Fix some typos: in to -> into.
[dragonfly.git] / sys / kern / uipc_mbuf.c
CommitLineData
984263bc 1/*
5bd48c1d
MD
2 * (MPSAFE)
3 *
0c33f36d 4 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved.
66d6c637
JH
5 * Copyright (c) 2004 The DragonFly Project. All rights reserved.
6 *
7 * This code is derived from software contributed to The DragonFly Project
8 * by Jeffrey M. Hsu.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
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.
21 *
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
33 * SUCH DAMAGE.
34 */
35
66d6c637 36/*
984263bc
MD
37 * Copyright (c) 1982, 1986, 1988, 1991, 1993
38 * The Regents of the University of California. All rights reserved.
39 *
40 * Redistribution and use in source and binary forms, with or without
41 * modification, are permitted provided that the following conditions
42 * are met:
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.
55 *
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
66 * SUCH DAMAGE.
67 *
8a3125c6 68 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94
984263bc
MD
69 * $FreeBSD: src/sys/kern/uipc_mbuf.c,v 1.51.2.24 2003/04/15 06:59:29 silby Exp $
70 */
71
72#include "opt_param.h"
73#include "opt_mbuf_stress_test.h"
74#include <sys/param.h>
75#include <sys/systm.h>
4e23f366 76#include <sys/file.h>
984263bc
MD
77#include <sys/malloc.h>
78#include <sys/mbuf.h>
79#include <sys/kernel.h>
80#include <sys/sysctl.h>
81#include <sys/domain.h>
7b6f875f 82#include <sys/objcache.h>
e9fa4b60 83#include <sys/tree.h>
984263bc 84#include <sys/protosw.h>
0c33f36d 85#include <sys/uio.h>
ef0fdad1 86#include <sys/thread.h>
a2a5ad0d 87#include <sys/globaldata.h>
5bd48c1d 88
90775e29 89#include <sys/thread2.h>
5bd48c1d 90#include <sys/spinlock2.h>
984263bc 91
1d16b2b5 92#include <machine/atomic.h>
e54488bb 93#include <machine/limits.h>
1d16b2b5 94
984263bc
MD
95#include <vm/vm.h>
96#include <vm/vm_kern.h>
97#include <vm/vm_extern.h>
98
99#ifdef INVARIANTS
100#include <machine/cpu.h>
101#endif
102
90775e29
MD
103/*
104 * mbuf cluster meta-data
105 */
7b6f875f 106struct mbcluster {
90775e29
MD
107 int32_t mcl_refs;
108 void *mcl_data;
7b6f875f 109};
90775e29 110
e9fa4b60
MD
111/*
112 * mbuf tracking for debugging purposes
113 */
114#ifdef MBUF_DEBUG
115
116static MALLOC_DEFINE(M_MTRACK, "mtrack", "mtrack");
117
118struct mbctrack;
119RB_HEAD(mbuf_rb_tree, mbtrack);
120RB_PROTOTYPE2(mbuf_rb_tree, mbtrack, rb_node, mbtrack_cmp, struct mbuf *);
121
122struct mbtrack {
123 RB_ENTRY(mbtrack) rb_node;
124 int trackid;
125 struct mbuf *m;
126};
127
128static int
129mbtrack_cmp(struct mbtrack *mb1, struct mbtrack *mb2)
130{
131 if (mb1->m < mb2->m)
132 return(-1);
133 if (mb1->m > mb2->m)
134 return(1);
135 return(0);
136}
137
138RB_GENERATE2(mbuf_rb_tree, mbtrack, rb_node, mbtrack_cmp, struct mbuf *, m);
139
140struct mbuf_rb_tree mbuf_track_root;
5bd48c1d 141static struct spinlock mbuf_track_spin = SPINLOCK_INITIALIZER(mbuf_track_spin);
e9fa4b60
MD
142
143static void
144mbuftrack(struct mbuf *m)
145{
146 struct mbtrack *mbt;
147
e9fa4b60 148 mbt = kmalloc(sizeof(*mbt), M_MTRACK, M_INTWAIT|M_ZERO);
5bd48c1d 149 spin_lock(&mbuf_track_spin);
e9fa4b60 150 mbt->m = m;
5bd48c1d
MD
151 if (mbuf_rb_tree_RB_INSERT(&mbuf_track_root, mbt)) {
152 spin_unlock(&mbuf_track_spin);
e9fa4b60 153 panic("mbuftrack: mbuf %p already being tracked\n", m);
5bd48c1d
MD
154 }
155 spin_unlock(&mbuf_track_spin);
e9fa4b60
MD
156}
157
158static void
159mbufuntrack(struct mbuf *m)
160{
161 struct mbtrack *mbt;
162
5bd48c1d 163 spin_lock(&mbuf_track_spin);
e9fa4b60
MD
164 mbt = mbuf_rb_tree_RB_LOOKUP(&mbuf_track_root, m);
165 if (mbt == NULL) {
5bd48c1d
MD
166 spin_unlock(&mbuf_track_spin);
167 panic("mbufuntrack: mbuf %p was not tracked\n", m);
e9fa4b60
MD
168 } else {
169 mbuf_rb_tree_RB_REMOVE(&mbuf_track_root, mbt);
6cef7136 170 spin_unlock(&mbuf_track_spin);
e9fa4b60
MD
171 kfree(mbt, M_MTRACK);
172 }
e9fa4b60
MD
173}
174
175void
176mbuftrackid(struct mbuf *m, int trackid)
177{
178 struct mbtrack *mbt;
179 struct mbuf *n;
180
5bd48c1d 181 spin_lock(&mbuf_track_spin);
e9fa4b60
MD
182 while (m) {
183 n = m->m_nextpkt;
184 while (m) {
185 mbt = mbuf_rb_tree_RB_LOOKUP(&mbuf_track_root, m);
5bd48c1d
MD
186 if (mbt == NULL) {
187 spin_unlock(&mbuf_track_spin);
188 panic("mbuftrackid: mbuf %p not tracked", m);
189 }
190 mbt->trackid = trackid;
e9fa4b60
MD
191 m = m->m_next;
192 }
193 m = n;
194 }
5bd48c1d 195 spin_unlock(&mbuf_track_spin);
e9fa4b60
MD
196}
197
198static int
199mbuftrack_callback(struct mbtrack *mbt, void *arg)
200{
201 struct sysctl_req *req = arg;
202 char buf[64];
203 int error;
204
205 ksnprintf(buf, sizeof(buf), "mbuf %p track %d\n", mbt->m, mbt->trackid);
206
5bd48c1d 207 spin_unlock(&mbuf_track_spin);
e9fa4b60 208 error = SYSCTL_OUT(req, buf, strlen(buf));
5bd48c1d 209 spin_lock(&mbuf_track_spin);
e9fa4b60
MD
210 if (error)
211 return(-error);
212 return(0);
213}
214
215static int
216mbuftrack_show(SYSCTL_HANDLER_ARGS)
217{
218 int error;
219
5bd48c1d 220 spin_lock(&mbuf_track_spin);
e9fa4b60
MD
221 error = mbuf_rb_tree_RB_SCAN(&mbuf_track_root, NULL,
222 mbuftrack_callback, req);
5bd48c1d 223 spin_unlock(&mbuf_track_spin);
e9fa4b60
MD
224 return (-error);
225}
226SYSCTL_PROC(_kern_ipc, OID_AUTO, showmbufs, CTLFLAG_RD|CTLTYPE_STRING,
227 0, 0, mbuftrack_show, "A", "Show all in-use mbufs");
228
229#else
230
231#define mbuftrack(m)
232#define mbufuntrack(m)
233
234#endif
235
7b6f875f 236static void mbinit(void *);
ba39e2e0 237SYSINIT(mbuf, SI_BOOT2_MACHDEP, SI_ORDER_FIRST, mbinit, NULL)
984263bc 238
4c1e2509 239static u_long mbtypes[SMP_MAXCPU][MT_NTYPES];
90775e29 240
4c1e2509 241static struct mbstat mbstat[SMP_MAXCPU];
984263bc
MD
242int max_linkhdr;
243int max_protohdr;
244int max_hdr;
245int max_datalen;
246int m_defragpackets;
247int m_defragbytes;
248int m_defraguseless;
249int m_defragfailure;
250#ifdef MBUF_STRESS_TEST
251int m_defragrandomfailures;
252#endif
253
7b6f875f 254struct objcache *mbuf_cache, *mbufphdr_cache;
94eaee9a 255struct objcache *mclmeta_cache, *mjclmeta_cache;
7b6f875f 256struct objcache *mbufcluster_cache, *mbufphdrcluster_cache;
94eaee9a 257struct objcache *mbufjcluster_cache, *mbufphdrjcluster_cache;
7b6f875f 258
984263bc
MD
259int nmbclusters;
260int nmbufs;
984263bc 261
984263bc 262SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW,
093e85dc 263 &max_linkhdr, 0, "Max size of a link-level header");
984263bc 264SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW,
093e85dc
SG
265 &max_protohdr, 0, "Max size of a protocol header");
266SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0,
267 "Max size of link+protocol headers");
984263bc 268SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW,
093e85dc 269 &max_datalen, 0, "Max data payload size without headers");
984263bc 270SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW,
093e85dc 271 &mbuf_wait, 0, "Time in ticks to sleep after failed mbuf allocations");
4c1e2509
JT
272static int do_mbstat(SYSCTL_HANDLER_ARGS);
273
274SYSCTL_PROC(_kern_ipc, KIPC_MBSTAT, mbstat, CTLTYPE_STRUCT|CTLFLAG_RD,
093e85dc 275 0, 0, do_mbstat, "S,mbstat", "mbuf usage statistics");
4c1e2509
JT
276
277static int do_mbtypes(SYSCTL_HANDLER_ARGS);
278
279SYSCTL_PROC(_kern_ipc, OID_AUTO, mbtypes, CTLTYPE_ULONG|CTLFLAG_RD,
280 0, 0, do_mbtypes, "LU", "");
281
282static int
283do_mbstat(SYSCTL_HANDLER_ARGS)
284{
285 struct mbstat mbstat_total;
286 struct mbstat *mbstat_totalp;
287 int i;
288
289 bzero(&mbstat_total, sizeof(mbstat_total));
290 mbstat_totalp = &mbstat_total;
291
292 for (i = 0; i < ncpus; i++)
293 {
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;
303
304 }
305 /*
306 * The following fields are not cumulative fields so just
307 * get their values once.
