2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include "opt_param.h"
38 #include "opt_mbuf_stress_test.h"
39 #include "opt_mbuf_profiling.h"
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/limits.h>
46 #include <sys/malloc.h>
48 #include <sys/sysctl.h>
49 #include <sys/domain.h>
50 #include <sys/protosw.h>
52 #include <sys/vmmeter.h>
55 #include <vm/vm_pageout.h>
56 #include <vm/vm_page.h>
58 SDT_PROBE_DEFINE5_XLATE(sdt, , , m__init,
59 "struct mbuf *", "mbufinfo_t *",
60 "uint32_t", "uint32_t",
61 "uint16_t", "uint16_t",
62 "uint32_t", "uint32_t",
63 "uint32_t", "uint32_t");
65 SDT_PROBE_DEFINE3_XLATE(sdt, , , m__gethdr,
66 "uint32_t", "uint32_t",
67 "uint16_t", "uint16_t",
68 "struct mbuf *", "mbufinfo_t *");
70 SDT_PROBE_DEFINE3_XLATE(sdt, , , m__get,
71 "uint32_t", "uint32_t",
72 "uint16_t", "uint16_t",
73 "struct mbuf *", "mbufinfo_t *");
75 SDT_PROBE_DEFINE4_XLATE(sdt, , , m__getcl,
76 "uint32_t", "uint32_t",
77 "uint16_t", "uint16_t",
78 "uint32_t", "uint32_t",
79 "struct mbuf *", "mbufinfo_t *");
81 SDT_PROBE_DEFINE3_XLATE(sdt, , , m__clget,
82 "struct mbuf *", "mbufinfo_t *",
83 "uint32_t", "uint32_t",
84 "uint32_t", "uint32_t");
86 SDT_PROBE_DEFINE4_XLATE(sdt, , , m__cljget,
87 "struct mbuf *", "mbufinfo_t *",
88 "uint32_t", "uint32_t",
89 "uint32_t", "uint32_t",
92 SDT_PROBE_DEFINE(sdt, , , m__cljset);
94 SDT_PROBE_DEFINE1_XLATE(sdt, , , m__free,
95 "struct mbuf *", "mbufinfo_t *");
97 SDT_PROBE_DEFINE1_XLATE(sdt, , , m__freem,
98 "struct mbuf *", "mbufinfo_t *");
100 #include <security/mac/mac_framework.h>
106 #ifdef MBUF_STRESS_TEST
111 int m_defragrandomfailures;
115 * sysctl(8) exported objects
117 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD,
118 &max_linkhdr, 0, "Size of largest link layer header");
119 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD,
120 &max_protohdr, 0, "Size of largest protocol layer header");
121 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD,
122 &max_hdr, 0, "Size of largest link plus protocol header");
123 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RD,
124 &max_datalen, 0, "Minimum space left in mbuf after max_hdr");
125 #ifdef MBUF_STRESS_TEST
126 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
127 &m_defragpackets, 0, "");
128 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
129 &m_defragbytes, 0, "");
130 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
131 &m_defraguseless, 0, "");
132 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
133 &m_defragfailure, 0, "");
134 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
135 &m_defragrandomfailures, 0, "");
139 * Ensure the correct size of various mbuf parameters. It could be off due
140 * to compiler-induced padding and alignment artifacts.
142 CTASSERT(MSIZE - offsetof(struct mbuf, m_dat) == MLEN);
143 CTASSERT(MSIZE - offsetof(struct mbuf, m_pktdat) == MHLEN);
146 * mbuf data storage should be 64-bit aligned regardless of architectural
147 * pointer size; check this is the case with and without a packet header.
149 CTASSERT(offsetof(struct mbuf, m_dat) % 8 == 0);
150 CTASSERT(offsetof(struct mbuf, m_pktdat) % 8 == 0);
153 * While the specific values here don't matter too much (i.e., +/- a few
154 * words), we do want to ensure that changes to these values are carefully
155 * reasoned about and properly documented. This is especially the case as
156 * network-protocol and device-driver modules encode these layouts, and must
157 * be recompiled if the structures change. Check these values at compile time
158 * against the ones documented in comments in mbuf.h.
160 * NB: Possibly they should be documented there via #define's and not just
163 #if defined(__LP64__)
164 CTASSERT(offsetof(struct mbuf, m_dat) == 32);
165 CTASSERT(sizeof(struct pkthdr) == 56);
166 CTASSERT(sizeof(struct m_ext) == 160);
168 CTASSERT(offsetof(struct mbuf, m_dat) == 24);
169 CTASSERT(sizeof(struct pkthdr) == 48);
170 #if defined(__powerpc__) && defined(BOOKE)
171 /* PowerPC booke has 64-bit physical pointers. */
172 CTASSERT(sizeof(struct m_ext) == 184);
174 CTASSERT(sizeof(struct m_ext) == 180);
179 * Assert that the queue(3) macros produce code of the same size as an old
180 * plain pointer does.
183 static struct mbuf __used m_assertbuf;
184 CTASSERT(sizeof(m_assertbuf.m_slist) == sizeof(m_assertbuf.m_next));
185 CTASSERT(sizeof(m_assertbuf.m_stailq) == sizeof(m_assertbuf.m_next));
186 CTASSERT(sizeof(m_assertbuf.m_slistpkt) == sizeof(m_assertbuf.m_nextpkt));
187 CTASSERT(sizeof(m_assertbuf.m_stailqpkt) == sizeof(m_assertbuf.m_nextpkt));
191 * Attach the cluster from *m to *n, set up m_ext in *n
192 * and bump the refcount of the cluster.
195 mb_dupcl(struct mbuf *n, struct mbuf *m)
197 volatile u_int *refcnt;
199 KASSERT(m->m_flags & (M_EXT|M_EXTPG),
200 ("%s: M_EXT|M_EXTPG not set on %p", __func__, m));
201 KASSERT(!(n->m_flags & (M_EXT|M_EXTPG)),
202 ("%s: M_EXT|M_EXTPG set on %p", __func__, n));
205 * Cache access optimization.
