
616 case M_EXT
| M_EXT_CLUSTER
:
617 return (((struct mbcluster
*)m
->m_ext
.ext_arg
)->mcl_refs
);
620 return (0); /* to shut up compiler */
624 * change mbuf to new type
627 m_chtype(struct mbuf
*m
, int type
)
629 struct globaldata
*gd
= mycpu
;
631 atomic_add_long_nonlocked(&mbtypes
[gd
->gd_cpuid
][type
], 1);
632 atomic_subtract_long_nonlocked(&mbtypes
[gd
->gd_cpuid
][m
->m_type
], 1);
633 atomic_set_short_nonlocked(&m
->m_type
, type
);
643 SLIST_FOREACH(dp
, &domains
, dom_next
) {
644 for (pr
= dp
->dom_protosw
; pr
< dp
->dom_protoswNPROTOSW
; pr
++) {
650 atomic_add_long_nonlocked(&mbstat
[mycpu
->gd_cpuid
].m_drain
, 1);
654 updatestats(struct mbuf
*m
, int type
)
656 struct globaldata
*gd
= mycpu
;
661 atomic_add_long_nonlocked(&mbtypes
[gd
->gd_cpuid
][type
], 1);
662 atomic_add_long_nonlocked(&mbstat
[mycpu
->gd_cpuid
].m_mbufs
, 1);
670 m_get(int how
, int type
)
674 int ocf
= MBTOM(how
);
678 m
= objcache_get(mbuf_cache
, ocf
);
681 if ((how
& MB_TRYWAIT
) && ntries
++ == 0) {
682 struct objcache
*reclaimlist
[] = {
684 mbufcluster_cache
, mbufphdrcluster_cache
686 const int nreclaims
= __arysize(reclaimlist
);
688 if (!objcache_reclaimlist(reclaimlist
, nreclaims
, ocf
))
695 updatestats(m
, type
);
700 m_gethdr(int how
, int type
)
703 int ocf
= MBTOM(how
);
708 m
= objcache_get(mbufphdr_cache
, ocf
);
711 if ((how
& MB_TRYWAIT
) && ntries
++ == 0) {
712 struct objcache
*reclaimlist
[] = {
714 mbufcluster_cache
, mbufphdrcluster_cache
716 const int nreclaims
= __arysize(reclaimlist
);
718 if (!objcache_reclaimlist(reclaimlist
, nreclaims
, ocf
))
725 updatestats(m
, type
);
730 * Get a mbuf (not a mbuf cluster!) and zero it.
734 m_getclr(int how
, int type
)
738 m
= m_get(how
, type
);
740 bzero(m
->m_data
, MLEN
);
745 * Returns an mbuf with an attached cluster.
746 * Because many network drivers use this kind of buffers a lot, it is
747 * convenient to keep a small pool of free buffers of this kind.
748 * Even a small size such as 10 gives about 10% improvement in the
749 * forwarding rate in a bridge or router.
752 m_getcl(int how
, short type
, int flags
)
755 int ocflags
= MBTOM(how
);
760 if (flags
& M_PKTHDR
)
761 m
= objcache_get(mbufphdrcluster_cache
, ocflags
);
763 m
= objcache_get(mbufcluster_cache
, ocflags
);
766 if ((how
& MB_TRYWAIT
) && ntries
++ == 0) {
767 struct objcache
*reclaimlist
[1];
769 if (flags
& M_PKTHDR
)
770 reclaimlist
[0] = mbufcluster_cache
;
772 reclaimlist
[0] = mbufphdrcluster_cache
;
773 if (!objcache_reclaimlist(reclaimlist
, 1, ocflags
))
784 atomic_add_long_nonlocked(&mbtypes
[mycpu
->gd_cpuid
][type
], 1);
785 atomic_add_long_nonlocked(&mbstat
[mycpu
->gd_cpuid
].m_clusters
, 1);
790 * Allocate chain of requested length.