308 */
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;
314
315 return(sysctl_handle_opaque(oidp, mbstat_totalp, sizeof(mbstat_total), req));
316}
317
318static int
319do_mbtypes(SYSCTL_HANDLER_ARGS)
320{
321 u_long totals[MT_NTYPES];
322 int i, j;
323
324 for (i = 0; i < MT_NTYPES; i++)
325 totals[i] = 0;
326
327 for (i = 0; i < ncpus; i++)
328 {
329 for (j = 0; j < MT_NTYPES; j++)
330 totals[j] += mbtypes[i][j];
331 }
332
333 return(sysctl_handle_opaque(oidp, totals, sizeof(totals), req));
334}
18c48b9c
MD
335
336/*
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.
340 */
341SYSCTL_INT(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLFLAG_RD,
984263bc 342 &nmbclusters, 0, "Maximum number of mbuf clusters available");
18c48b9c 343SYSCTL_INT(_kern_ipc, OID_AUTO, nmbufs, CTLFLAG_RD, &nmbufs, 0,
984263bc 344 "Maximum number of mbufs available");
7b6f875f 345
984263bc 346SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
093e85dc 347 &m_defragpackets, 0, "Number of defragment packets");
984263bc 348SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
093e85dc 349 &m_defragbytes, 0, "Number of defragment bytes");
984263bc 350SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
093e85dc 351 &m_defraguseless, 0, "Number of useless defragment mbuf chain operations");
984263bc 352SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
093e85dc 353 &m_defragfailure, 0, "Number of failed defragment mbuf chain operations");
984263bc
MD
354#ifdef MBUF_STRESS_TEST
355SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
356 &m_defragrandomfailures, 0, "");
357#endif
358
90775e29
MD
359static MALLOC_DEFINE(M_MBUF, "mbuf", "mbuf");
360static MALLOC_DEFINE(M_MBUFCL, "mbufcl", "mbufcl");
94eaee9a 361static MALLOC_DEFINE(M_MJBUFCL, "mbufcl", "mbufcl");
7b6f875f 362static MALLOC_DEFINE(M_MCLMETA, "mclmeta", "mclmeta");
94eaee9a 363static MALLOC_DEFINE(M_MJCLMETA, "mjclmeta", "mjclmeta");
90775e29
MD
364
365static void m_reclaim (void);
90775e29
MD
366static void m_mclref(void *arg);
367static void m_mclfree(void *arg);
984263bc 368
4e23f366
MD
369/*
370 * NOTE: Default NMBUFS must take into account a possible DOS attack
371 * using fd passing on unix domain sockets.
372 */
984263bc
MD
373#ifndef NMBCLUSTERS
374#define NMBCLUSTERS (512 + maxusers * 16)
375#endif
376#ifndef NMBUFS
4e23f366 377#define NMBUFS (nmbclusters * 2 + maxfiles)
984263bc
MD
378#endif
379
380/*
381 * Perform sanity checks of tunables declared above.
382 */
383static void
384tunable_mbinit(void *dummy)
385{
984263bc
MD
386 /*
387 * This has to be done before VM init.
388 */
389 nmbclusters = NMBCLUSTERS;
390 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
391 nmbufs = NMBUFS;
392 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs);
393 /* Sanity checks */
394 if (nmbufs < nmbclusters * 2)
395 nmbufs = nmbclusters * 2;
984263bc 396}
ba39e2e0
MD
397SYSINIT(tunable_mbinit, SI_BOOT1_TUNABLES, SI_ORDER_ANY,
398 tunable_mbinit, NULL);
984263bc
MD
399
400/* "number of clusters of pages" */
401#define NCL_INIT 1
402
403#define NMB_INIT 16
404
7b6f875f
JH
405/*
406 * The mbuf object cache only guarantees that m_next and m_nextpkt are
407 * NULL and that m_data points to the beginning of the data area. In
408 * particular, m_len and m_pkthdr.len are uninitialized. It is the
409 * responsibility of the caller to initialize those fields before use.
410 */
411
db11cb20 412static __inline boolean_t
7b6f875f 413mbuf_ctor(void *obj, void *private, int ocflags)
984263bc 414{
7b6f875f 415 struct mbuf *m = obj;
984263bc 416
7b6f875f
JH
417 m->m_next = NULL;
418 m->m_nextpkt = NULL;
419 m->m_data = m->m_dat;
420 m->m_flags = 0;
421
422 return (TRUE);
984263bc
MD
423}
424
425/*
7b6f875f 426 * Initialize the mbuf and the packet header fields.
984263bc 427 */
7b6f875f
JH
428static boolean_t
429mbufphdr_ctor(void *obj, void *private, int ocflags)
984263bc 430{
7b6f875f 431 struct mbuf *m = obj;
984263bc 432
7b6f875f
JH
433 m->m_next = NULL;
434 m->m_nextpkt = NULL;
435 m->m_data = m->m_pktdat;
77e294a1 436 m->m_flags = M_PKTHDR | M_PHCACHE;
984263bc 437
7b6f875f
JH
438 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */
439 SLIST_INIT(&m->m_pkthdr.tags);
440 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */
441 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */
442
443 return (TRUE);
984263bc
MD
444}
445
446/*
7b6f875f 447 * A mbcluster object consists of 2K (MCLBYTES) cluster and a refcount.
984263bc 448 */
7b6f875f
JH
449static boolean_t
450mclmeta_ctor(void *obj, void *private, int ocflags)
984263bc 451{
7b6f875f
JH
452 struct mbcluster *cl = obj;
453 void *buf;
454
455 if (ocflags & M_NOWAIT)
efda3bd0 456 buf = kmalloc(MCLBYTES, M_MBUFCL, M_NOWAIT | M_ZERO);
7b6f875f 457 else
efda3bd0 458 buf = kmalloc(MCLBYTES, M_MBUFCL, M_INTWAIT | M_ZERO);
7b6f875f
JH
459 if (buf == NULL)
460 return (FALSE);
77e294a1 461 cl->mcl_refs = 0;
7b6f875f
JH
462 cl->mcl_data = buf;
463 return (TRUE);
464}
984263bc 465
94eaee9a
JT
466static boolean_t
467mjclmeta_ctor(void *obj, void *private, int ocflags)
468{
469 struct mbcluster *cl = obj;
470 void *buf;
471
472 if (ocflags & M_NOWAIT)
473 buf = kmalloc(MJUMPAGESIZE, M_MBUFCL, M_NOWAIT | M_ZERO);
474 else
475 buf = kmalloc(MJUMPAGESIZE, M_MBUFCL, M_INTWAIT | M_ZERO);
476 if (buf == NULL)
477 return (FALSE);
478 cl->mcl_refs = 0;
479 cl->mcl_data = buf;
480 return (TRUE);
481}
482
c3ef87ca
MD
483static void
484mclmeta_dtor(void *obj, void *private)
485{
486 struct mbcluster *mcl = obj;
487
488 KKASSERT(mcl->mcl_refs == 0);
efda3bd0 489 kfree(mcl->mcl_data, M_MBUFCL);
c3ef87ca
MD
490}
491
7b6f875f 492static void
94eaee9a 493linkjcluster(struct mbuf *m, struct mbcluster *cl, uint size)
7b6f875f 494{
984263bc 495 /*
7b6f875f
JH
496 * Add the cluster to the mbuf. The caller will detect that the
497 * mbuf now has an attached cluster.
984263bc 498 */
7b6f875f
JH
499 m->m_ext.ext_arg = cl;
500 m->m_ext.ext_buf = cl->mcl_data;
501 m->m_ext.ext_ref = m_mclref;
502 m->m_ext.ext_free = m_mclfree;
94eaee9a 503 m->m_ext.ext_size = size;
df8d1020 504 atomic_add_int(&cl->mcl_refs, 1);
984263bc 505
7b6f875f
JH
506 m->m_data = m->m_ext.ext_buf;
507 m->m_flags |= M_EXT | M_EXT_CLUSTER;
984263bc
MD
508}
509
94eaee9a
JT
510static void
511linkcluster(struct mbuf *m, struct mbcluster *cl)
512{
513 linkjcluster(m, cl, MCLBYTES);
514}
515
7b6f875f
JH
516static boolean_t
517mbufphdrcluster_ctor(void *obj, void *private, int ocflags)
518{
519 struct mbuf *m = obj;
520 struct mbcluster *cl;
521
522 mbufphdr_ctor(obj, private, ocflags);
523 cl = objcache_get(mclmeta_cache, ocflags);
a5955b15
MD
524 if (cl == NULL) {
525 ++mbstat[mycpu->gd_cpuid].m_drops;
7b6f875f 526 return (FALSE);
a5955b15 527 }
77e294a1 528 m->m_flags |= M_CLCACHE;
7b6f875f
JH
529 linkcluster(m, cl);
530 return (TRUE);
531}
984263bc 532
94eaee9a
JT
533static boolean_t
534mbufphdrjcluster_ctor(void *obj, void *private, int ocflags)
535{
536 struct mbuf *m = obj;
537 struct mbcluster *cl;
538
539 mbufphdr_ctor(obj, private, ocflags);
540 cl = objcache_get(mjclmeta_cache, ocflags);
541 if (cl == NULL) {
542 ++mbstat[mycpu->gd_cpuid].m_drops;
543 return (FALSE);
544 }
545 m->m_flags |= M_CLCACHE;
546 linkjcluster(m, cl, MJUMPAGESIZE);
547 return (TRUE);
548}
549
7b6f875f
JH
550static boolean_t
551mbufcluster_ctor(void *obj, void *private, int ocflags)
984263bc 552{
7b6f875f
JH
553 struct mbuf *m = obj;
554 struct mbcluster *cl;
555
556 mbuf_ctor(obj, private, ocflags);
557 cl = objcache_get(mclmeta_cache, ocflags);
a5955b15
MD
558 if (cl == NULL) {
559 ++mbstat[mycpu->gd_cpuid].m_drops;
7b6f875f 560 return (FALSE);
a5955b15 561 }
77e294a1 562 m->m_flags |= M_CLCACHE;
7b6f875f
JH
563 linkcluster(m, cl);
564 return (TRUE);
565}
984263bc 566
94eaee9a
JT
567static boolean_t
568mbufjcluster_ctor(void *obj, void *private, int ocflags)
569{
570 struct mbuf *m = obj;
571 struct mbcluster *cl;
572
573 mbuf_ctor(obj, private, ocflags);
574 cl = objcache_get(mjclmeta_cache, ocflags);
575 if (cl == NULL) {
576 ++mbstat[mycpu->gd_cpuid].m_drops;
577 return (FALSE);
578 }
579 m->m_flags |= M_CLCACHE;
580 linkjcluster(m, cl, MJUMPAGESIZE);
581 return (TRUE);
582}
583
77e294a1
MD
584/*
585 * Used for both the cluster and cluster PHDR caches.
586 *
587 * The mbuf may have lost its cluster due to sharing, deal
588 * with the situation by checking M_EXT.