207 * o Regular M_EXT storage doesn't need full copy of m_ext, since
208 * the holder of the 'ext_count' is responsible to carry the free
209 * routine and its arguments.
210 * o M_EXTPG data is split between main part of mbuf and m_ext, the
211 * main part is copied in full, the m_ext part is similar to M_EXT.
212 * o EXT_EXTREF, where 'ext_cnt' doesn't point into mbuf at all, is
213 * special - it needs full copy of m_ext into each mbuf, since any
214 * copy could end up as the last to free.
216 if (m->m_flags & M_EXTPG) {
217 bcopy(&m->m_epg_startcopy, &n->m_epg_startcopy,
218 __rangeof(struct mbuf, m_epg_startcopy, m_epg_endcopy));
219 bcopy(&m->m_ext, &n->m_ext, m_epg_ext_copylen);
220 } else if (m->m_ext.ext_type == EXT_EXTREF)
221 bcopy(&m->m_ext, &n->m_ext, sizeof(struct m_ext));
223 bcopy(&m->m_ext, &n->m_ext, m_ext_copylen);
225 n->m_flags |= m->m_flags & (M_RDONLY | M_EXT | M_EXTPG);
227 /* See if this is the mbuf that holds the embedded refcount. */
228 if (m->m_ext.ext_flags & EXT_FLAG_EMBREF) {
229 refcnt = n->m_ext.ext_cnt = &m->m_ext.ext_count;
230 n->m_ext.ext_flags &= ~EXT_FLAG_EMBREF;
232 KASSERT(m->m_ext.ext_cnt != NULL,
233 ("%s: no refcounting pointer on %p", __func__, m));
234 refcnt = m->m_ext.ext_cnt;
240 atomic_add_int(refcnt, 1);
244 m_demote_pkthdr(struct mbuf *m)
249 m_tag_delete_chain(m, NULL);
250 m->m_flags &= ~M_PKTHDR;
251 bzero(&m->m_pkthdr, sizeof(struct pkthdr));
255 * Clean up mbuf (chain) from any tags and packet headers.
256 * If "all" is set then the first mbuf in the chain will be
260 m_demote(struct mbuf *m0, int all, int flags)
264 for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) {
265 KASSERT(m->m_nextpkt == NULL, ("%s: m_nextpkt in m %p, m0 %p",
267 if (m->m_flags & M_PKTHDR)
269 m->m_flags = m->m_flags & (M_EXT | M_RDONLY | M_NOFREE |
275 * Sanity checks on mbuf (chain) for use in KASSERT() and general
277 * Returns 0 or panics when bad and 1 on all tests passed.
278 * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they
282 m_sanity(struct mbuf *m0, int sanitize)
289 #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m)
291 #define M_SANITY_ACTION(s) printf("mbuf %p: " s, m)
294 for (m = m0; m != NULL; m = m->m_next) {
296 * Basic pointer checks. If any of these fails then some
297 * unrelated kernel memory before or after us is trashed.
298 * No way to recover from that.
302 if ((caddr_t)m->m_data < a)
303 M_SANITY_ACTION("m_data outside mbuf data range left");
304 if ((caddr_t)m->m_data > b)
305 M_SANITY_ACTION("m_data outside mbuf data range right");
306 if ((caddr_t)m->m_data + m->m_len > b)
307 M_SANITY_ACTION("m_data + m_len exeeds mbuf space");
309 /* m->m_nextpkt may only be set on first mbuf in chain. */
310 if (m != m0 && m->m_nextpkt != NULL) {
312 m_freem(m->m_nextpkt);
313 m->m_nextpkt = (struct mbuf *)0xDEADC0DE;
315 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf");
318 /* packet length (not mbuf length!) calculation */
319 if (m0->m_flags & M_PKTHDR)
322 /* m_tags may only be attached to first mbuf in chain. */
323 if (m != m0 && m->m_flags & M_PKTHDR &&
324 !SLIST_EMPTY(&m->m_pkthdr.tags)) {
326 m_tag_delete_chain(m, NULL);
327 /* put in 0xDEADC0DE perhaps? */
329 M_SANITY_ACTION("m_tags on in-chain mbuf");
332 /* M_PKTHDR may only be set on first mbuf in chain */
333 if (m != m0 && m->m_flags & M_PKTHDR) {
335 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
336 m->m_flags &= ~M_PKTHDR;
337 /* put in 0xDEADCODE and leave hdr flag in */
339 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf");
343 if (pktlen && pktlen != m->m_pkthdr.len) {
347 M_SANITY_ACTION("m_pkthdr.len != mbuf chain length");
351 #undef M_SANITY_ACTION
355 * Non-inlined part of m_init().
358 m_pkthdr_init(struct mbuf *m, int how)
363 m->m_data = m->m_pktdat;
364 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
366 m->m_pkthdr.numa_domain = M_NODOM;
369 /* If the label init fails, fail the alloc */
370 error = mac_mbuf_init(m, how);
379 * "Move" mbuf pkthdr from "from" to "to".
380 * "from" must have M_PKTHDR set, and "to" must be empty.
383 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
387 /* see below for why these are not enabled */
389 /* Note: with MAC, this may not be a good assertion. */
390 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
391 ("m_move_pkthdr: to has tags"));
395 * XXXMAC: It could be this should also occur for non-MAC?
397 if (to->m_flags & M_PKTHDR)
398 m_tag_delete_chain(to, NULL);
400 to->m_flags = (from->m_flags & M_COPYFLAGS) |
401 (to->m_flags & (M_EXT | M_EXTPG));
402 if ((to->m_flags & M_EXT) == 0)
403 to->m_data = to->m_pktdat;
404 to->m_pkthdr = from->m_pkthdr; /* especially tags */
405 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
406 from->m_flags &= ~M_PKTHDR;
407 if (from->m_pkthdr.csum_flags & CSUM_SND_TAG) {
408 from->m_pkthdr.csum_flags &= ~CSUM_SND_TAG;
409 from->m_pkthdr.snd_tag = NULL;
414 * Duplicate "from"'s mbuf pkthdr in "to".
415 * "from" must have M_PKTHDR set, and "to" must be empty.
416 * In particular, this does a deep copy of the packet tags.