793 m_getc(int len
, int how
, int type
)
795 struct mbuf
*n
, *nfirst
= NULL
, **ntail
= &nfirst
;
799 n
= m_getl(len
, how
, type
, 0, &nsize
);
815 * Allocate len-worth of mbufs and/or mbuf clusters (whatever fits best)
816 * and return a pointer to the head of the allocated chain. If m0 is
817 * non-null, then we assume that it is a single mbuf or an mbuf chain to
818 * which we want len bytes worth of mbufs and/or clusters attached, and so
819 * if we succeed in allocating it, we will just return a pointer to m0.
821 * If we happen to fail at any point during the allocation, we will free
822 * up everything we have already allocated and return NULL.
824 * Deprecated. Use m_getc() and m_cat() instead.
827 m_getm(struct mbuf
*m0
, int len
, int type
, int how
)
831 nfirst
= m_getc(len
, how
, type
);
834 m_last(m0
)->m_next
= nfirst
;
842 * Adds a cluster to a normal mbuf, M_EXT is set on success.
843 * Deprecated. Use m_getcl() instead.
846 m_mclget(struct mbuf
*m
, int how
)
848 struct mbcluster
*mcl
;
850 KKASSERT((m
->m_flags
& M_EXT
) == 0);
851 mcl
= objcache_get(mclmeta_cache
, MBTOM(how
));
854 atomic_add_long_nonlocked(&mbstat
[mycpu
->gd_cpuid
].m_clusters
, 1);
859 * Updates to mbcluster must be MPSAFE. Only an entity which already has
860 * a reference to the cluster can ref it, so we are in no danger of
861 * racing an add with a subtract. But the operation must still be atomic
862 * since multiple entities may have a reference on the cluster.
864 * m_mclfree() is almost the same but it must contend with two entities
865 * freeing the cluster at the same time. If there is only one reference
866 * count we are the only entity referencing the cluster and no further
867 * locking is required. Otherwise we must protect against a race to 0
868 * with the serializer.
873 struct mbcluster
*mcl
= arg
;
875 atomic_add_int(&mcl
->mcl_refs
, 1);
879 * When dereferencing a cluster we have to deal with a N->0 race, where
880 * N entities free their references simultaniously. To do this we use
881 * atomic_fetchadd_int().
886 struct mbcluster
*mcl
= arg
;
888 if (atomic_fetchadd_int(&mcl
->mcl_refs
, -1) == 1)
889 objcache_put(mclmeta_cache
, mcl
);
893 * Free a single mbuf and any associated external storage. The successor,
894 * if any, is returned.
896 * We do need to check non-first mbuf for m_aux, since some of existing
897 * code does not call M_PREPEND properly.
898 * (example: call to bpf_mtap from drivers)
901 m_free(struct mbuf
*m
)
904 struct globaldata
*gd
= mycpu
;
906 KASSERT(m
->m_type
!= MT_FREE
, ("freeing free mbuf %p", m
));
907 atomic_subtract_long_nonlocked(&mbtypes
[gd
->gd_cpuid
][m
->m_type
], 1);
912 * Make sure the mbuf is in constructed state before returning it
918 KKASSERT(m
->m_nextpkt
== NULL
);
920 if (m
->m_nextpkt
!= NULL
) {
921 static int afewtimes
= 10;
923 if (afewtimes
-- > 0) {
924 kprintf("mfree: m->m_nextpkt != NULL\n");
930 if (m
->m_flags
& M_PKTHDR
) {
931 m_tag_delete_chain(m
); /* eliminate XXX JH */
934 m
->m_flags
&= (M_EXT
| M_EXT_CLUSTER
| M_CLCACHE
| M_PHCACHE
);
937 * Clean the M_PKTHDR state so we can return the mbuf to its original
938 * cache. This is based on the PHCACHE flag which tells us whether
939 * the mbuf was originally allocated out of a packet-header cache
940 * or a non-packet-header cache.
942 if (m
->m_flags
& M_PHCACHE
) {
943 m
->m_flags
|= M_PKTHDR
;
944 m
->m_pkthdr
.rcvif
= NULL
; /* eliminate XXX JH */
945 m
->m_pkthdr
.csum_flags
= 0; /* eliminate XXX JH */
946 m
->m_pkthdr
.fw_flags
= 0; /* eliminate XXX JH */
947 SLIST_INIT(&m
->m_pkthdr
.tags
);
951 * Handle remaining flags combinations. M_CLCACHE tells us whether
952 * the mbuf was originally allocated from a cluster cache or not,
953 * and is totally separate from whether the mbuf is currently
954 * associated with a cluster.