589 */
7b6f875f
JH
590static void
591mbufcluster_dtor(void *obj, void *private)
984263bc 592{
7b6f875f 593 struct mbuf *m = obj;
77e294a1 594 struct mbcluster *mcl;
984263bc 595
77e294a1
MD
596 if (m->m_flags & M_EXT) {
597 KKASSERT((m->m_flags & M_EXT_CLUSTER) != 0);
598 mcl = m->m_ext.ext_arg;
599 KKASSERT(mcl->mcl_refs == 1);
600 mcl->mcl_refs = 0;
94eaee9a
JT
601 if (m->m_flags & M_EXT && m->m_ext.ext_size != MCLBYTES)
602 objcache_put(mjclmeta_cache, mcl);
603 else
604 objcache_put(mclmeta_cache, mcl);
77e294a1 605 }
984263bc
MD
606}
607
7b6f875f
JH
608struct objcache_malloc_args mbuf_malloc_args = { MSIZE, M_MBUF };
609struct objcache_malloc_args mclmeta_malloc_args =
610 { sizeof(struct mbcluster), M_MCLMETA };
611
612/* ARGSUSED*/
90775e29 613static void
7b6f875f 614mbinit(void *dummy)
984263bc 615{
6f21e2f4 616 int mb_limit, cl_limit;
0aa16b5d 617 int limit;
4c1e2509
JT
618 int i;
619
0aa16b5d
SZ
620 /*
621 * Initialize statistics
622 */
623 for (i = 0; i < ncpus; i++) {
4c1e2509
JT
624 atomic_set_long_nonlocked(&mbstat[i].m_msize, MSIZE);
625 atomic_set_long_nonlocked(&mbstat[i].m_mclbytes, MCLBYTES);
94eaee9a 626 atomic_set_long_nonlocked(&mbstat[i].m_mjumpagesize, MJUMPAGESIZE);
4c1e2509
JT
627 atomic_set_long_nonlocked(&mbstat[i].m_minclsize, MINCLSIZE);
628 atomic_set_long_nonlocked(&mbstat[i].m_mlen, MLEN);
629 atomic_set_long_nonlocked(&mbstat[i].m_mhlen, MHLEN);
630 }
984263bc 631
0aa16b5d
SZ
632 /*
633 * Create objtect caches and save cluster limits, which will
634 * be used to adjust backing kmalloc pools' limit later.
635 */
636
6f21e2f4 637 mb_limit = cl_limit = 0;
0aa16b5d
SZ
638
639 limit = nmbufs;
3508d9a1
MD
640 mbuf_cache = objcache_create("mbuf",
641 &limit, 0,
5b7da64a 642 mbuf_ctor, NULL, NULL,
7b6f875f 643 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
6f21e2f4 644 mb_limit += limit;
0aa16b5d
SZ
645
646 limit = nmbufs;
3508d9a1
MD
647 mbufphdr_cache = objcache_create("mbuf pkt hdr",
648 &limit, nmbufs / 4,
5b7da64a 649 mbufphdr_ctor, NULL, NULL,
7b6f875f 650 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
6f21e2f4 651 mb_limit += limit;
0aa16b5d
SZ
652
653 cl_limit = nmbclusters;
3508d9a1
MD
654 mclmeta_cache = objcache_create("cluster mbuf",
655 &cl_limit, 0,
7b6f875f
JH
656 mclmeta_ctor, mclmeta_dtor, NULL,
657 objcache_malloc_alloc, objcache_malloc_free, &mclmeta_malloc_args);
0aa16b5d 658
94eaee9a 659 cl_limit = nmbclusters;
3508d9a1
MD
660 mjclmeta_cache = objcache_create("jcluster mbuf",
661 &cl_limit, 0,
94eaee9a
JT
662 mjclmeta_ctor, mclmeta_dtor, NULL,
663 objcache_malloc_alloc, objcache_malloc_free, &mclmeta_malloc_args);
664
0aa16b5d 665 limit = nmbclusters;
3508d9a1
MD
666 mbufcluster_cache = objcache_create("mbuf + cluster",
667 &limit, 0,
7b6f875f
JH
668 mbufcluster_ctor, mbufcluster_dtor, NULL,
669 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
6f21e2f4 670 mb_limit += limit;
0aa16b5d
SZ
671
672 limit = nmbclusters;
7b6f875f 673 mbufphdrcluster_cache = objcache_create("mbuf pkt hdr + cluster",
3508d9a1
MD
674 &limit, nmbclusters / 16,
675 mbufphdrcluster_ctor, mbufcluster_dtor, NULL,
7b6f875f 676 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
6f21e2f4 677 mb_limit += limit;
0aa16b5d 678
94eaee9a 679 limit = nmbclusters;
3508d9a1
MD
680 mbufjcluster_cache = objcache_create("mbuf + jcluster",
681 &limit, 0,
94eaee9a
JT
682 mbufjcluster_ctor, mbufcluster_dtor, NULL,
683 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
684 mb_limit += limit;
685
686 limit = nmbclusters;
687 mbufphdrjcluster_cache = objcache_create("mbuf pkt hdr + jcluster",
3508d9a1
MD
688 &limit, nmbclusters / 16,
689 mbufphdrjcluster_ctor, mbufcluster_dtor, NULL,
94eaee9a
JT
690 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args);
691 mb_limit += limit;
692
0aa16b5d
SZ
693 /*
694 * Adjust backing kmalloc pools' limit
3f98f485
SZ
695 *
696 * NOTE: We raise the limit by another 1/8 to take the effect
697 * of loosememuse into account.
0aa16b5d 698 */
3f98f485 699 cl_limit += cl_limit / 8;
0aa16b5d
SZ
700 kmalloc_raise_limit(mclmeta_malloc_args.mtype,
701 mclmeta_malloc_args.objsize * cl_limit);
94eaee9a
JT
702 kmalloc_raise_limit(M_MBUFCL, MCLBYTES * cl_limit * 3/4 + MJUMPAGESIZE * cl_limit / 4);
703 /*kmalloc_raise_limit(M_MBUFCL, MCLBYTES * cl_limit);*/
0aa16b5d 704
3f98f485 705 mb_limit += mb_limit / 8;
0aa16b5d
SZ
706 kmalloc_raise_limit(mbuf_malloc_args.mtype,
707 mbuf_malloc_args.objsize * mb_limit);
90775e29 708}
984263bc 709
90775e29
MD
710/*
711 * Return the number of references to this mbuf's data. 0 is returned
712 * if the mbuf is not M_EXT, a reference count is returned if it is
7b6f875f 713 * M_EXT | M_EXT_CLUSTER, and 99 is returned if it is a special M_EXT.
90775e29
MD
714 */
715int
716m_sharecount(struct mbuf *m)
717{
7b6f875f
JH
718 switch (m->m_flags & (M_EXT | M_EXT_CLUSTER)) {
719 case 0:
720 return (0);
721 case M_EXT:
722 return (99);
723 case M_EXT | M_EXT_CLUSTER:
724 return (((struct mbcluster *)m->m_ext.ext_arg)->mcl_refs);
725 }
726 /* NOTREACHED */
727 return (0); /* to shut up compiler */
90775e29
MD
728}
729
730/*
731 * change mbuf to new type
732 */
733void
734m_chtype(struct mbuf *m, int type)
735{
4c1e2509
JT
736 struct globaldata *gd = mycpu;
737
738 atomic_add_long_nonlocked(&mbtypes[gd->gd_cpuid][type], 1);
739 atomic_subtract_long_nonlocked(&mbtypes[gd->gd_cpuid][m->m_type], 1);
740 atomic_set_short_nonlocked(&m->m_type, type);
984263bc
MD
741}
742
984263bc 743static void
8a3125c6 744m_reclaim(void)
984263bc 745{
1fd87d54
RG
746 struct domain *dp;
747 struct protosw *pr;
984263bc 748
5bd48c1d
MD
749 kprintf("Debug: m_reclaim() called\n");
750
9c70fe43 751 SLIST_FOREACH(dp, &domains, dom_next) {
8a3125c6 752 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
984263bc
MD
753 if (pr->pr_drain)
754 (*pr->pr_drain)();
8a3125c6
MD
755 }
756 }
4c1e2509 757 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_drain, 1);
984263bc
MD
758}
759
db11cb20 760static __inline void
7b6f875f
JH
761updatestats(struct mbuf *m, int type)
762{
4c1e2509 763 struct globaldata *gd = mycpu;
7b6f875f 764
fcd1202a 765 m->m_type = type;
e9fa4b60 766 mbuftrack(m);
982f999d
MD
767#ifdef MBUF_DEBUG
768 KASSERT(m->m_next == NULL, ("mbuf %p: bad m_next in get", m));
769 KASSERT(m->m_nextpkt == NULL, ("mbuf %p: bad m_nextpkt in get", m));
770#endif
4c1e2509
JT
771
772 atomic_add_long_nonlocked(&mbtypes[gd->gd_cpuid][type], 1);
773 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mbufs, 1);
774
7b6f875f
JH
775}
776
984263bc 777/*
7b6f875f 778 * Allocate an mbuf.
984263bc
MD
779 */
780struct mbuf *
8a3125c6 781m_get(int how, int type)
984263bc 782{
12496bdf 783 struct mbuf *m;
7b6f875f
JH
784 int ntries = 0;
785 int ocf = MBTOM(how);
12496bdf 786
7b6f875f
JH
787retryonce:
788
789 m = objcache_get(mbuf_cache, ocf);
790
791 if (m == NULL) {
792 if ((how & MB_TRYWAIT) && ntries++ == 0) {
793 struct objcache *reclaimlist[] = {
794 mbufphdr_cache,
5bd48c1d 795 mbufcluster_cache,
94eaee9a
JT
796 mbufphdrcluster_cache,
797 mbufjcluster_cache,
798 mbufphdrjcluster_cache
7b6f875f 799 };
a3034532 800 const int nreclaims = NELEM(reclaimlist);
7b6f875f
JH
801
802 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf))
803 m_reclaim();
804 goto retryonce;
c6339e39 805 }
a5955b15 806 ++mbstat[mycpu->gd_cpuid].m_drops;
7b6f875f 807 return (NULL);
12496bdf 808 }
982f999d
MD
809#ifdef MBUF_DEBUG
810 KASSERT(m->m_data == m->m_dat, ("mbuf %p: bad m_data in get", m));
811#endif
5bd08532 812 m->m_len = 0;
c6339e39 813
7b6f875f 814 updatestats(m, type);
984263bc
MD
815 return (m);
816}
817
818struct mbuf *
8a3125c6 819m_gethdr(int how, int type)
984263bc 820{
12496bdf 821 struct mbuf *m;
7b6f875f
JH
822 int ocf = MBTOM(how);
823 int ntries = 0;
12496bdf 824
7b6f875f
JH
825retryonce:
826
827 m = objcache_get(mbufphdr_cache, ocf);
828
829 if (m == NULL) {
830 if ((how & MB_TRYWAIT) && ntries++ == 0) {
831 struct objcache *reclaimlist[] = {
832 mbuf_cache,
94eaee9a
JT
833 mbufcluster_cache, mbufphdrcluster_cache,
834 mbufjcluster_cache, mbufphdrjcluster_cache
7b6f875f 835 };
a3034532 836 const int nreclaims = NELEM(reclaimlist);
7b6f875f
JH
837
838 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf))
839 m_reclaim();
840 goto retryonce;
c6339e39 841 }
a5955b15 842 ++mbstat[mycpu->gd_cpuid].m_drops;
7b6f875f 843 return (NULL);
12496bdf 844 }
982f999d
MD
845#ifdef MBUF_DEBUG
846 KASSERT(m->m_data == m->m_pktdat, ("mbuf %p: bad m_data in get", m));
847#endif
5bd08532
MD
848 m->m_len = 0;
849 m->m_pkthdr.len = 0;
c6339e39 850
7b6f875f 851 updatestats(m, type);
984263bc
MD
852 return (m);
853}
854
7b6f875f
JH
855/*
856 * Get a mbuf (not a mbuf cluster!) and zero it.