419 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how)
424 * The mbuf allocator only initializes the pkthdr
425 * when the mbuf is allocated with m_gethdr(). Many users
426 * (e.g. m_copy*, m_prepend) use m_get() and then
427 * smash the pkthdr as needed causing these
428 * assertions to trip. For now just disable them.
431 /* Note: with MAC, this may not be a good assertion. */
432 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
434 MBUF_CHECKSLEEP(how);
436 if (to->m_flags & M_PKTHDR)
437 m_tag_delete_chain(to, NULL);
439 to->m_flags = (from->m_flags & M_COPYFLAGS) |
440 (to->m_flags & (M_EXT | M_EXTPG));
441 if ((to->m_flags & M_EXT) == 0)
442 to->m_data = to->m_pktdat;
443 to->m_pkthdr = from->m_pkthdr;
444 if (from->m_pkthdr.csum_flags & CSUM_SND_TAG)
445 m_snd_tag_ref(from->m_pkthdr.snd_tag);
446 SLIST_INIT(&to->m_pkthdr.tags);
447 return (m_tag_copy_chain(to, from, how));
451 * Lesser-used path for M_PREPEND:
452 * allocate new mbuf to prepend to chain,
456 m_prepend(struct mbuf *m, int len, int how)
460 if (m->m_flags & M_PKTHDR)
461 mn = m_gethdr(how, m->m_type);
463 mn = m_get(how, m->m_type);
468 if (m->m_flags & M_PKTHDR)
469 m_move_pkthdr(mn, m);
479 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
480 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
481 * The wait parameter is a choice of M_WAITOK/M_NOWAIT from caller.
482 * Note that the copy is read-only, because clusters are not copied,
483 * only their reference counts are incremented.
486 m_copym(struct mbuf *m, int off0, int len, int wait)
488 struct mbuf *n, **np;
493 KASSERT(off >= 0, ("m_copym, negative off %d", off));
494 KASSERT(len >= 0, ("m_copym, negative len %d", len));
495 MBUF_CHECKSLEEP(wait);
496 if (off == 0 && m->m_flags & M_PKTHDR)
499 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
509 KASSERT(len == M_COPYALL,
510 ("m_copym, length > size of mbuf chain"));
514 n = m_gethdr(wait, m->m_type);
516 n = m_get(wait, m->m_type);
521 if (!m_dup_pkthdr(n, m, wait))
523 if (len == M_COPYALL)
524 n->m_pkthdr.len -= off0;
526 n->m_pkthdr.len = len;
529 n->m_len = min(len, m->m_len - off);
530 if (m->m_flags & (M_EXT|M_EXTPG)) {
531 n->m_data = m->m_data + off;
534 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
536 if (len != M_COPYALL)
550 * Copy an entire packet, including header (which must be present).
551 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
552 * Note that the copy is read-only, because clusters are not copied,
553 * only their reference counts are incremented.
554 * Preserve alignment of the first mbuf so if the creator has left
555 * some room at the beginning (e.g. for inserting protocol headers)
556 * the copies still have the room available.
559 m_copypacket(struct mbuf *m, int how)
561 struct mbuf *top, *n, *o;
563 MBUF_CHECKSLEEP(how);
564 n = m_get(how, m->m_type);
569 if (!m_dup_pkthdr(n, m, how))
572 if (m->m_flags & (M_EXT|M_EXTPG)) {
573 n->m_data = m->m_data;
576 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
577 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
582 o = m_get(how, m->m_type);
590 if (m->m_flags & (M_EXT|M_EXTPG)) {
591 n->m_data = m->m_data;
594 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
606 m_copyfromunmapped(const struct mbuf *m, int off, int len, caddr_t cp)
612 KASSERT(off >= 0, ("m_copyfromunmapped: negative off %d", off));
613 KASSERT(len >= 0, ("m_copyfromunmapped: negative len %d", len));
614 KASSERT(off < m->m_len,
615 ("m_copyfromunmapped: len exceeds mbuf length"));
620 uio.uio_segflg = UIO_SYSSPACE;
623 uio.uio_rw = UIO_READ;
624 error = m_unmappedtouio(m, off, &uio, len);
625 KASSERT(error == 0, ("m_unmappedtouio failed: off %d, len %d", off,
630 * Copy data from an mbuf chain starting "off" bytes from the beginning,
631 * continuing for "len" bytes, into the indicated buffer.
634 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
638 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
639 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
641 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
648 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
649 count = min(m->m_len - off, len);
650 if ((m->m_flags & M_EXTPG) != 0)
651 m_copyfromunmapped(m, off, count, cp);
653 bcopy(mtod(m, caddr_t) + off, cp, count);
662 * Copy a packet header mbuf chain into a completely new chain, including
663 * copying any mbuf clusters. Use this instead of m_copypacket() when
664 * you need a writable copy of an mbuf chain.
667 m_dup(const struct mbuf *m, int how)
669 struct mbuf **p, *top = NULL;
670 int remain, moff, nsize;
672 MBUF_CHECKSLEEP(how);
678 /* While there's more data, get a new mbuf, tack it on, and fill it */
679 remain = m->m_pkthdr.len;
682 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
685 /* Get the next new mbuf */
686 if (remain >= MINCLSIZE) {
687 n = m_getcl(how, m->m_type, 0);
690 n = m_get(how, m->m_type);
696 if (top == NULL) { /* First one, must be PKTHDR */
697 if (!m_dup_pkthdr(n, m, how)) {
701 if ((n->m_flags & M_EXT) == 0)
703 n->m_flags &= ~M_RDONLY;
707 /* Link it into the new chain */
711 /* Copy data from original mbuf(s) into new mbuf */
712 while (n->m_len < nsize && m != NULL) {
713 int chunk = min(nsize - n->m_len, m->m_len - moff);
715 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
719 if (moff == m->m_len) {
725 /* Check correct total mbuf length */
726 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
727 ("%s: bogus m_pkthdr.len", __func__));
737 * Concatenate mbuf chain n to m.
738 * Both chains must be of the same type (e.g. MT_DATA).
739 * Any m_pkthdr is not updated.