957 switch(m
->m_flags
& (M_CLCACHE
| M_EXT
| M_EXT_CLUSTER
)) {
958 case M_CLCACHE
| M_EXT
| M_EXT_CLUSTER
:
960 * mbuf+cluster cache case. The mbuf was allocated from the
961 * combined mbuf_cluster cache and can be returned to the
962 * cache if the cluster hasn't been shared.
964 if (m_sharecount(m
) == 1) {
966 * The cluster has not been shared, we can just
967 * reset the data pointer and return the mbuf
968 * to the cluster cache. Note that the reference
969 * count is left intact (it is still associated with
972 m
->m_data
= m
->m_ext
.ext_buf
;
973 if (m
->m_flags
& M_PHCACHE
)
974 objcache_put(mbufphdrcluster_cache
, m
);
976 objcache_put(mbufcluster_cache
, m
);
977 atomic_subtract_long_nonlocked(&mbstat
[mycpu
->gd_cpuid
].m_clusters
, 1);
980 * Hell. Someone else has a ref on this cluster,
981 * we have to disconnect it which means we can't
982 * put it back into the mbufcluster_cache, we
983 * have to destroy the mbuf.
985 * Other mbuf references to the cluster will typically
986 * be M_EXT | M_EXT_CLUSTER but without M_CLCACHE.
988 * XXX we could try to connect another cluster to
991 m
->m_ext
.ext_free(m
->m_ext
.ext_arg
);
992 m
->m_flags
&= ~(M_EXT
| M_EXT_CLUSTER
);
993 if (m
->m_flags
& M_PHCACHE
)
994 objcache_dtor(mbufphdrcluster_cache
, m
);
996 objcache_dtor(mbufcluster_cache
, m
);
999 case M_EXT
| M_EXT_CLUSTER
:
1001 * Normal cluster associated with an mbuf that was allocated
1002 * from the normal mbuf pool rather then the cluster pool.
1003 * The cluster has to be independantly disassociated from the
1006 if (m_sharecount(m
) == 1)
1007 atomic_subtract_long_nonlocked(&mbstat
[mycpu
->gd_cpuid
].m_clusters
, 1);
1011 * Normal cluster association case, disconnect the cluster from
1012 * the mbuf. The cluster may or may not be custom.
1014 m
->m_ext
.ext_free(m
->m_ext
.ext_arg
);
1015 m
->m_flags
&= ~(M_EXT
| M_EXT_CLUSTER
);
1019 * return the mbuf to the mbuf cache.
1021 if (m
->m_flags
& M_PHCACHE
) {
1022 m
->m_data
= m
->m_pktdat
;
1023 objcache_put(mbufphdr_cache
, m
);
1025 m
->m_data
= m
->m_dat
;
1026 objcache_put(mbuf_cache
, m
);
1028 atomic_subtract_long_nonlocked(&mbstat
[mycpu
->gd_cpuid
].m_mbufs
, 1);
1032 panic("bad mbuf flags %p %08x\n", m
, m
->m_flags
);
1040 m_freem(struct mbuf
*m
)
1049 * mbuf utility routines
1053 * Lesser-used path for M_PREPEND: allocate new mbuf to prepend to chain and
1057 m_prepend(struct mbuf
*m
, int len
, int how
)
1061 if (m
->m_flags
& M_PKTHDR
)
1062 mn
= m_gethdr(how
, m
->m_type
);
1064 mn
= m_get(how
, m
->m_type
);
1069 if (m
->m_flags
& M_PKTHDR
)
1070 M_MOVE_PKTHDR(mn
, m
);
1080 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
1081 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
1082 * The wait parameter is a choice of MB_WAIT/MB_DONTWAIT from caller.
1083 * Note that the copy is read-only, because clusters are not copied,
1084 * only their reference counts are incremented.