857 * Deprecated.
858 */
984263bc 859struct mbuf *
8a3125c6 860m_getclr(int how, int type)
984263bc 861{
1fd87d54 862 struct mbuf *m;
984263bc 863
7b6f875f
JH
864 m = m_get(how, type);
865 if (m != NULL)
866 bzero(m->m_data, MLEN);
984263bc
MD
867 return (m);
868}
869
94eaee9a
JT
870struct mbuf *
871m_getjcl(int how, short type, int flags, size_t size)
872{
873 struct mbuf *m = NULL;
2e7afdb4 874 struct objcache *mbclc, *mbphclc;
94eaee9a
JT
875 int ocflags = MBTOM(how);
876 int ntries = 0;
877
2e7afdb4
JT
878 switch (size) {
879 case MCLBYTES:
880 mbclc = mbufcluster_cache;
881 mbphclc = mbufphdrcluster_cache;
882 break;
883 default:
884 mbclc = mbufjcluster_cache;
885 mbphclc = mbufphdrjcluster_cache;
886 break;
887 }
888
94eaee9a
JT
889retryonce:
890
891 if (flags & M_PKTHDR)
2e7afdb4 892 m = objcache_get(mbphclc, ocflags);
94eaee9a 893 else
2e7afdb4 894 m = objcache_get(mbclc, ocflags);
94eaee9a
JT
895
896 if (m == NULL) {
897 if ((how & MB_TRYWAIT) && ntries++ == 0) {
898 struct objcache *reclaimlist[1];
899
900 if (flags & M_PKTHDR)
2e7afdb4 901 reclaimlist[0] = mbclc;
94eaee9a 902 else
2e7afdb4 903 reclaimlist[0] = mbphclc;
94eaee9a
JT
904 if (!objcache_reclaimlist(reclaimlist, 1, ocflags))
905 m_reclaim();
906 goto retryonce;
907 }
908 ++mbstat[mycpu->gd_cpuid].m_drops;
909 return (NULL);
910 }
911
912#ifdef MBUF_DEBUG
913 KASSERT(m->m_data == m->m_ext.ext_buf,
914 ("mbuf %p: bad m_data in get", m));
915#endif
916 m->m_type = type;
917 m->m_len = 0;
918 m->m_pkthdr.len = 0; /* just do it unconditonally */
919
920 mbuftrack(m);
921
922 atomic_add_long_nonlocked(&mbtypes[mycpu->gd_cpuid][type], 1);
923 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters, 1);
924 return (m);
925}
926
984263bc 927/*
7b6f875f 928 * Returns an mbuf with an attached cluster.
984263bc
MD
929 * Because many network drivers use this kind of buffers a lot, it is
930 * convenient to keep a small pool of free buffers of this kind.
931 * Even a small size such as 10 gives about 10% improvement in the
932 * forwarding rate in a bridge or router.
984263bc 933 */
984263bc
MD
934struct mbuf *
935m_getcl(int how, short type, int flags)
936{
2e7afdb4 937 return (m_getjcl(how, type, flags, MCLBYTES));
984263bc
MD
938}
939
940/*
50503f0f
JH
941 * Allocate chain of requested length.
942 */
943struct mbuf *
944m_getc(int len, int how, int type)
945{
946 struct mbuf *n, *nfirst = NULL, **ntail = &nfirst;
947 int nsize;
948
949 while (len > 0) {
950 n = m_getl(len, how, type, 0, &nsize);
951 if (n == NULL)
952 goto failed;
953 n->m_len = 0;
954 *ntail = n;
955 ntail = &n->m_next;
956 len -= nsize;
957 }
958 return (nfirst);
959
960failed:
961 m_freem(nfirst);
962 return (NULL);
963}
964
965/*
966 * Allocate len-worth of mbufs and/or mbuf clusters (whatever fits best)
967 * and return a pointer to the head of the allocated chain. If m0 is
984263bc
MD
968 * non-null, then we assume that it is a single mbuf or an mbuf chain to
969 * which we want len bytes worth of mbufs and/or clusters attached, and so
50503f0f 970 * if we succeed in allocating it, we will just return a pointer to m0.
984263bc
MD
971 *
972 * If we happen to fail at any point during the allocation, we will free
973 * up everything we have already allocated and return NULL.
974 *
50503f0f 975 * Deprecated. Use m_getc() and m_cat() instead.
984263bc
MD
976 */
977struct mbuf *
dc14b0a9 978m_getm(struct mbuf *m0, int len, int type, int how)
984263bc 979{
50503f0f 980 struct mbuf *nfirst;
984263bc 981
50503f0f 982 nfirst = m_getc(len, how, type);
984263bc 983
50503f0f
JH
984 if (m0 != NULL) {
985 m_last(m0)->m_next = nfirst;
986 return (m0);
984263bc
MD
987 }
988
50503f0f 989 return (nfirst);
984263bc
MD
990}
991
992/*
7b6f875f
JH
993 * Adds a cluster to a normal mbuf, M_EXT is set on success.
994 * Deprecated. Use m_getcl() instead.
b6650ec0 995 */
90775e29
MD
996void
997m_mclget(struct mbuf *m, int how)
b6650ec0 998{
7b6f875f 999 struct mbcluster *mcl;
b6650ec0 1000
77e294a1 1001 KKASSERT((m->m_flags & M_EXT) == 0);
7b6f875f 1002 mcl = objcache_get(mclmeta_cache, MBTOM(how));
c3ef87ca
MD
1003 if (mcl != NULL) {
1004 linkcluster(m, mcl);
5bd48c1d
MD
1005 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters,
1006 1);
a5955b15
MD
1007 } else {
1008 ++mbstat[mycpu->gd_cpuid].m_drops;
c3ef87ca 1009 }
b6650ec0
MD
1010}
1011
df8d1020
MD
1012/*
1013 * Updates to mbcluster must be MPSAFE. Only an entity which already has
1014 * a reference to the cluster can ref it, so we are in no danger of
1015 * racing an add with a subtract. But the operation must still be atomic
1016 * since multiple entities may have a reference on the cluster.
1017 *
1018 * m_mclfree() is almost the same but it must contend with two entities
5bd48c1d 1019 * freeing the cluster at the same time.
df8d1020 1020 */
90775e29 1021static void
7b6f875f 1022m_mclref(void *arg)
b6650ec0 1023{
7b6f875f 1024 struct mbcluster *mcl = arg;
90775e29 1025
7b6f875f 1026 atomic_add_int(&mcl->mcl_refs, 1);
b6650ec0
MD
1027}
1028
1d16b2b5
MD
1029/*
1030 * When dereferencing a cluster we have to deal with a N->0 race, where
1031 * N entities free their references simultaniously. To do this we use
dee87a60 1032 * atomic_fetchadd_int().
1d16b2b5 1033 */
90775e29 1034static void
7b6f875f 1035m_mclfree(void *arg)
b6650ec0 1036{
7b6f875f 1037 struct mbcluster *mcl = arg;
90775e29 1038
dee87a60 1039 if (atomic_fetchadd_int(&mcl->mcl_refs, -1) == 1)
77e294a1 1040 objcache_put(mclmeta_cache, mcl);
b6650ec0
MD
1041}
1042
1043/*
b6650ec0
MD
1044 * Free a single mbuf and any associated external storage. The successor,
1045 * if any, is returned.
984263bc 1046 *
b6650ec0 1047 * We do need to check non-first mbuf for m_aux, since some of existing
984263bc
MD
1048 * code does not call M_PREPEND properly.
1049 * (example: call to bpf_mtap from drivers)
1050 */
982f999d
MD
1051
1052#ifdef MBUF_DEBUG
1053
1054struct mbuf *
1055_m_free(struct mbuf *m, const char *func)
1056
1057#else
1058
984263bc 1059struct mbuf *
b6650ec0 1060m_free(struct mbuf *m)
982f999d
MD
1061
1062#endif
984263bc 1063{
b6650ec0 1064 struct mbuf *n;
4c1e2509 1065 struct globaldata *gd = mycpu;
b6650ec0 1066
361af367 1067 KASSERT(m->m_type != MT_FREE, ("freeing free mbuf %p", m));
f3f0fc49 1068 KASSERT(M_TRAILINGSPACE(m) >= 0, ("overflowed mbuf %p", m));
4c1e2509 1069 atomic_subtract_long_nonlocked(&mbtypes[gd->gd_cpuid][m->m_type], 1);
90775e29 1070
7b6f875f 1071 n = m->m_next;
90775e29
MD
1072
1073 /*
7b6f875f
JH
1074 * Make sure the mbuf is in constructed state before returning it
1075 * to the objcache.
90775e29 1076 */
90775e29 1077 m->m_next = NULL;
e9fa4b60 1078 mbufuntrack(m);
982f999d
MD
1079#ifdef MBUF_DEBUG
1080 m->m_hdr.mh_lastfunc = func;
1081#endif
7b6f875f
JH
1082#ifdef notyet
1083 KKASSERT(m->m_nextpkt == NULL);
1084#else
1085 if (m->m_nextpkt != NULL) {
7b6f875f
JH
1086 static int afewtimes = 10;
1087
1088 if (afewtimes-- > 0) {
6ea70f76 1089 kprintf("mfree: m->m_nextpkt != NULL\n");
7ce2998e 1090 print_backtrace(-1);
90775e29 1091 }
7b6f875f
JH
1092 m->m_nextpkt = NULL;
1093 }
1094#endif
1095 if (m->m_flags & M_PKTHDR) {
7b6f875f 1096 m_tag_delete_chain(m); /* eliminate XXX JH */
77e294a1
MD
1097 }
1098
1099 m->m_flags &= (M_EXT | M_EXT_CLUSTER | M_CLCACHE | M_PHCACHE);
1100
1101 /*
1102 * Clean the M_PKTHDR state so we can return the mbuf to its original
1103 * cache. This is based on the PHCACHE flag which tells us whether
1104 * the mbuf was originally allocated out of a packet-header cache
1105 * or a non-packet-header cache.