742 m_cat(struct mbuf *m, struct mbuf *n)
747 if (!M_WRITABLE(m) ||
748 (n->m_flags & M_EXTPG) != 0 ||
749 M_TRAILINGSPACE(m) < n->m_len) {
750 /* just join the two chains */
754 /* splat the data from one into the other */
755 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
757 m->m_len += n->m_len;
763 * Concatenate two pkthdr mbuf chains.
766 m_catpkt(struct mbuf *m, struct mbuf *n)
772 m->m_pkthdr.len += n->m_pkthdr.len;
779 m_adj(struct mbuf *mp, int req_len)
785 if ((m = mp) == NULL)
791 while (m != NULL && len > 0) {
792 if (m->m_len <= len) {
802 if (mp->m_flags & M_PKTHDR)
803 mp->m_pkthdr.len -= (req_len - len);
806 * Trim from tail. Scan the mbuf chain,
807 * calculating its length and finding the last mbuf.
808 * If the adjustment only affects this mbuf, then just
809 * adjust and return. Otherwise, rescan and truncate
810 * after the remaining size.
816 if (m->m_next == (struct mbuf *)0)
820 if (m->m_len >= len) {
822 if (mp->m_flags & M_PKTHDR)
823 mp->m_pkthdr.len -= len;
830 * Correct length for chain is "count".
831 * Find the mbuf with last data, adjust its length,
832 * and toss data from remaining mbufs on chain.
835 if (m->m_flags & M_PKTHDR)
836 m->m_pkthdr.len = count;
837 for (; m; m = m->m_next) {
838 if (m->m_len >= count) {
840 if (m->m_next != NULL) {
852 * Rearange an mbuf chain so that len bytes are contiguous
853 * and in the data area of an mbuf (so that mtod will work
854 * for a structure of size len). Returns the resulting
855 * mbuf chain on success, frees it and returns null on failure.
856 * If there is room, it will add up to max_protohdr-len extra bytes to the
857 * contiguous region in an attempt to avoid being called next time.
860 m_pullup(struct mbuf *n, int len)
866 KASSERT((n->m_flags & M_EXTPG) == 0,
867 ("%s: unmapped mbuf %p", __func__, n));
870 * If first mbuf has no cluster, and has room for len bytes
871 * without shifting current data, pullup into it,
872 * otherwise allocate a new mbuf to prepend to the chain.
874 if ((n->m_flags & M_EXT) == 0 &&
875 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
884 m = m_get(M_NOWAIT, n->m_type);
887 if (n->m_flags & M_PKTHDR)
890 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
892 count = min(min(max(len, max_protohdr), space), n->m_len);
893 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
903 } while (len > 0 && n);
916 * Like m_pullup(), except a new mbuf is always allocated, and we allow
917 * the amount of empty space before the data in the new mbuf to be specified
918 * (in the event that the caller expects to prepend later).
921 m_copyup(struct mbuf *n, int len, int dstoff)
926 if (len > (MHLEN - dstoff))
928 m = m_get(M_NOWAIT, n->m_type);
931 if (n->m_flags & M_PKTHDR)
934 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
936 count = min(min(max(len, max_protohdr), space), n->m_len);
937 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
947 } while (len > 0 && n);
960 * Partition an mbuf chain in two pieces, returning the tail --
961 * all but the first len0 bytes. In case of failure, it returns NULL and
962 * attempts to restore the chain to its original state.
964 * Note that the resulting mbufs might be read-only, because the new
965 * mbuf can end up sharing an mbuf cluster with the original mbuf if
966 * the "breaking point" happens to lie within a cluster mbuf. Use the
967 * M_WRITABLE() macro to check for this case.
970 m_split(struct mbuf *m0, int len0, int wait)
973 u_int len = len0, remain;
975 MBUF_CHECKSLEEP(wait);
976 for (m = m0; m && len > m->m_len; m = m->m_next)
980 remain = m->m_len - len;
981 if (m0->m_flags & M_PKTHDR && remain == 0) {
982 n = m_gethdr(wait, m0->m_type);
985 n->m_next = m->m_next;
987 if (m0->m_pkthdr.csum_flags & CSUM_SND_TAG) {
988 n->m_pkthdr.snd_tag =
989 m_snd_tag_ref(m0->m_pkthdr.snd_tag);
990 n->m_pkthdr.csum_flags |= CSUM_SND_TAG;
992 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
993 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
994 m0->m_pkthdr.len = len0;
996 } else if (m0->m_flags & M_PKTHDR) {
997 n = m_gethdr(wait, m0->m_type);
1000 if (m0->m_pkthdr.csum_flags & CSUM_SND_TAG) {
1001 n->m_pkthdr.snd_tag =
1002 m_snd_tag_ref(m0->m_pkthdr.snd_tag);
1003 n->m_pkthdr.csum_flags |= CSUM_SND_TAG;
1005 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1006 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1007 m0->m_pkthdr.len = len0;
1008 if (m->m_flags & (M_EXT|M_EXTPG))
1010 if (remain > MHLEN) {
1011 /* m can't be the lead packet */
1013 n->m_next = m_split(m, len, wait);
1014 if (n->m_next == NULL) {
1023 } else if (remain == 0) {
1028 n = m_get(wait, m->m_type);
1034 if (m->m_flags & (M_EXT|M_EXTPG)) {
1035 n->m_data = m->m_data + len;
1038 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1042 n->m_next = m->m_next;
1047 * Routine to copy from device local memory into mbufs.
1048 * Note that `off' argument is offset into first mbuf of target chain from
1049 * which to begin copying the data to.