1087 m_copym(const struct mbuf
*m
, int off0
, int len
, int wait
)
1089 struct mbuf
*n
, **np
;
1094 KASSERT(off
>= 0, ("m_copym, negative off %d", off
));
1095 KASSERT(len
>= 0, ("m_copym, negative len %d", len
));
1096 if (off
== 0 && m
->m_flags
& M_PKTHDR
)
1099 KASSERT(m
!= NULL
, ("m_copym, offset > size of mbuf chain"));
1109 KASSERT(len
== M_COPYALL
,
1110 ("m_copym, length > size of mbuf chain"));
1114 * Because we are sharing any cluster attachment below,
1115 * be sure to get an mbuf that does not have a cluster
1116 * associated with it.
1119 n
= m_gethdr(wait
, m
->m_type
);
1121 n
= m_get(wait
, m
->m_type
);
1126 if (!m_dup_pkthdr(n
, m
, wait
))
1128 if (len
== M_COPYALL
)
1129 n
->m_pkthdr
.len
-= off0
;
1131 n
->m_pkthdr
.len
= len
;
1134 n
->m_len
= min(len
, m
->m_len
- off
);
1135 if (m
->m_flags
& M_EXT
) {
1136 KKASSERT((n
->m_flags
& M_EXT
) == 0);
1137 n
->m_data
= m
->m_data
+ off
;
1138 m
->m_ext
.ext_ref(m
->m_ext
.ext_arg
);
1139 n
->m_ext
= m
->m_ext
;
1140 n
->m_flags
|= m
->m_flags
& (M_EXT
| M_EXT_CLUSTER
);
1142 bcopy(mtod(m
, caddr_t
)+off
, mtod(n
, caddr_t
),
1143 (unsigned)n
->m_len
);
1145 if (len
!= M_COPYALL
)
1152 atomic_add_long_nonlocked(&mbstat
[mycpu
->gd_cpuid
].m_mcfail
, 1);
1156 atomic_add_long_nonlocked(&mbstat
[mycpu
->gd_cpuid
].m_mcfail
, 1);
1161 * Copy an entire packet, including header (which must be present).
1162 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
1163 * Note that the copy is read-only, because clusters are not copied,
1164 * only their reference counts are incremented.
1165 * Preserve alignment of the first mbuf so if the creator has left
1166 * some room at the beginning (e.g. for inserting protocol headers)
1167 * the copies also have the room available.
1170 m_copypacket(struct mbuf
*m
, int how
)
1172 struct mbuf
*top
, *n
, *o
;
1174 n
= m_gethdr(how
, m
->m_type
);
1179 if (!m_dup_pkthdr(n
, m
, how
))
1181 n
->m_len
= m
->m_len
;
1182 if (m
->m_flags
& M_EXT
) {
1183 KKASSERT((n
->m_flags
& M_EXT
) == 0);
1184 n
->m_data
= m
->m_data
;
1185 m
->m_ext
.ext_ref(m
->m_ext
.ext_arg
);
1186 n
->m_ext
= m
->m_ext
;
1187 n
->m_flags
|= m
->m_flags
& (M_EXT
| M_EXT_CLUSTER
);
1189 n
->m_data
= n
->m_pktdat
+ (m
->m_data
- m
->m_pktdat
);
1190 bcopy(mtod(m
, char *), mtod(n
, char *), n
->m_len
);
1195 o
= m_get(how
, m
->m_type
);
1202 n
->m_len
= m
->m_len
;
1203 if (m
->m_flags
& M_EXT
) {
1204 KKASSERT((n
->m_flags
& M_EXT
) == 0);
1205 n
->m_data
= m
->m_data
;
1206 m
->m_ext
.ext_ref(m
->m_ext
.ext_arg
);
1207 n
->m_ext
= m
->m_ext
;
1208 n
->m_flags
|= m
->m_flags
& (M_EXT
| M_EXT_CLUSTER
);
1210 bcopy(mtod(m
, char *), mtod(n
, char *), n
->m_len
);
1218 atomic_add_long_nonlocked(&mbstat
[mycpu
->gd_cpuid
].m_mcfail
, 1);
1223 * Copy data from an mbuf chain starting "off" bytes from the beginning,
1224 * continuing for "len" bytes, into the indicated buffer.