1106 */
1107 if (m->m_flags & M_PHCACHE) {
1108 m->m_flags |= M_PKTHDR;
1109 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */
7b6f875f
JH
1110 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */
1111 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */
6b1d6bed 1112 SLIST_INIT(&m->m_pkthdr.tags);
90775e29 1113 }
7b6f875f 1114
77e294a1
MD
1115 /*
1116 * Handle remaining flags combinations. M_CLCACHE tells us whether
1117 * the mbuf was originally allocated from a cluster cache or not,
1118 * and is totally separate from whether the mbuf is currently
1119 * associated with a cluster.
1120 */
77e294a1
MD
1121 switch(m->m_flags & (M_CLCACHE | M_EXT | M_EXT_CLUSTER)) {
1122 case M_CLCACHE | M_EXT | M_EXT_CLUSTER:
1123 /*
1124 * mbuf+cluster cache case. The mbuf was allocated from the
1125 * combined mbuf_cluster cache and can be returned to the
1126 * cache if the cluster hasn't been shared.
1127 */
1128 if (m_sharecount(m) == 1) {
1129 /*
1130 * The cluster has not been shared, we can just
1131 * reset the data pointer and return the mbuf
1132 * to the cluster cache. Note that the reference
1133 * count is left intact (it is still associated with
1134 * an mbuf).
1135 */
1136 m->m_data = m->m_ext.ext_buf;
94eaee9a
JT
1137 if (m->m_flags & M_EXT && m->m_ext.ext_size != MCLBYTES) {
1138 if (m->m_flags & M_PHCACHE)
1139 objcache_put(mbufphdrjcluster_cache, m);
1140 else
1141 objcache_put(mbufjcluster_cache, m);
1142 } else {
1143 if (m->m_flags & M_PHCACHE)
1144 objcache_put(mbufphdrcluster_cache, m);
1145 else
1146 objcache_put(mbufcluster_cache, m);
1147 }
4c1e2509 1148 atomic_subtract_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters, 1);
77e294a1
MD
1149 } else {
1150 /*
1151 * Hell. Someone else has a ref on this cluster,
1152 * we have to disconnect it which means we can't
1153 * put it back into the mbufcluster_cache, we
1154 * have to destroy the mbuf.
1155 *
cb086467
MD
1156 * Other mbuf references to the cluster will typically
1157 * be M_EXT | M_EXT_CLUSTER but without M_CLCACHE.
1158 *
77e294a1
MD
1159 * XXX we could try to connect another cluster to
1160 * it.
1161 */
94eaee9a 1162
7b6f875f
JH
1163 m->m_ext.ext_free(m->m_ext.ext_arg);
1164 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER);
94eaee9a
JT
1165 if (m->m_ext.ext_size == MCLBYTES) {
1166 if (m->m_flags & M_PHCACHE)
1167 objcache_dtor(mbufphdrcluster_cache, m);
1168 else
1169 objcache_dtor(mbufcluster_cache, m);
1170 } else {
1171 if (m->m_flags & M_PHCACHE)
1172 objcache_dtor(mbufphdrjcluster_cache, m);
1173 else
1174 objcache_dtor(mbufjcluster_cache, m);
1175 }
7b6f875f 1176 }
77e294a1
MD
1177 break;
1178 case M_EXT | M_EXT_CLUSTER:
1179 /*
1180 * Normal cluster associated with an mbuf that was allocated
1181 * from the normal mbuf pool rather then the cluster pool.
1182 * The cluster has to be independantly disassociated from the
1183 * mbuf.
1184 */
cb086467 1185 if (m_sharecount(m) == 1)
4c1e2509 1186 atomic_subtract_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_clusters, 1);
77e294a1
MD
1187 /* fall through */
1188 case M_EXT:
1189 /*
1190 * Normal cluster association case, disconnect the cluster from
1191 * the mbuf. The cluster may or may not be custom.
1192 */
1193 m->m_ext.ext_free(m->m_ext.ext_arg);
1194 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER);
1195 /* fall through */
1196 case 0:
1197 /*
1198 * return the mbuf to the mbuf cache.
1199 */
1200 if (m->m_flags & M_PHCACHE) {
7b6f875f
JH
1201 m->m_data = m->m_pktdat;
1202 objcache_put(mbufphdr_cache, m);
90775e29 1203 } else {
7b6f875f
JH
1204 m->m_data = m->m_dat;
1205 objcache_put(mbuf_cache, m);
90775e29 1206 }
4c1e2509 1207 atomic_subtract_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mbufs, 1);
77e294a1
MD
1208 break;
1209 default:
1210 if (!panicstr)
1211 panic("bad mbuf flags %p %08x\n", m, m->m_flags);
1212 break;
b6650ec0 1213 }
984263bc
MD
1214 return (n);
1215}
1216
982f999d
MD
1217#ifdef MBUF_DEBUG
1218
1219void
1220_m_freem(struct mbuf *m, const char *func)
1221{
1222 while (m)
1223 m = _m_free(m, func);
1224}
1225
1226#else
1227
984263bc 1228void
b6650ec0 1229m_freem(struct mbuf *m)
984263bc 1230{
90775e29
MD
1231 while (m)
1232 m = m_free(m);
984263bc
MD
1233}
1234
982f999d
MD
1235#endif
1236
984263bc 1237/*
df80f2ea 1238 * mbuf utility routines
984263bc
MD
1239 */
1240
1241/*
7b6f875f 1242 * Lesser-used path for M_PREPEND: allocate new mbuf to prepend to chain and
984263bc
MD
1243 * copy junk along.
1244 */
1245struct mbuf *
8a3125c6 1246m_prepend(struct mbuf *m, int len, int how)
984263bc
MD
1247{
1248 struct mbuf *mn;
1249
c3ef87ca
MD
1250 if (m->m_flags & M_PKTHDR)
1251 mn = m_gethdr(how, m->m_type);
1252 else
1253 mn = m_get(how, m->m_type);
7b6f875f 1254 if (mn == NULL) {
984263bc 1255 m_freem(m);
7b6f875f 1256 return (NULL);
984263bc
MD
1257 }
1258 if (m->m_flags & M_PKTHDR)
1259 M_MOVE_PKTHDR(mn, m);
1260 mn->m_next = m;
1261 m = mn;
1262 if (len < MHLEN)
1263 MH_ALIGN(m, len);
1264 m->m_len = len;
1265 return (m);
1266}
1267
1268/*
1269 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
1270 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
74f1caca 1271 * The wait parameter is a choice of MB_WAIT/MB_DONTWAIT from caller.
984263bc
MD
1272 * Note that the copy is read-only, because clusters are not copied,
1273 * only their reference counts are incremented.
1274 */
984263bc 1275struct mbuf *
8a3125c6 1276m_copym(const struct mbuf *m, int off0, int len, int wait)
984263bc 1277{
1fd87d54
RG
1278 struct mbuf *n, **np;
1279 int off = off0;
984263bc
MD
1280 struct mbuf *top;
1281 int copyhdr = 0;
1282
1283 KASSERT(off >= 0, ("m_copym, negative off %d", off));
1284 KASSERT(len >= 0, ("m_copym, negative len %d", len));
5bd48c1d 1285 if (off == 0 && (m->m_flags & M_PKTHDR))
984263bc
MD
1286 copyhdr = 1;
1287 while (off > 0) {
1288 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
1289 if (off < m->m_len)
1290 break;
1291 off -= m->m_len;
1292 m = m->m_next;
1293 }
1294 np = &top;
5bd48c1d 1295 top = NULL;
984263bc 1296 while (len > 0) {
7b6f875f 1297 if (m == NULL) {
984263bc
MD
1298 KASSERT(len == M_COPYALL,
1299 ("m_copym, length > size of mbuf chain"));
1300 break;
1301 }
c3ef87ca
MD
1302 /*
1303 * Because we are sharing any cluster attachment below,
1304 * be sure to get an mbuf that does not have a cluster
1305 * associated with it.
1306 */
1307 if (copyhdr)
1308 n = m_gethdr(wait, m->m_type);
1309 else
1310 n = m_get(wait, m->m_type);
984263bc 1311 *np = n;
7b6f875f 1312 if (n == NULL)
984263bc
MD
1313 goto nospace;
1314 if (copyhdr) {
1315 if (!m_dup_pkthdr(n, m, wait))
1316 goto nospace;
1317 if (len == M_COPYALL)
1318 n->m_pkthdr.len -= off0;
1319 else
1320 n->m_pkthdr.len = len;
1321 copyhdr = 0;
1322 }
1323 n->m_len = min(len, m->m_len - off);
1324 if (m->m_flags & M_EXT) {
c3ef87ca 1325 KKASSERT((n->m_flags & M_EXT) == 0);
984263bc 1326 n->m_data = m->m_data + off;
7b6f875f 1327 m->m_ext.ext_ref(m->m_ext.ext_arg);
984263bc 1328 n->m_ext = m->m_ext;
b542cd49 1329 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
7eccf245 1330 } else {
984263bc
MD
1331 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
1332 (unsigned)n->m_len);
7eccf245 1333 }
984263bc
MD
1334 if (len != M_COPYALL)
1335 len -= n->m_len;
1336 off = 0;
1337 m = m->m_next;
1338 np = &n->m_next;
1339 }
7b6f875f 1340 if (top == NULL)
4c1e2509 1341 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
984263bc
MD
1342 return (top);
1343nospace:
1344 m_freem(top);
4c1e2509 1345 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
7b6f875f 1346 return (NULL);
984263bc
MD
1347}
1348
1349/*
1350 * Copy an entire packet, including header (which must be present).
1351 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
1352 * Note that the copy is read-only, because clusters are not copied,
1353 * only their reference counts are incremented.
1354 * Preserve alignment of the first mbuf so if the creator has left
1355 * some room at the beginning (e.g. for inserting protocol headers)
1356 * the copies also have the room available.