1052 m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
1053 void (*copy)(char *from, caddr_t to, u_int len))
1056 struct mbuf *top = NULL, **mp = ⊤
1059 if (off < 0 || off > MHLEN)
1062 while (totlen > 0) {
1063 if (top == NULL) { /* First one, must be PKTHDR */
1064 if (totlen + off >= MINCLSIZE) {
1065 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1068 m = m_gethdr(M_NOWAIT, MT_DATA);
1071 /* Place initial small packet/header at end of mbuf */
1072 if (m && totlen + off + max_linkhdr <= MHLEN) {
1073 m->m_data += max_linkhdr;
1079 m->m_pkthdr.rcvif = ifp;
1080 m->m_pkthdr.len = totlen;
1082 if (totlen + off >= MINCLSIZE) {
1083 m = m_getcl(M_NOWAIT, MT_DATA, 0);
1086 m = m_get(M_NOWAIT, MT_DATA);
1099 m->m_len = len = min(totlen, len);
1101 copy(buf, mtod(m, caddr_t), (u_int)len);
1103 bcopy(buf, mtod(m, caddr_t), (u_int)len);
1113 * Copy data from a buffer back into the indicated mbuf chain,
1114 * starting "off" bytes from the beginning, extending the mbuf
1115 * chain if necessary.
1118 m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
1121 struct mbuf *m = m0, *n;
1126 while (off > (mlen = m->m_len)) {
1129 if (m->m_next == NULL) {
1130 n = m_get(M_NOWAIT, m->m_type);
1133 bzero(mtod(n, caddr_t), MLEN);
1134 n->m_len = min(MLEN, len + off);
1140 if (m->m_next == NULL && (len > m->m_len - off)) {
1141 m->m_len += min(len - (m->m_len - off),
1142 M_TRAILINGSPACE(m));
1144 mlen = min (m->m_len - off, len);
1145 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
1153 if (m->m_next == NULL) {
1154 n = m_get(M_NOWAIT, m->m_type);
1157 n->m_len = min(MLEN, len);
1162 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1163 m->m_pkthdr.len = totlen;
1167 * Append the specified data to the indicated mbuf chain,
1168 * Extend the mbuf chain if the new data does not fit in
1171 * Return 1 if able to complete the job; otherwise 0.
1174 m_append(struct mbuf *m0, int len, c_caddr_t cp)
1177 int remainder, space;
1179 for (m = m0; m->m_next != NULL; m = m->m_next)
1182 space = M_TRAILINGSPACE(m);
1185 * Copy into available space.
1187 if (space > remainder)
1189 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1191 cp += space, remainder -= space;
1193 while (remainder > 0) {
1195 * Allocate a new mbuf; could check space
1196 * and allocate a cluster instead.
1198 n = m_get(M_NOWAIT, m->m_type);
1201 n->m_len = min(MLEN, remainder);
1202 bcopy(cp, mtod(n, caddr_t), n->m_len);
1203 cp += n->m_len, remainder -= n->m_len;
1207 if (m0->m_flags & M_PKTHDR)
1208 m0->m_pkthdr.len += len - remainder;
1209 return (remainder == 0);
1213 * Apply function f to the data in an mbuf chain starting "off" bytes from
1214 * the beginning, continuing for "len" bytes.
1217 m_apply(struct mbuf *m, int off, int len,
1218 int (*f)(void *, void *, u_int), void *arg)
1223 KASSERT(off >= 0, ("m_apply, negative off %d", off));
1224 KASSERT(len >= 0, ("m_apply, negative len %d", len));
1226 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1233 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1234 count = min(m->m_len - off, len);
1235 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1246 * Return a pointer to mbuf/offset of location in mbuf chain.
1249 m_getptr(struct mbuf *m, int loc, int *off)
1253 /* Normal end of search. */
1254 if (m->m_len > loc) {
1259 if (m->m_next == NULL) {
1261 /* Point at the end of valid data. */
1274 m_print(const struct mbuf *m, int maxlen)
1278 const struct mbuf *m2;
1281 printf("mbuf: %p\n", m);
1285 if (m->m_flags & M_PKTHDR)
1286 len = m->m_pkthdr.len;
1290 while (m2 != NULL && (len == -1 || len)) {
1292 if (maxlen != -1 && pdata > maxlen)
1294 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
1295 m2->m_next, m2->m_flags, "\20\20freelist\17skipfw"
1296 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
1297 "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
1299 printf(", %*D\n", pdata, (u_char *)m2->m_data, "-");
1305 printf("%d bytes unaccounted for.\n", len);
1310 m_fixhdr(struct mbuf *m0)
1314 len = m_length(m0, NULL);
1315 m0->m_pkthdr.len = len;
1320 m_length(struct mbuf *m0, struct mbuf **last)
1326 for (m = m0; m != NULL; m = m->m_next) {
1328 if (m->m_next == NULL)
1337 * Defragment a mbuf chain, returning the shortest possible
1338 * chain of mbufs and clusters. If allocation fails and
1339 * this cannot be completed, NULL will be returned, but
1340 * the passed in chain will be unchanged. Upon success,
1341 * the original chain will be freed, and the new chain
1344 * If a non-packet header is passed in, the original
1345 * mbuf (chain?) will be returned unharmed.
1348 m_defrag(struct mbuf *m0, int how)
1350 struct mbuf *m_new = NULL, *m_final = NULL;
1351 int progress = 0, length;
1353 MBUF_CHECKSLEEP(how);
1354 if (!(m0->m_flags & M_PKTHDR))
1357 m_fixhdr(m0); /* Needed sanity check */
1359 #ifdef MBUF_STRESS_TEST
1360 if (m_defragrandomfailures) {
1361 int temp = arc4random() & 0xff;
1367 if (m0->m_pkthdr.len > MHLEN)
1368 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1370 m_final = m_gethdr(how, MT_DATA);
1372 if (m_final == NULL)
1375 if (m_dup_pkthdr(m_final, m0, how) == 0)
1380 while (progress < m0->m_pkthdr.len) {
1381 length = m0->m_pkthdr.len - progress;
1382 if (length > MCLBYTES)
1385 if (m_new == NULL) {
1387 m_new = m_getcl(how, MT_DATA, 0);
1389 m_new = m_get(how, MT_DATA);
1394 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1396 m_new->m_len = length;
1397 if (m_new != m_final)
1398 m_cat(m_final, m_new);
1401 #ifdef MBUF_STRESS_TEST
1402 if (m0->m_next == NULL)
1407 #ifdef MBUF_STRESS_TEST
1409 m_defragbytes += m0->m_pkthdr.len;
1413 #ifdef MBUF_STRESS_TEST
1422 * Return the number of fragments an mbuf will use. This is usually
1423 * used as a proxy for the number of scatter/gather elements needed by
1424 * a DMA engine to access an mbuf. In general mapped mbufs are
1425 * assumed to be backed by physically contiguous buffers that only
1426 * need a single fragment. Unmapped mbufs, on the other hand, can
1427 * span disjoint physical pages.