1227 m_copydata(const struct mbuf
*m
, int off
, int len
, caddr_t cp
)
1231 KASSERT(off
>= 0, ("m_copydata, negative off %d", off
));
1232 KASSERT(len
>= 0, ("m_copydata, negative len %d", len
));
1234 KASSERT(m
!= NULL
, ("m_copydata, offset > size of mbuf chain"));
1241 KASSERT(m
!= NULL
, ("m_copydata, length > size of mbuf chain"));
1242 count
= min(m
->m_len
- off
, len
);
1243 bcopy(mtod(m
, caddr_t
) + off
, cp
, count
);
1252 * Copy a packet header mbuf chain into a completely new chain, including
1253 * copying any mbuf clusters. Use this instead of m_copypacket() when
1254 * you need a writable copy of an mbuf chain.
1257 m_dup(struct mbuf
*m
, int how
)
1259 struct mbuf
**p
, *top
= NULL
;
1260 int remain
, moff
, nsize
;
1265 KASSERT((m
->m_flags
& M_PKTHDR
) != 0, ("%s: !PKTHDR", __func__
));
1267 /* While there's more data, get a new mbuf, tack it on, and fill it */
1268 remain
= m
->m_pkthdr
.len
;
1271 while (remain
> 0 || top
== NULL
) { /* allow m->m_pkthdr.len == 0 */
1274 /* Get the next new mbuf */
1275 n
= m_getl(remain
, how
, m
->m_type
, top
== NULL
? M_PKTHDR
: 0,
1280 if (!m_dup_pkthdr(n
, m
, how
))
1283 /* Link it into the new chain */
1287 /* Copy data from original mbuf(s) into new mbuf */
1289 while (n
->m_len
< nsize
&& m
!= NULL
) {
1290 int chunk
= min(nsize
- n
->m_len
, m
->m_len
- moff
);
1292 bcopy(m
->m_data
+ moff
, n
->m_data
+ n
->m_len
, chunk
);
1296 if (moff
== m
->m_len
) {
1302 /* Check correct total mbuf length */
1303 KASSERT((remain
> 0 && m
!= NULL
) || (remain
== 0 && m
== NULL
),
1304 ("%s: bogus m_pkthdr.len", __func__
));
1311 atomic_add_long_nonlocked(&mbstat
[mycpu
->gd_cpuid
].m_mcfail
, 1);
1316 * Concatenate mbuf chain n to m.
1317 * Both chains must be of the same type (e.g. MT_DATA).
1318 * Any m_pkthdr is not updated.
1321 m_cat(struct mbuf
*m
, struct mbuf
*n
)
1325 if (m
->m_flags
& M_EXT
||
1326 m
->m_data
+ m
->m_len
+ n
->m_len
>= &m
->m_dat
[MLEN
]) {
1327 /* just join the two chains */
1331 /* splat the data from one into the other */
1332 bcopy(mtod(n
, caddr_t
), mtod(m
, caddr_t
) + m
->m_len
,
1334 m
->m_len
+= n
->m_len
;
1340 m_adj(struct mbuf
*mp
, int req_len
)
1346 if ((m
= mp
) == NULL
)
1352 while (m
!= NULL
&& len
> 0) {
1353 if (m
->m_len
<= len
) {
1364 if (mp
->m_flags
& M_PKTHDR
)
1365 m
->m_pkthdr
.len
-= (req_len
- len
);
1368 * Trim from tail. Scan the mbuf chain,
1369 * calculating its length and finding the last mbuf.
1370 * If the adjustment only affects this mbuf, then just
1371 * adjust and return. Otherwise, rescan and truncate
1372 * after the remaining size.
1378 if (m
->m_next
== (struct mbuf
*)0)
1382 if (m
->m_len
>= len
) {
1384 if (mp
->m_flags
& M_PKTHDR
)
1385 mp
->m_pkthdr
.len
-= len
;
1392 * Correct length for chain is "count".
1393 * Find the mbuf with last data, adjust its length,
1394 * and toss data from remaining mbufs on chain.