1357 */
1358struct mbuf *
8a3125c6 1359m_copypacket(struct mbuf *m, int how)
984263bc
MD
1360{
1361 struct mbuf *top, *n, *o;
1362
7f3602fe 1363 n = m_gethdr(how, m->m_type);
984263bc
MD
1364 top = n;
1365 if (!n)
1366 goto nospace;
1367
1368 if (!m_dup_pkthdr(n, m, how))
1369 goto nospace;
1370 n->m_len = m->m_len;
1371 if (m->m_flags & M_EXT) {
c3ef87ca 1372 KKASSERT((n->m_flags & M_EXT) == 0);
984263bc 1373 n->m_data = m->m_data;
7b6f875f 1374 m->m_ext.ext_ref(m->m_ext.ext_arg);
984263bc 1375 n->m_ext = m->m_ext;
b542cd49 1376 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
984263bc
MD
1377 } else {
1378 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
1379 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
1380 }
1381
1382 m = m->m_next;
1383 while (m) {
7b6f875f 1384 o = m_get(how, m->m_type);
984263bc
MD
1385 if (!o)
1386 goto nospace;
1387
1388 n->m_next = o;
1389 n = n->m_next;
1390
1391 n->m_len = m->m_len;
1392 if (m->m_flags & M_EXT) {
c3ef87ca 1393 KKASSERT((n->m_flags & M_EXT) == 0);
984263bc 1394 n->m_data = m->m_data;
7b6f875f 1395 m->m_ext.ext_ref(m->m_ext.ext_arg);
984263bc 1396 n->m_ext = m->m_ext;
b542cd49 1397 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
984263bc
MD
1398 } else {
1399 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
1400 }
1401
1402 m = m->m_next;
1403 }
1404 return top;
1405nospace:
1406 m_freem(top);
4c1e2509 1407 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
7b6f875f 1408 return (NULL);
984263bc
MD
1409}
1410
1411/*
1412 * Copy data from an mbuf chain starting "off" bytes from the beginning,
1413 * continuing for "len" bytes, into the indicated buffer.
1414 */
1415void
8a3125c6 1416m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
984263bc 1417{
1fd87d54 1418 unsigned count;
984263bc
MD
1419
1420 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
1421 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
1422 while (off > 0) {
1423 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
1424 if (off < m->m_len)
1425 break;
1426 off -= m->m_len;
1427 m = m->m_next;
1428 }
1429 while (len > 0) {
1430 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
1431 count = min(m->m_len - off, len);
1432 bcopy(mtod(m, caddr_t) + off, cp, count);
1433 len -= count;
1434 cp += count;
1435 off = 0;
1436 m = m->m_next;
1437 }
1438}
1439
1440/*
1441 * Copy a packet header mbuf chain into a completely new chain, including
1442 * copying any mbuf clusters. Use this instead of m_copypacket() when
1443 * you need a writable copy of an mbuf chain.
1444 */
1445struct mbuf *
8a3125c6 1446m_dup(struct mbuf *m, int how)
984263bc
MD
1447{
1448 struct mbuf **p, *top = NULL;
1449 int remain, moff, nsize;
1450
1451 /* Sanity check */
1452 if (m == NULL)
50503f0f 1453 return (NULL);
5e2195bf 1454 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __func__));
984263bc
MD
1455
1456 /* While there's more data, get a new mbuf, tack it on, and fill it */
1457 remain = m->m_pkthdr.len;
1458 moff = 0;
1459 p = &top;
1460 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
1461 struct mbuf *n;
1462
1463 /* Get the next new mbuf */
50503f0f
JH
1464 n = m_getl(remain, how, m->m_type, top == NULL ? M_PKTHDR : 0,
1465 &nsize);
984263bc
MD
1466 if (n == NULL)
1467 goto nospace;
50503f0f 1468 if (top == NULL)
984263bc 1469 if (!m_dup_pkthdr(n, m, how))
50503f0f 1470 goto nospace0;
984263bc
MD
1471
1472 /* Link it into the new chain */
1473 *p = n;
1474 p = &n->m_next;
1475
1476 /* Copy data from original mbuf(s) into new mbuf */
50503f0f 1477 n->m_len = 0;
984263bc
MD
1478 while (n->m_len < nsize && m != NULL) {
1479 int chunk = min(nsize - n->m_len, m->m_len - moff);
1480
1481 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
1482 moff += chunk;
1483 n->m_len += chunk;
1484 remain -= chunk;
1485 if (moff == m->m_len) {
1486 m = m->m_next;
1487 moff = 0;
1488 }
1489 }
1490
1491 /* Check correct total mbuf length */
1492 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
50503f0f 1493 ("%s: bogus m_pkthdr.len", __func__));
984263bc
MD
1494 }
1495 return (top);
1496
1497nospace:
1498 m_freem(top);
50503f0f 1499nospace0:
4c1e2509 1500 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
50503f0f 1501 return (NULL);
984263bc
MD
1502}
1503
3bf6fec3
MD
1504/*
1505 * Copy the non-packet mbuf data chain into a new set of mbufs, including
1506 * copying any mbuf clusters. This is typically used to realign a data
1507 * chain by nfs_realign().
1508 *
1509 * The original chain is left intact. how should be MB_WAIT or MB_DONTWAIT
1510 * and NULL can be returned if MB_DONTWAIT is passed.
1511 *
1512 * Be careful to use cluster mbufs, a large mbuf chain converted to non
1513 * cluster mbufs can exhaust our supply of mbufs.
1514 */
1515struct mbuf *
1516m_dup_data(struct mbuf *m, int how)
1517{
1518 struct mbuf **p, *n, *top = NULL;
1519 int mlen, moff, chunk, gsize, nsize;
1520
1521 /*
1522 * Degenerate case
1523 */
1524 if (m == NULL)
1525 return (NULL);
1526
1527 /*
1528 * Optimize the mbuf allocation but do not get too carried away.
1529 */
1530 if (m->m_next || m->m_len > MLEN)
94eaee9a
JT
1531 if (m->m_flags & M_EXT && m->m_ext.ext_size == MCLBYTES)
1532 gsize = MCLBYTES;
1533 else
1534 gsize = MJUMPAGESIZE;
3bf6fec3
MD
1535 else
1536 gsize = MLEN;
1537
1538 /* Chain control */
1539 p = &top;
1540 n = NULL;
1541 nsize = 0;
1542
1543 /*
1544 * Scan the mbuf chain until nothing is left, the new mbuf chain
1545 * will be allocated on the fly as needed.
1546 */
1547 while (m) {
1548 mlen = m->m_len;
1549 moff = 0;
1550
1551 while (mlen) {
1552 KKASSERT(m->m_type == MT_DATA);
1553 if (n == NULL) {
1554 n = m_getl(gsize, how, MT_DATA, 0, &nsize);
1555 n->m_len = 0;
1556 if (n == NULL)
1557 goto nospace;
1558 *p = n;
1559 p = &n->m_next;
1560 }
1561 chunk = imin(mlen, nsize);
1562 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
1563 mlen -= chunk;
1564 moff += chunk;
1565 n->m_len += chunk;
1566 nsize -= chunk;
1567 if (nsize == 0)
1568 n = NULL;
1569 }
1570 m = m->m_next;
1571 }
1572 *p = NULL;
1573 return(top);
1574nospace:
1575 *p = NULL;
1576 m_freem(top);
1577 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
1578 return (NULL);
1579}
1580
984263bc
MD
1581/*
1582 * Concatenate mbuf chain n to m.
1583 * Both chains must be of the same type (e.g. MT_DATA).
1584 * Any m_pkthdr is not updated.
1585 */
1586void
8a3125c6 1587m_cat(struct mbuf *m, struct mbuf *n)
984263bc 1588{
50503f0f 1589 m = m_last(m);
984263bc
MD
1590 while (n) {
1591 if (m->m_flags & M_EXT ||
1592 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
1593 /* just join the two chains */
1594 m->m_next = n;
1595 return;
1596 }
1597 /* splat the data from one into the other */
1598 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1599 (u_int)n->m_len);
1600 m->m_len += n->m_len;
1601 n = m_free(n);
1602 }
1603}
1604
1605void
8a3125c6 1606m_adj(struct mbuf *mp, int req_len)
984263bc 1607{
1fd87d54
RG
1608 int len = req_len;
1609 struct mbuf *m;
1610 int count;
984263bc
MD
1611
1612 if ((m = mp) == NULL)
1613 return;
1614 if (len >= 0) {
1615 /*
1616 * Trim from head.
1617 */
1618 while (m != NULL && len > 0) {
1619 if (m->m_len <= len) {
1620 len -= m->m_len;
1621 m->m_len = 0;
1622 m = m->m_next;
1623 } else {
1624 m->m_len -= len;
1625 m->m_data += len;
1626 len = 0;
1627 }
1628 }
1629 m = mp;
1630 if (mp->m_flags & M_PKTHDR)
1631 m->m_pkthdr.len -= (req_len - len);
1632 } else {
1633 /*
1634 * Trim from tail. Scan the mbuf chain,
1635 * calculating its length and finding the last mbuf.
1636 * If the adjustment only affects this mbuf, then just
1637 * adjust and return. Otherwise, rescan and truncate
1638 * after the remaining size.
1639 */
1640 len = -len;
1641 count = 0;
1642 for (;;) {
1643 count += m->m_len;
60233e58 1644 if (m->m_next == NULL)
984263bc
MD
1645 break;
1646 m = m->m_next;
1647 }
1648 if (m->m_len >= len) {
1649 m->m_len -= len;
1650 if (mp->m_flags & M_PKTHDR)
1651 mp->m_pkthdr.len -= len;
1652 return;
1653 }
1654 count -= len;
1655 if (count < 0)
1656 count = 0;
1657 /*
1658 * Correct length for chain is "count".
1659 * Find the mbuf with last data, adjust its length,
1660 * and toss data from remaining mbufs on chain.
1661 */
1662 m = mp;
1663 if (m->m_flags & M_PKTHDR)
1664 m->m_pkthdr.len = count;
1665 for (; m; m = m->m_next) {
1666 if (m->m_len >= count) {
1667 m->m_len = count;
1668 break;
1669 }
1670 count -= m->m_len;
1671 }
1672 while (m->m_next)
1673 (m = m->m_next) ->m_len = 0;
1674 }
1675}
1676
a3768f58
RP
1677/*
1678 * Set the m_data pointer of a newly-allocated mbuf
1679 * to place an object of the specified size at the
1680 * end of the mbuf, longword aligned.
1681 */
1682void
1683m_align(struct mbuf *m, int len)
1684{
1685 int adjust;
1686
1687 if (m->m_flags & M_EXT)
1688 adjust = m->m_ext.ext_size - len;
1689 else if (m->m_flags & M_PKTHDR)
1690 adjust = MHLEN - len;
1691 else
1692 adjust = MLEN - len;
1693 m->m_data += adjust &~ (sizeof(long)-1);
1694}
1695
984263bc 1696/*
7b6f875f 1697 * Rearrange an mbuf chain so that len bytes are contiguous
9e4465af
MD
1698 * and in the data area of an mbuf (so that mtod will work for a structure
1699 * of size len). Returns the resulting mbuf chain on success, frees it and
1700 * returns null on failure. If there is room, it will add up to
1701 * max_protohdr-len extra bytes to the contiguous region in an attempt to
1702 * avoid being called next time.