1430 frags_per_mbuf(struct mbuf *m)
1434 if ((m->m_flags & M_EXTPG) == 0)
1438 * The header and trailer are counted as a single fragment
1439 * each when present.
1441 * XXX: This overestimates the number of fragments by assuming
1442 * all the backing physical pages are disjoint.
1445 if (m->m_epg_hdrlen != 0)
1447 frags += m->m_epg_npgs;
1448 if (m->m_epg_trllen != 0)
1455 * Defragment an mbuf chain, returning at most maxfrags separate
1456 * mbufs+clusters. If this is not possible NULL is returned and
1457 * the original mbuf chain is left in its present (potentially
1458 * modified) state. We use two techniques: collapsing consecutive
1459 * mbufs and replacing consecutive mbufs by a cluster.
1461 * NB: this should really be named m_defrag but that name is taken
1464 m_collapse(struct mbuf *m0, int how, int maxfrags)
1466 struct mbuf *m, *n, *n2, **prev;
1470 * Calculate the current number of frags.
1473 for (m = m0; m != NULL; m = m->m_next)
1474 curfrags += frags_per_mbuf(m);
1476 * First, try to collapse mbufs. Note that we always collapse
1477 * towards the front so we don't need to deal with moving the
1478 * pkthdr. This may be suboptimal if the first mbuf has much
1479 * less data than the following.
1487 if (M_WRITABLE(m) &&
1488 n->m_len < M_TRAILINGSPACE(m)) {
1489 m_copydata(n, 0, n->m_len,
1490 mtod(m, char *) + m->m_len);
1491 m->m_len += n->m_len;
1492 m->m_next = n->m_next;
1493 curfrags -= frags_per_mbuf(n);
1495 if (curfrags <= maxfrags)
1500 KASSERT(maxfrags > 1,
1501 ("maxfrags %u, but normal collapse failed", maxfrags));
1503 * Collapse consecutive mbufs to a cluster.
1505 prev = &m0->m_next; /* NB: not the first mbuf */
1506 while ((n = *prev) != NULL) {
1507 if ((n2 = n->m_next) != NULL &&
1508 n->m_len + n2->m_len < MCLBYTES) {
1509 m = m_getcl(how, MT_DATA, 0);
1512 m_copydata(n, 0, n->m_len, mtod(m, char *));
1513 m_copydata(n2, 0, n2->m_len,
1514 mtod(m, char *) + n->m_len);
1515 m->m_len = n->m_len + n2->m_len;
1516 m->m_next = n2->m_next;
1518 curfrags += 1; /* For the new cluster */
1519 curfrags -= frags_per_mbuf(n);
1520 curfrags -= frags_per_mbuf(n2);
1523 if (curfrags <= maxfrags)
1526 * Still not there, try the normal collapse
1527 * again before we allocate another cluster.
1534 * No place where we can collapse to a cluster; punt.
1535 * This can occur if, for example, you request 2 frags
1536 * but the packet requires that both be clusters (we
1537 * never reallocate the first mbuf to avoid moving the
1544 #ifdef MBUF_STRESS_TEST
1547 * Fragment an mbuf chain. There's no reason you'd ever want to do
1548 * this in normal usage, but it's great for stress testing various
1551 * If fragmentation is not possible, the original chain will be
1554 * Possible length values:
1555 * 0 no fragmentation will occur
1556 * > 0 each fragment will be of the specified length
1557 * -1 each fragment will be the same random value in length
1558 * -2 each fragment's length will be entirely random
1559 * (Random values range from 1 to 256)
1562 m_fragment(struct mbuf *m0, int how, int length)
1564 struct mbuf *m_first, *m_last;
1565 int divisor = 255, progress = 0, fraglen;
1567 if (!(m0->m_flags & M_PKTHDR))
1570 if (length == 0 || length < -2)
1572 if (length > MCLBYTES)
1574 if (length < 0 && divisor > MCLBYTES)
1577 length = 1 + (arc4random() % divisor);
1581 m_fixhdr(m0); /* Needed sanity check */
1583 m_first = m_getcl(how, MT_DATA, M_PKTHDR);
1584 if (m_first == NULL)
1587 if (m_dup_pkthdr(m_first, m0, how) == 0)
1592 while (progress < m0->m_pkthdr.len) {
1594 fraglen = 1 + (arc4random() % divisor);
1595 if (fraglen > m0->m_pkthdr.len - progress)
1596 fraglen = m0->m_pkthdr.len - progress;
1598 if (progress != 0) {
1599 struct mbuf *m_new = m_getcl(how, MT_DATA, 0);
1603 m_last->m_next = m_new;
1607 m_copydata(m0, progress, fraglen, mtod(m_last, caddr_t));
1608 progress += fraglen;
1609 m_last->m_len = fraglen;
1617 /* Return the original chain on failure */
1624 * Free pages from mbuf_ext_pgs, assuming they were allocated via
1625 * vm_page_alloc() and aren't associated with any object. Complement
1626 * to allocator from m_uiotombuf_nomap().
1629 mb_free_mext_pgs(struct mbuf *m)
1634 for (int i = 0; i < m->m_epg_npgs; i++) {
1635 pg = PHYS_TO_VM_PAGE(m->m_epg_pa[i]);
1636 vm_page_unwire_noq(pg);
1641 static struct mbuf *
1642 m_uiotombuf_nomap(struct uio *uio, int how, int len, int maxseg, int flags)
1644 struct mbuf *m, *mb, *prev;
1645 vm_page_t pg_array[MBUF_PEXT_MAX_PGS];
1646 int error, length, i, needed;
1648 int pflags = malloc2vm_flags(how) | VM_ALLOC_NOOBJ | VM_ALLOC_NODUMP |
1652 * len can be zero or an arbitrary large value bound by
1653 * the total data supplied by the uio.