1397 if (m
->m_flags
& M_PKTHDR
)
1398 m
->m_pkthdr
.len
= count
;
1399 for (; m
; m
= m
->m_next
) {
1400 if (m
->m_len
>= count
) {
1407 (m
= m
->m_next
) ->m_len
= 0;
1412 * Rearrange an mbuf chain so that len bytes are contiguous
1413 * and in the data area of an mbuf (so that mtod will work for a structure
1414 * of size len). Returns the resulting mbuf chain on success, frees it and
1415 * returns null on failure. If there is room, it will add up to
1416 * max_protohdr-len extra bytes to the contiguous region in an attempt to
1417 * avoid being called next time.
1420 m_pullup(struct mbuf
*n
, int len
)
1427 * If first mbuf has no cluster, and has room for len bytes
1428 * without shifting current data, pullup into it,
1429 * otherwise allocate a new mbuf to prepend to the chain.
1431 if (!(n
->m_flags
& M_EXT
) &&
1432 n
->m_data
+ len
< &n
->m_dat
[MLEN
] &&
1434 if (n
->m_len
>= len
)
1442 if (n
->m_flags
& M_PKTHDR
)
1443 m
= m_gethdr(MB_DONTWAIT
, n
->m_type
);
1445 m
= m_get(MB_DONTWAIT
, n
->m_type
);
1449 if (n
->m_flags
& M_PKTHDR
)
1450 M_MOVE_PKTHDR(m
, n
);
1452 space
= &m
->m_dat
[MLEN
] - (m
->m_data
+ m
->m_len
);
1454 count
= min(min(max(len
, max_protohdr
), space
), n
->m_len
);
1455 bcopy(mtod(n
, caddr_t
), mtod(m
, caddr_t
) + m
->m_len
,
1465 } while (len
> 0 && n
);
1474 atomic_add_long_nonlocked(&mbstat
[mycpu
->gd_cpuid
].m_mcfail
, 1);
1479 * Partition an mbuf chain in two pieces, returning the tail --
1480 * all but the first len0 bytes. In case of failure, it returns NULL and
1481 * attempts to restore the chain to its original state.
1483 * Note that the resulting mbufs might be read-only, because the new
1484 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1485 * the "breaking point" happens to lie within a cluster mbuf. Use the
1486 * M_WRITABLE() macro to check for this case.
1489 m_split(struct mbuf
*m0
, int len0
, int wait
)
1492 unsigned len
= len0
, remain
;
1494 for (m
= m0
; m
&& len
> m
->m_len
; m
= m
->m_next
)
1498 remain
= m
->m_len
- len
;
1499 if (m0
->m_flags
& M_PKTHDR
) {
1500 n
= m_gethdr(wait
, m0
->m_type
);
1503 n
->m_pkthdr
.rcvif
= m0
->m_pkthdr
.rcvif
;
1504 n
->m_pkthdr
.len
= m0
->m_pkthdr
.len
- len0
;
1505 m0
->m_pkthdr
.len
= len0
;
1506 if (m
->m_flags
& M_EXT
)
1508 if (remain
> MHLEN
) {
1509 /* m can't be the lead packet */
1511 n
->m_next
= m_split(m
, len
, wait
);
1512 if (n
->m_next
== NULL
) {
1520 MH_ALIGN(n
, remain
);
1521 } else if (remain
== 0) {
1526 n
= m_get(wait
, m
->m_type
);
1532 if (m
->m_flags
& M_EXT
) {
1533 KKASSERT((n
->m_flags
& M_EXT
) == 0);
1534 n
->m_data
= m
->m_data
+ len
;
1535 m
->m_ext
.ext_ref(m
->m_ext
.ext_arg
);
1536 n
->m_ext
= m
->m_ext
;
1537 n
->m_flags
|= m
->m_flags
& (M_EXT
| M_EXT_CLUSTER
);
1539 bcopy(mtod(m
, caddr_t
) + len
, mtod(n
, caddr_t
), remain
);
1543 n
->m_next
= m
->m_next
;
1549 * Routine to copy from device local memory into mbufs.
1550 * Note: "offset" is ill-defined and always called as 0, so ignore it.