984263bc 1703 */
984263bc 1704struct mbuf *
8a3125c6 1705m_pullup(struct mbuf *n, int len)
984263bc 1706{
1fd87d54
RG
1707 struct mbuf *m;
1708 int count;
984263bc
MD
1709 int space;
1710
1711 /*
1712 * If first mbuf has no cluster, and has room for len bytes
1713 * without shifting current data, pullup into it,
1714 * otherwise allocate a new mbuf to prepend to the chain.
1715 */
7b6f875f
JH
1716 if (!(n->m_flags & M_EXT) &&
1717 n->m_data + len < &n->m_dat[MLEN] &&
1718 n->m_next) {
984263bc
MD
1719 if (n->m_len >= len)
1720 return (n);
1721 m = n;
1722 n = n->m_next;
1723 len -= m->m_len;
1724 } else {
1725 if (len > MHLEN)
1726 goto bad;
c3ef87ca
MD
1727 if (n->m_flags & M_PKTHDR)
1728 m = m_gethdr(MB_DONTWAIT, n->m_type);
1729 else
1730 m = m_get(MB_DONTWAIT, n->m_type);
7b6f875f 1731 if (m == NULL)
984263bc
MD
1732 goto bad;
1733 m->m_len = 0;
1734 if (n->m_flags & M_PKTHDR)
1735 M_MOVE_PKTHDR(m, n);
1736 }
1737 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1738 do {
1739 count = min(min(max(len, max_protohdr), space), n->m_len);
1740 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1741 (unsigned)count);
1742 len -= count;
1743 m->m_len += count;
1744 n->m_len -= count;
1745 space -= count;
1746 if (n->m_len)
1747 n->m_data += count;
1748 else
1749 n = m_free(n);
1750 } while (len > 0 && n);
1751 if (len > 0) {
7b6f875f 1752 m_free(m);
984263bc
MD
1753 goto bad;
1754 }
1755 m->m_next = n;
1756 return (m);
1757bad:
1758 m_freem(n);
4c1e2509 1759 atomic_add_long_nonlocked(&mbstat[mycpu->gd_cpuid].m_mcfail, 1);
7b6f875f 1760 return (NULL);
984263bc
MD
1761}
1762
1763/*
1764 * Partition an mbuf chain in two pieces, returning the tail --
1765 * all but the first len0 bytes. In case of failure, it returns NULL and
1766 * attempts to restore the chain to its original state.
1767 *
1768 * Note that the resulting mbufs might be read-only, because the new
1769 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1770 * the "breaking point" happens to lie within a cluster mbuf. Use the
1771 * M_WRITABLE() macro to check for this case.
1772 */
1773struct mbuf *
8a3125c6 1774m_split(struct mbuf *m0, int len0, int wait)
984263bc 1775{
1fd87d54 1776 struct mbuf *m, *n;
984263bc
MD
1777 unsigned len = len0, remain;
1778
1779 for (m = m0; m && len > m->m_len; m = m->m_next)
1780 len -= m->m_len;
7b6f875f
JH
1781 if (m == NULL)
1782 return (NULL);
984263bc
MD
1783 remain = m->m_len - len;
1784 if (m0->m_flags & M_PKTHDR) {
7b6f875f
JH
1785 n = m_gethdr(wait, m0->m_type);
1786 if (n == NULL)
1787 return (NULL);
984263bc
MD
1788 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1789 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1790 m0->m_pkthdr.len = len0;
1791 if (m->m_flags & M_EXT)
1792 goto extpacket;
1793 if (remain > MHLEN) {
1794 /* m can't be the lead packet */
1795 MH_ALIGN(n, 0);
1796 n->m_next = m_split(m, len, wait);
7b6f875f
JH
1797 if (n->m_next == NULL) {
1798 m_free(n);
1799 return (NULL);
984263bc
MD
1800 } else {
1801 n->m_len = 0;
1802 return (n);
1803 }
1804 } else
1805 MH_ALIGN(n, remain);
1806 } else if (remain == 0) {
1807 n = m->m_next;
1808 m->m_next = 0;
1809 return (n);
1810 } else {
7b6f875f
JH
1811 n = m_get(wait, m->m_type);
1812 if (n == NULL)
1813 return (NULL);
984263bc
MD
1814 M_ALIGN(n, remain);
1815 }
1816extpacket:
1817 if (m->m_flags & M_EXT) {
c3ef87ca 1818 KKASSERT((n->m_flags & M_EXT) == 0);
984263bc 1819 n->m_data = m->m_data + len;
7b6f875f 1820 m->m_ext.ext_ref(m->m_ext.ext_arg);
7eccf245 1821 n->m_ext = m->m_ext;
b542cd49 1822 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER);
984263bc
MD
1823 } else {
1824 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1825 }
1826 n->m_len = remain;
1827 m->m_len = len;
1828 n->m_next = m->m_next;
1829 m->m_next = 0;
1830 return (n);
1831}
50503f0f 1832
984263bc
MD
1833/*
1834 * Routine to copy from device local memory into mbufs.
50503f0f 1835 * Note: "offset" is ill-defined and always called as 0, so ignore it.
984263bc
MD
1836 */
1837struct mbuf *
50503f0f
JH
1838m_devget(char *buf, int len, int offset, struct ifnet *ifp,
1839 void (*copy)(volatile const void *from, volatile void *to, size_t length))
984263bc 1840{
50503f0f
JH
1841 struct mbuf *m, *mfirst = NULL, **mtail;
1842 int nsize, flags;
1843
1844 if (copy == NULL)
1845 copy = bcopy;
1846 mtail = &mfirst;
1847 flags = M_PKTHDR;
1848
1849 while (len > 0) {
1850 m = m_getl(len, MB_DONTWAIT, MT_DATA, flags, &nsize);
1851 if (m == NULL) {
1852 m_freem(mfirst);
1853 return (NULL);
984263bc 1854 }
50503f0f
JH
1855 m->m_len = min(len, nsize);
1856
1857 if (flags & M_PKTHDR) {
1858 if (len + max_linkhdr <= nsize)
1859 m->m_data += max_linkhdr;
1860 m->m_pkthdr.rcvif = ifp;
1861 m->m_pkthdr.len = len;
1862 flags = 0;
984263bc 1863 }
50503f0f
JH
1864
1865 copy(buf, m->m_data, (unsigned)m->m_len);
1866 buf += m->m_len;
1867 len -= m->m_len;
1868 *mtail = m;
1869 mtail = &m->m_next;
984263bc 1870 }
50503f0f
JH
1871
1872 return (mfirst);
984263bc
MD
1873}
1874
cf12ba3c
SZ
1875/*
1876 * Routine to pad mbuf to the specified length 'padto'.
1877 */
1878int
1879m_devpad(struct mbuf *m, int padto)
1880{
1881 struct mbuf *last = NULL;
1882 int padlen;
1883
1884 if (padto <= m->m_pkthdr.len)
1885 return 0;
1886
1887 padlen = padto - m->m_pkthdr.len;
1888
1889 /* if there's only the packet-header and we can pad there, use it. */
1890 if (m->m_pkthdr.len == m->m_len && M_TRAILINGSPACE(m) >= padlen) {
1891 last = m;
1892 } else {
1893 /*
1894 * Walk packet chain to find last mbuf. We will either
1895 * pad there, or append a new mbuf and pad it
1896 */
1897 for (last = m; last->m_next != NULL; last = last->m_next)
1898 ; /* EMPTY */
1899
1900 /* `last' now points to last in chain. */
1901 if (M_TRAILINGSPACE(last) < padlen) {
1902 struct mbuf *n;
1903
1904 /* Allocate new empty mbuf, pad it. Compact later. */
1905 MGET(n, MB_DONTWAIT, MT_DATA);
1906 if (n == NULL)
1907 return ENOBUFS;
1908 n->m_len = 0;
1909 last->m_next = n;
1910 last = n;
1911 }
1912 }
1913 KKASSERT(M_TRAILINGSPACE(last) >= padlen);
1914 KKASSERT(M_WRITABLE(last));
1915
1916 /* Now zero the pad area */
1917 bzero(mtod(last, char *) + last->m_len, padlen);
1918 last->m_len += padlen;
1919 m->m_pkthdr.len += padlen;
1920 return 0;
1921}
1922
984263bc
MD
1923/*
1924 * Copy data from a buffer back into the indicated mbuf chain,
1925 * starting "off" bytes from the beginning, extending the mbuf
1926 * chain if necessary.
1927 */
1928void
8a3125c6 1929m_copyback(struct mbuf *m0, int off, int len, caddr_t cp)
984263bc 1930{
1fd87d54
RG
1931 int mlen;
1932 struct mbuf *m = m0, *n;
984263bc
MD
1933 int totlen = 0;
1934
7b6f875f 1935 if (m0 == NULL)
984263bc
MD
1936 return;
1937 while (off > (mlen = m->m_len)) {
1938 off -= mlen;
1939 totlen += mlen;
7b6f875f 1940 if (m->m_next == NULL) {
74f1caca 1941 n = m_getclr(MB_DONTWAIT, m->m_type);
7b6f875f 1942 if (n == NULL)
984263bc
MD
1943 goto out;
1944 n->m_len = min(MLEN, len + off);
1945 m->m_next = n;
1946 }
1947 m = m->m_next;
1948 }
1949 while (len > 0) {
1950 mlen = min (m->m_len - off, len);
1951 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen);
1952 cp += mlen;
1953 len -= mlen;
1954 mlen += off;
1955 off = 0;
1956 totlen += mlen;
1957 if (len == 0)
1958 break;
7b6f875f 1959 if (m->m_next == NULL) {
74f1caca 1960 n = m_get(MB_DONTWAIT, m->m_type);
7b6f875f 1961 if (n == NULL)
984263bc
MD
1962 break;
1963 n->m_len = min(MLEN, len);
1964 m->m_next = n;
1965 }
1966 m = m->m_next;
1967 }
1968out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1969 m->m_pkthdr.len = totlen;
1970}
1971
bf2cc98c
RP
1972/*
1973 * Append the specified data to the indicated mbuf chain,
1974 * Extend the mbuf chain if the new data does not fit in
1975 * existing space.
1976 *
1977 * Return 1 if able to complete the job; otherwise 0.
1978 */
1979int
1980m_append(struct mbuf *m0, int len, c_caddr_t cp)
1981{
1982 struct mbuf *m, *n;
1983 int remainder, space;
1984
1985 for (m = m0; m->m_next != NULL; m = m->m_next)
1986 ;
1987 remainder = len;
1988 space = M_TRAILINGSPACE(m);
1989 if (space > 0) {
1990 /*
1991 * Copy into available space.