1656 total = MIN(uio->uio_resid, len);
1658 total = uio->uio_resid;
1661 maxseg = MBUF_PEXT_MAX_PGS * PAGE_SIZE;
1664 * Allocate the pages
1667 MPASS((flags & M_PKTHDR) == 0);
1669 mb = mb_alloc_ext_pgs(how, mb_free_mext_pgs);
1677 mb->m_epg_flags = EPG_FLAG_ANON;
1678 needed = length = MIN(maxseg, total);
1679 for (i = 0; needed > 0; i++, needed -= PAGE_SIZE) {
1681 pg_array[i] = vm_page_alloc(NULL, 0, pflags);
1682 if (pg_array[i] == NULL) {
1683 if (how & M_NOWAIT) {
1690 pg_array[i]->flags &= ~PG_ZERO;
1691 mb->m_epg_pa[i] = VM_PAGE_TO_PHYS(pg_array[i]);
1694 mb->m_epg_last_len = length - PAGE_SIZE * (mb->m_epg_npgs - 1);
1695 MBUF_EXT_PGS_ASSERT_SANITY(mb);
1697 error = uiomove_fromphys(pg_array, 0, length, uio);
1701 mb->m_ext.ext_size += PAGE_SIZE * mb->m_epg_npgs;
1702 if (flags & M_PKTHDR)
1703 m->m_pkthdr.len += length;
1713 * Copy the contents of uio into a properly sized mbuf chain.
1716 m_uiotombuf(struct uio *uio, int how, int len, int align, int flags)
1718 struct mbuf *m, *mb;
1723 if (flags & M_EXTPG)
1724 return (m_uiotombuf_nomap(uio, how, len, align, flags));
1727 * len can be zero or an arbitrary large value bound by
1728 * the total data supplied by the uio.
1731 total = (uio->uio_resid < len) ? uio->uio_resid : len;
1733 total = uio->uio_resid;
1736 * The smallest unit returned by m_getm2() is a single mbuf
1737 * with pkthdr. We can't align past it.
1743 * Give us the full allocation or nothing.
1744 * If len is zero return the smallest empty mbuf.
1746 m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags);
1751 /* Fill all mbufs with uio data and update header information. */
1752 for (mb = m; mb != NULL; mb = mb->m_next) {
1753 length = min(M_TRAILINGSPACE(mb), total - progress);
1755 error = uiomove(mtod(mb, void *), length, uio);
1763 if (flags & M_PKTHDR)
1764 m->m_pkthdr.len += length;
1766 KASSERT(progress == total, ("%s: progress != total", __func__));
1772 * Copy data from an unmapped mbuf into a uio limited by len if set.
1775 m_unmappedtouio(const struct mbuf *m, int m_off, struct uio *uio, int len)
1778 int error, i, off, pglen, pgoff, seglen, segoff;
1783 /* Skip over any data removed from the front. */
1784 off = mtod(m, vm_offset_t);
1787 if (m->m_epg_hdrlen != 0) {
1788 if (off >= m->m_epg_hdrlen) {
1789 off -= m->m_epg_hdrlen;
1791 seglen = m->m_epg_hdrlen - off;
1793 seglen = min(seglen, len);
1796 error = uiomove(__DECONST(void *,
1797 &m->m_epg_hdr[segoff]), seglen, uio);
1800 pgoff = m->m_epg_1st_off;
1801 for (i = 0; i < m->m_epg_npgs && error == 0 && len > 0; i++) {
1802 pglen = m_epg_pagelen(m, i, pgoff);
1808 seglen = pglen - off;
1809 segoff = pgoff + off;
1811 seglen = min(seglen, len);
1813 pg = PHYS_TO_VM_PAGE(m->m_epg_pa[i]);
1814 error = uiomove_fromphys(&pg, segoff, seglen, uio);
1817 if (len != 0 && error == 0) {
1818 KASSERT((off + len) <= m->m_epg_trllen,
1819 ("off + len > trail (%d + %d > %d, m_off = %d)", off, len,
1820 m->m_epg_trllen, m_off));
1821 error = uiomove(__DECONST(void *, &m->m_epg_trail[off]),
1828 * Copy an mbuf chain into a uio limited by len if set.
1831 m_mbuftouio(struct uio *uio, const struct mbuf *m, int len)
1833 int error, length, total;
1837 total = min(uio->uio_resid, len);
1839 total = uio->uio_resid;
1841 /* Fill the uio with data from the mbufs. */
1842 for (; m != NULL; m = m->m_next) {
1843 length = min(m->m_len, total - progress);
1845 if ((m->m_flags & M_EXTPG) != 0)
1846 error = m_unmappedtouio(m, 0, uio, length);
1848 error = uiomove(mtod(m, void *), length, uio);
1859 * Create a writable copy of the mbuf chain. While doing this
1860 * we compact the chain with a goal of producing a chain with
1861 * at most two mbufs. The second mbuf in this chain is likely
1862 * to be a cluster. The primary purpose of this work is to create
1863 * a writable packet for encryption, compression, etc. The
1864 * secondary goal is to linearize the data so the data can be
1865 * passed to crypto hardware in the most efficient manner possible.
1868 m_unshare(struct mbuf *m0, int how)
1870 struct mbuf *m, *mprev;
1871 struct mbuf *n, *mfirst, *mlast;
1875 for (m = m0; m != NULL; m = mprev->m_next) {
1877 * Regular mbufs are ignored unless there's a cluster
1878 * in front of it that we can use to coalesce. We do
1879 * the latter mainly so later clusters can be coalesced
1880 * also w/o having to handle them specially (i.e. convert
1881 * mbuf+cluster -> cluster). This optimization is heavily
1882 * influenced by the assumption that we're running over
1883 * Ethernet where MCLBYTES is large enough that the max
1884 * packet size will permit lots of coalescing into a
1885 * single cluster. This in turn permits efficient
1886 * crypto operations, especially when using hardware.