1553 m_devget(char *buf
, int len
, int offset
, struct ifnet
*ifp
,
1554 void (*copy
)(volatile const void *from
, volatile void *to
, size_t length
))
1556 struct mbuf
*m
, *mfirst
= NULL
, **mtail
;
1565 m
= m_getl(len
, MB_DONTWAIT
, MT_DATA
, flags
, &nsize
);
1570 m
->m_len
= min(len
, nsize
);
1572 if (flags
& M_PKTHDR
) {
1573 if (len
+ max_linkhdr
<= nsize
)
1574 m
->m_data
+= max_linkhdr
;
1575 m
->m_pkthdr
.rcvif
= ifp
;
1576 m
->m_pkthdr
.len
= len
;
1580 copy(buf
, m
->m_data
, (unsigned)m
->m_len
);
1591 * Routine to pad mbuf to the specified length 'padto'.
1594 m_devpad(struct mbuf
*m
, int padto
)
1596 struct mbuf
*last
= NULL
;
1599 if (padto
<= m
->m_pkthdr
.len
)
1602 padlen
= padto
- m
->m_pkthdr
.len
;
1604 /* if there's only the packet-header and we can pad there, use it. */
1605 if (m
->m_pkthdr
.len
== m
->m_len
&& M_TRAILINGSPACE(m
) >= padlen
) {
1609 * Walk packet chain to find last mbuf. We will either
1610 * pad there, or append a new mbuf and pad it
1612 for (last
= m
; last
->m_next
!= NULL
; last
= last
->m_next
)
1615 /* `last' now points to last in chain. */
1616 if (M_TRAILINGSPACE(last
) < padlen
) {
1619 /* Allocate new empty mbuf, pad it. Compact later. */
1620 MGET(n
, MB_DONTWAIT
, MT_DATA
);
1628 KKASSERT(M_TRAILINGSPACE(last
) >= padlen
);
1629 KKASSERT(M_WRITABLE(last
));
1631 /* Now zero the pad area */
1632 bzero(mtod(last
, char *) + last
->m_len
, padlen
);
1633 last
->m_len
+= padlen
;
1634 m
->m_pkthdr
.len
+= padlen
;
1639 * Copy data from a buffer back into the indicated mbuf chain,
1640 * starting "off" bytes from the beginning, extending the mbuf
1641 * chain if necessary.
1644 m_copyback(struct mbuf
*m0
, int off
, int len
, caddr_t cp
)
1647 struct mbuf
*m
= m0
, *n
;
1652 while (off
> (mlen
= m
->m_len
)) {
1655 if (m
->m_next
== NULL
) {
1656 n
= m_getclr(MB_DONTWAIT
, m
->m_type
);
1659 n
->m_len
= min(MLEN
, len
+ off
);
1665 mlen
= min (m
->m_len
- off
, len
);
1666 bcopy(cp
, off
+ mtod(m
, caddr_t
), (unsigned)mlen
);
1674 if (m
->m_next
== NULL
) {
1675 n
= m_get(MB_DONTWAIT
, m
->m_type
);
1678 n
->m_len
= min(MLEN
, len
);
1683 out
: if (((m
= m0
)->m_flags
& M_PKTHDR
) && (m
->m_pkthdr
.len
< totlen
))
1684 m
->m_pkthdr
.len
= totlen
;
1688 m_print(const struct mbuf
*m
)
1691 const struct mbuf
*m2
;
1693 len
= m
->m_pkthdr
.len
;
1696 kprintf("%p %*D\n", m2
, m2
->m_len
, (u_char
*)m2
->m_data
, "-");
1704 * "Move" mbuf pkthdr from "from" to "to".
1705 * "from" must have M_PKTHDR set, and "to" must be empty.
1708 m_move_pkthdr(struct mbuf
*to
, struct mbuf
*from
)
1710 KASSERT((to
->m_flags
& M_PKTHDR
), ("m_move_pkthdr: not packet header"));
1712 to
->m_flags
|= from
->m_flags
& M_COPYFLAGS
;
1713 to
->m_pkthdr
= from
->m_pkthdr
; /* especially tags */
1714 SLIST_INIT(&from
->m_pkthdr
.tags
); /* purge tags from src */
1718 * Duplicate "from"'s mbuf pkthdr in "to".
1719 * "from" must have M_PKTHDR set, and "to" must be empty.
1720 * In particular, this does a deep copy of the packet tags.