1992 */
1993 if (space > remainder)
1994 space = remainder;
1995 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1996 m->m_len += space;
1997 cp += space, remainder -= space;
1998 }
1999 while (remainder > 0) {
2000 /*
2001 * Allocate a new mbuf; could check space
2002 * and allocate a cluster instead.
2003 */
2004 n = m_get(MB_DONTWAIT, m->m_type);
2005 if (n == NULL)
2006 break;
2007 n->m_len = min(MLEN, remainder);
2008 bcopy(cp, mtod(n, caddr_t), n->m_len);
2009 cp += n->m_len, remainder -= n->m_len;
2010 m->m_next = n;
2011 m = n;
2012 }
2013 if (m0->m_flags & M_PKTHDR)
2014 m0->m_pkthdr.len += len - remainder;
2015 return (remainder == 0);
2016}
2017
920c9f10
AH
2018/*
2019 * Apply function f to the data in an mbuf chain starting "off" bytes from
2020 * the beginning, continuing for "len" bytes.
2021 */
2022int
2023m_apply(struct mbuf *m, int off, int len,
2024 int (*f)(void *, void *, u_int), void *arg)
2025{
2026 u_int count;
2027 int rval;
2028
2029 KASSERT(off >= 0, ("m_apply, negative off %d", off));
2030 KASSERT(len >= 0, ("m_apply, negative len %d", len));
2031 while (off > 0) {
2032 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
2033 if (off < m->m_len)
2034 break;
2035 off -= m->m_len;
2036 m = m->m_next;
2037 }
2038 while (len > 0) {
2039 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
2040 count = min(m->m_len - off, len);
2041 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
2042 if (rval)
2043 return (rval);
2044 len -= count;
2045 off = 0;
2046 m = m->m_next;
2047 }
2048 return (0);
2049}
2050
2051/*
2052 * Return a pointer to mbuf/offset of location in mbuf chain.
2053 */
2054struct mbuf *
2055m_getptr(struct mbuf *m, int loc, int *off)
2056{
2057
2058 while (loc >= 0) {
2059 /* Normal end of search. */
2060 if (m->m_len > loc) {
2061 *off = loc;
2062 return (m);
2063 } else {
2064 loc -= m->m_len;
2065 if (m->m_next == NULL) {
2066 if (loc == 0) {
2067 /* Point at the end of valid data. */
2068 *off = m->m_len;
2069 return (m);
2070 }
2071 return (NULL);
2072 }
2073 m = m->m_next;
2074 }
2075 }
2076 return (NULL);
2077}
2078
984263bc
MD
2079void
2080m_print(const struct mbuf *m)
2081{
2082 int len;
2083 const struct mbuf *m2;
2084
2085 len = m->m_pkthdr.len;
2086 m2 = m;
2087 while (len) {
6ea70f76 2088 kprintf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-");
984263bc
MD
2089 len -= m2->m_len;
2090 m2 = m2->m_next;
2091 }
2092 return;
2093}
2094
2095/*
2096 * "Move" mbuf pkthdr from "from" to "to".
2097 * "from" must have M_PKTHDR set, and "to" must be empty.
2098 */
2099void
2100m_move_pkthdr(struct mbuf *to, struct mbuf *from)
2101{
e0d05288 2102 KASSERT((to->m_flags & M_PKTHDR), ("m_move_pkthdr: not packet header"));
984263bc 2103
77e294a1 2104 to->m_flags |= from->m_flags & M_COPYFLAGS;
984263bc
MD
2105 to->m_pkthdr = from->m_pkthdr; /* especially tags */
2106 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
984263bc
MD
2107}
2108
2109/*
2110 * Duplicate "from"'s mbuf pkthdr in "to".
2111 * "from" must have M_PKTHDR set, and "to" must be empty.
2112 * In particular, this does a deep copy of the packet tags.
2113 */
2114int
f15db79e 2115m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how)
984263bc 2116{
7f3602fe
JH
2117 KASSERT((to->m_flags & M_PKTHDR), ("m_dup_pkthdr: not packet header"));
2118
4bac35fc 2119 to->m_flags = (from->m_flags & M_COPYFLAGS) |
c4da22e4 2120 (to->m_flags & ~M_COPYFLAGS);
984263bc
MD
2121 to->m_pkthdr = from->m_pkthdr;
2122 SLIST_INIT(&to->m_pkthdr.tags);
2123 return (m_tag_copy_chain(to, from, how));
2124}
2125
2126/*
2127 * Defragment a mbuf chain, returning the shortest possible
2128 * chain of mbufs and clusters. If allocation fails and
2129 * this cannot be completed, NULL will be returned, but
2130 * the passed in chain will be unchanged. Upon success,
2131 * the original chain will be freed, and the new chain
2132 * will be returned.
2133 *
2134 * If a non-packet header is passed in, the original
2135 * mbuf (chain?) will be returned unharmed.
c8f5127a
JS
2136 *
2137 * m_defrag_nofree doesn't free the passed in mbuf.
984263bc
MD
2138 */
2139struct mbuf *
2140m_defrag(struct mbuf *m0, int how)
c8f5127a
JS
2141{
2142 struct mbuf *m_new;
2143
2144 if ((m_new = m_defrag_nofree(m0, how)) == NULL)
2145 return (NULL);
2146 if (m_new != m0)
2147 m_freem(m0);
2148 return (m_new);
2149}
2150
2151struct mbuf *
2152m_defrag_nofree(struct mbuf *m0, int how)
984263bc
MD
2153{
2154 struct mbuf *m_new = NULL, *m_final = NULL;
61721e90 2155 int progress = 0, length, nsize;
984263bc
MD
2156
2157 if (!(m0->m_flags & M_PKTHDR))
2158 return (m0);
2159
2160#ifdef MBUF_STRESS_TEST
2161 if (m_defragrandomfailures) {
0ced1954 2162 int temp = karc4random() & 0xff;
984263bc
MD
2163 if (temp == 0xba)
2164 goto nospace;
2165 }
2166#endif
2167
61721e90 2168 m_final = m_getl(m0->m_pkthdr.len, how, MT_DATA, M_PKTHDR, &nsize);
984263bc
MD
2169 if (m_final == NULL)
2170 goto nospace;
61721e90 2171 m_final->m_len = 0; /* in case m0->m_pkthdr.len is zero */
984263bc 2172
3641b7ca 2173 if (m_dup_pkthdr(m_final, m0, how) == 0)
984263bc
MD
2174 goto nospace;
2175
2176 m_new = m_final;
2177
2178 while (progress < m0->m_pkthdr.len) {
2179 length = m0->m_pkthdr.len - progress;
2180 if (length > MCLBYTES)
2181 length = MCLBYTES;
2182
2183 if (m_new == NULL) {
61721e90 2184 m_new = m_getl(length, how, MT_DATA, 0, &nsize);
984263bc
MD
2185 if (m_new == NULL)
2186 goto nospace;
2187 }
2188
2189 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
2190 progress += length;
2191 m_new->m_len = length;
2192 if (m_new != m_final)
2193 m_cat(m_final, m_new);
2194 m_new = NULL;
2195 }
2196 if (m0->m_next == NULL)
2197 m_defraguseless++;
984263bc 2198 m_defragpackets++;
c8f5127a
JS
2199 m_defragbytes += m_final->m_pkthdr.len;
2200 return (m_final);
984263bc
MD
2201nospace:
2202 m_defragfailure++;
2203 if (m_new)
2204 m_free(m_new);
61721e90 2205 m_freem(m_final);
984263bc
MD
2206 return (NULL);
2207}
0c33f36d
JH
2208
2209/*
2210 * Move data from uio into mbufs.
0c33f36d
JH
2211 */
2212struct mbuf *
e12241e1 2213m_uiomove(struct uio *uio)
0c33f36d 2214{
0c33f36d 2215 struct mbuf *m; /* current working mbuf */
e12241e1
JH
2216 struct mbuf *head = NULL; /* result mbuf chain */
2217 struct mbuf **mp = &head;
e54488bb
MD
2218 int flags = M_PKTHDR;
2219 int nsize;
2220 int error;
2221 int resid;
0c33f36d 2222
0c33f36d 2223 do {
e54488bb
MD
2224 if (uio->uio_resid > INT_MAX)
2225 resid = INT_MAX;
2226 else
2227 resid = (int)uio->uio_resid;
e12241e1 2228 m = m_getl(resid, MB_WAIT, MT_DATA, flags, &nsize);
61721e90
JH
2229 if (flags) {
2230 m->m_pkthdr.len = 0;
2231 /* Leave room for protocol headers. */
2232 if (resid < MHLEN)
2233 MH_ALIGN(m, resid);
2234 flags = 0;
0c33f36d 2235 }
e54488bb 2236 m->m_len = imin(nsize, resid);
61721e90 2237 error = uiomove(mtod(m, caddr_t), m->m_len, uio);
0c33f36d
JH
2238 if (error) {
2239 m_free(m);
2240 goto failed;
2241 }
0c33f36d
JH
2242 *mp = m;
2243 mp = &m->m_next;
61721e90 2244 head->m_pkthdr.len += m->m_len;
e54488bb 2245 } while (uio->uio_resid > 0);
0c33f36d
JH
2246
2247 return (head);
2248
2249failed:
61721e90 2250 m_freem(head);
0c33f36d
JH
2251 return (NULL);
2252}
df80f2ea 2253
50503f0f
JH
2254struct mbuf *
2255m_last(struct mbuf *m)
2256{
2257 while (m->m_next)
2258 m = m->m_next;
2259 return (m);
2260}
2261
df80f2ea
JH
2262/*
2263 * Return the number of bytes in an mbuf chain.
2264 * If lastm is not NULL, also return the last mbuf.
2265 */
2266u_int
2267m_lengthm(struct mbuf *m, struct mbuf **lastm)
2268{
2269 u_int len = 0;
2270 struct mbuf *prev = m;
2271
2272 while (m) {
2273 len += m->m_len;
2274 prev = m;
2275 m = m->m_next;
2276 }
2277 if (lastm != NULL)
2278 *lastm = prev;
2279 return (len);
2280}
2281
2282/*
2283 * Like m_lengthm(), except also keep track of mbuf usage.
2284 */
2285u_int
2286m_countm(struct mbuf *m, struct mbuf **lastm, u_int *pmbcnt)
2287{
2288 u_int len = 0, mbcnt = 0;
2289 struct mbuf *prev = m;
2290
2291 while (m) {
2292 len += m->m_len;
2293 mbcnt += MSIZE;
2294 if (m->m_flags & M_EXT)
2295 mbcnt += m->m_ext.ext_size;
2296 prev = m;
2297 m = m->m_next;
2298 }
2299 if (lastm != NULL)
2300 *lastm = prev;
2301 *pmbcnt = mbcnt;
2302 return (len);
2303}