1888 if ((m->m_flags & M_EXT) == 0) {
1889 if (mprev && (mprev->m_flags & M_EXT) &&
1890 m->m_len <= M_TRAILINGSPACE(mprev)) {
1891 /* XXX: this ignores mbuf types */
1892 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1893 mtod(m, caddr_t), m->m_len);
1894 mprev->m_len += m->m_len;
1895 mprev->m_next = m->m_next; /* unlink from chain */
1896 m_free(m); /* reclaim mbuf */
1903 * Writable mbufs are left alone (for now).
1905 if (M_WRITABLE(m)) {
1911 * Not writable, replace with a copy or coalesce with
1912 * the previous mbuf if possible (since we have to copy
1913 * it anyway, we try to reduce the number of mbufs and
1914 * clusters so that future work is easier).
1916 KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags));
1917 /* NB: we only coalesce into a cluster or larger */
1918 if (mprev != NULL && (mprev->m_flags & M_EXT) &&
1919 m->m_len <= M_TRAILINGSPACE(mprev)) {
1920 /* XXX: this ignores mbuf types */
1921 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1922 mtod(m, caddr_t), m->m_len);
1923 mprev->m_len += m->m_len;
1924 mprev->m_next = m->m_next; /* unlink from chain */
1925 m_free(m); /* reclaim mbuf */
1930 * Allocate new space to hold the copy and copy the data.
1931 * We deal with jumbo mbufs (i.e. m_len > MCLBYTES) by
1932 * splitting them into clusters. We could just malloc a
1933 * buffer and make it external but too many device drivers
1934 * don't know how to break up the non-contiguous memory when
1937 n = m_getcl(how, m->m_type, m->m_flags & M_COPYFLAGS);
1942 if (m->m_flags & M_PKTHDR) {
1943 KASSERT(mprev == NULL, ("%s: m0 %p, m %p has M_PKTHDR",
1945 m_move_pkthdr(n, m);
1952 int cc = min(len, MCLBYTES);
1953 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc);
1959 newipsecstat.ips_clcopied++;
1967 n = m_getcl(how, m->m_type, m->m_flags & M_COPYFLAGS);
1974 n->m_next = m->m_next;
1976 m0 = mfirst; /* new head of chain */
1978 mprev->m_next = mfirst; /* replace old mbuf */
1979 m_free(m); /* release old mbuf */
1985 #ifdef MBUF_PROFILING
1987 #define MP_BUCKETS 32 /* don't just change this as things may overflow.*/
1988 struct mbufprofile {
1989 uintmax_t wasted[MP_BUCKETS];
1990 uintmax_t used[MP_BUCKETS];
1991 uintmax_t segments[MP_BUCKETS];
1994 #define MP_MAXDIGITS 21 /* strlen("16,000,000,000,000,000,000") == 21 */
1995 #define MP_NUMLINES 6
1996 #define MP_NUMSPERLINE 16
1997 #define MP_EXTRABYTES 64 /* > strlen("used:\nwasted:\nsegments:\n") */
1998 /* work out max space needed and add a bit of spare space too */
1999 #define MP_MAXLINE ((MP_MAXDIGITS+1) * MP_NUMSPERLINE)
2000 #define MP_BUFSIZE ((MP_MAXLINE * MP_NUMLINES) + 1 + MP_EXTRABYTES)
2002 char mbprofbuf[MP_BUFSIZE];
2005 m_profile(struct mbuf *m)
2014 if (m->m_flags & M_EXT) {
2015 wasted += MHLEN - sizeof(m->m_ext) +
2016 m->m_ext.ext_size - m->m_len;
2018 if (m->m_flags & M_PKTHDR)
2019 wasted += MHLEN - m->m_len;
2021 wasted += MLEN - m->m_len;
2025 /* be paranoid.. it helps */
2026 if (segments > MP_BUCKETS - 1)
2027 segments = MP_BUCKETS - 1;
2030 if (wasted > 100000)
2032 /* store in the appropriate bucket */
2033 /* don't bother locking. if it's slightly off, so what? */
2034 mbprof.segments[segments]++;
2035 mbprof.used[fls(used)]++;
2036 mbprof.wasted[fls(wasted)]++;
2040 mbprof_textify(void)
2046 p = &mbprof.wasted[0];
2048 offset = snprintf(c, MP_MAXLINE + 10,
2050 "%ju %ju %ju %ju %ju %ju %ju %ju "
2051 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2052 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2053 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2055 p = &mbprof.wasted[16];
2057 offset = snprintf(c, MP_MAXLINE,
2058 "%ju %ju %ju %ju %ju %ju %ju %ju "
2059 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2060 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2061 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2063 p = &mbprof.used[0];
2065 offset = snprintf(c, MP_MAXLINE + 10,
2067 "%ju %ju %ju %ju %ju %ju %ju %ju "
2068 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2069 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2070 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2072 p = &mbprof.used[16];
2074 offset = snprintf(c, MP_MAXLINE,
2075 "%ju %ju %ju %ju %ju %ju %ju %ju "
2076 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2077 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2078 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2080 p = &mbprof.segments[0];
2082 offset = snprintf(c, MP_MAXLINE + 10,
2084 "%ju %ju %ju %ju %ju %ju %ju %ju "
2085 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2086 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2087 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2089 p = &mbprof.segments[16];
2091 offset = snprintf(c, MP_MAXLINE,
2092 "%ju %ju %ju %ju %ju %ju %ju %ju "
2093 "%ju %ju %ju %ju %ju %ju %ju %jju",
2094 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2095 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2100 mbprof_handler(SYSCTL_HANDLER_ARGS)
2105 error = SYSCTL_OUT(req, mbprofbuf, strlen(mbprofbuf) + 1);
2110 mbprof_clr_handler(SYSCTL_HANDLER_ARGS)
2115 error = sysctl_handle_int(oidp, &clear, 0, req);
2116 if (error || !req->newptr)
2120 bzero(&mbprof, sizeof(mbprof));
2126 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofile,
2127 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, NULL, 0,
2128 mbprof_handler, "A",
2129 "mbuf profiling statistics");
2131 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofileclr,
2132 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, NULL, 0,
2133 mbprof_clr_handler, "I",
2134 "clear mbuf profiling statistics");