1723 m_dup_pkthdr(struct mbuf
*to
, const struct mbuf
*from
, int how
)
1725 KASSERT((to
->m_flags
& M_PKTHDR
), ("m_dup_pkthdr: not packet header"));
1727 to
->m_flags
= (from
->m_flags
& M_COPYFLAGS
) |
1728 (to
->m_flags
& ~M_COPYFLAGS
);
1729 to
->m_pkthdr
= from
->m_pkthdr
;
1730 SLIST_INIT(&to
->m_pkthdr
.tags
);
1731 return (m_tag_copy_chain(to
, from
, how
));
1735 * Defragment a mbuf chain, returning the shortest possible
1736 * chain of mbufs and clusters. If allocation fails and
1737 * this cannot be completed, NULL will be returned, but
1738 * the passed in chain will be unchanged. Upon success,
1739 * the original chain will be freed, and the new chain
1742 * If a non-packet header is passed in, the original
1743 * mbuf (chain?) will be returned unharmed.
1745 * m_defrag_nofree doesn't free the passed in mbuf.
1748 m_defrag(struct mbuf
*m0
, int how
)
1752 if ((m_new
= m_defrag_nofree(m0
, how
)) == NULL
)
1760 m_defrag_nofree(struct mbuf
*m0
, int how
)
1762 struct mbuf
*m_new
= NULL
, *m_final
= NULL
;
1763 int progress
= 0, length
, nsize
;
1765 if (!(m0
->m_flags
& M_PKTHDR
))
1768 #ifdef MBUF_STRESS_TEST
1769 if (m_defragrandomfailures
) {
1770 int temp
= karc4random() & 0xff;
1776 m_final
= m_getl(m0
->m_pkthdr
.len
, how
, MT_DATA
, M_PKTHDR
, &nsize
);
1777 if (m_final
== NULL
)
1779 m_final
->m_len
= 0; /* in case m0->m_pkthdr.len is zero */
1781 if (m_dup_pkthdr(m_final
, m0
, how
) == 0)
1786 while (progress
< m0
->m_pkthdr
.len
) {
1787 length
= m0
->m_pkthdr
.len
- progress
;
1788 if (length
> MCLBYTES
)
1791 if (m_new
== NULL
) {
1792 m_new
= m_getl(length
, how
, MT_DATA
, 0, &nsize
);
1797 m_copydata(m0
, progress
, length
, mtod(m_new
, caddr_t
));
1799 m_new
->m_len
= length
;
1800 if (m_new
!= m_final
)
1801 m_cat(m_final
, m_new
);
1804 if (m0
->m_next
== NULL
)
1807 m_defragbytes
+= m_final
->m_pkthdr
.len
;
1818 * Move data from uio into mbufs.
1821 m_uiomove(struct uio
*uio
)
1823 struct mbuf
*m
; /* current working mbuf */
1824 struct mbuf
*head
= NULL
; /* result mbuf chain */
1825 struct mbuf
**mp
= &head
;
1826 int resid
= uio
->uio_resid
, nsize
, flags
= M_PKTHDR
, error
;
1829 m
= m_getl(resid
, MB_WAIT
, MT_DATA
, flags
, &nsize
);
1831 m
->m_pkthdr
.len
= 0;
1832 /* Leave room for protocol headers. */
1837 m
->m_len
= min(nsize
, resid
);
1838 error
= uiomove(mtod(m
, caddr_t
), m
->m_len
, uio
);
1845 head
->m_pkthdr
.len
+= m
->m_len
;
1847 } while (resid
> 0);
1857 m_last(struct mbuf
*m
)
1865 * Return the number of bytes in an mbuf chain.
1866 * If lastm is not NULL, also return the last mbuf.
1869 m_lengthm(struct mbuf
*m
, struct mbuf
**lastm
)
1872 struct mbuf
*prev
= m
;
1885 * Like m_lengthm(), except also keep track of mbuf usage.
1888 m_countm(struct mbuf
*m
, struct mbuf
**lastm
, u_int
*pmbcnt
)
1890 u_int len
= 0, mbcnt
= 0;
1891 struct mbuf
*prev
= m
;
1896 if (m
->m_flags
& M_EXT
)
1897 mbcnt
+= m
->m_ext
.ext_size
;