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 CTASSERT(sizeof(struct m_ext) == 180);
174 * Assert that the queue(3) macros produce code of the same size as an old
175 * plain pointer does.
178 static struct mbuf __used m_assertbuf;
179 CTASSERT(sizeof(m_assertbuf.m_slist) == sizeof(m_assertbuf.m_next));
180 CTASSERT(sizeof(m_assertbuf.m_stailq) == sizeof(m_assertbuf.m_next));
181 CTASSERT(sizeof(m_assertbuf.m_slistpkt) == sizeof(m_assertbuf.m_nextpkt));
182 CTASSERT(sizeof(m_assertbuf.m_stailqpkt) == sizeof(m_assertbuf.m_nextpkt));
186 * Attach the cluster from *m to *n, set up m_ext in *n
187 * and bump the refcount of the cluster.
190 mb_dupcl(struct mbuf *n, struct mbuf *m)
192 volatile u_int *refcnt;
194 KASSERT(m->m_flags & M_EXT, ("%s: M_EXT not set on %p", __func__, m));
195 KASSERT(!(n->m_flags & M_EXT), ("%s: M_EXT set on %p", __func__, n));
198 * Cache access optimization.
200 * o Regular M_EXT storage doesn't need full copy of m_ext, since
201 * the holder of the 'ext_count' is responsible to carry the free
202 * routine and its arguments.
203 * o EXT_PGS data is split between main part of mbuf and m_ext, the
204 * main part is copied in full, the m_ext part is similar to M_EXT.
205 * o EXT_EXTREF, where 'ext_cnt' doesn't point into mbuf at all, is
206 * special - it needs full copy of m_ext into each mbuf, since any
207 * copy could end up as the last to free.
209 switch (m->m_ext.ext_type) {
211 bcopy(&m->m_ext, &n->m_ext, m_epg_copylen);
212 bcopy(&m->m_ext_pgs, &n->m_ext_pgs,
213 sizeof(struct mbuf_ext_pgs));
216 bcopy(&m->m_ext, &n->m_ext, sizeof(struct m_ext));
219 bcopy(&m->m_ext, &n->m_ext, m_ext_copylen);
223 n->m_flags |= m->m_flags & (M_RDONLY | M_NOMAP);
225 /* See if this is the mbuf that holds the embedded refcount. */
226 if (m->m_ext.ext_flags & EXT_FLAG_EMBREF) {
227 refcnt = n->m_ext.ext_cnt = &m->m_ext.ext_count;
228 n->m_ext.ext_flags &= ~EXT_FLAG_EMBREF;
230 KASSERT(m->m_ext.ext_cnt != NULL,
231 ("%s: no refcounting pointer on %p", __func__, m));
232 refcnt = m->m_ext.ext_cnt;
238 atomic_add_int(refcnt, 1);
242 m_demote_pkthdr(struct mbuf *m)
247 m_tag_delete_chain(m, NULL);
248 m->m_flags &= ~M_PKTHDR;
249 bzero(&m->m_pkthdr, sizeof(struct pkthdr));
253 * Clean up mbuf (chain) from any tags and packet headers.
254 * If "all" is set then the first mbuf in the chain will be
258 m_demote(struct mbuf *m0, int all, int flags)
262 for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) {
263 KASSERT(m->m_nextpkt == NULL, ("%s: m_nextpkt in m %p, m0 %p",
265 if (m->m_flags & M_PKTHDR)
267 m->m_flags = m->m_flags & (M_EXT | M_RDONLY | M_NOFREE |
273 * Sanity checks on mbuf (chain) for use in KASSERT() and general
275 * Returns 0 or panics when bad and 1 on all tests passed.
276 * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they
280 m_sanity(struct mbuf *m0, int sanitize)
287 #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m)
289 #define M_SANITY_ACTION(s) printf("mbuf %p: " s, m)
292 for (m = m0; m != NULL; m = m->m_next) {
294 * Basic pointer checks. If any of these fails then some
295 * unrelated kernel memory before or after us is trashed.
296 * No way to recover from that.
300 if ((caddr_t)m->m_data < a)
301 M_SANITY_ACTION("m_data outside mbuf data range left");
302 if ((caddr_t)m->m_data > b)
303 M_SANITY_ACTION("m_data outside mbuf data range right");
304 if ((caddr_t)m->m_data + m->m_len > b)
305 M_SANITY_ACTION("m_data + m_len exeeds mbuf space");
307 /* m->m_nextpkt may only be set on first mbuf in chain. */
308 if (m != m0 && m->m_nextpkt != NULL) {
310 m_freem(m->m_nextpkt);
311 m->m_nextpkt = (struct mbuf *)0xDEADC0DE;
313 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf");
316 /* packet length (not mbuf length!) calculation */
317 if (m0->m_flags & M_PKTHDR)
320 /* m_tags may only be attached to first mbuf in chain. */
321 if (m != m0 && m->m_flags & M_PKTHDR &&
322 !SLIST_EMPTY(&m->m_pkthdr.tags)) {
324 m_tag_delete_chain(m, NULL);
325 /* put in 0xDEADC0DE perhaps? */
327 M_SANITY_ACTION("m_tags on in-chain mbuf");
330 /* M_PKTHDR may only be set on first mbuf in chain */
331 if (m != m0 && m->m_flags & M_PKTHDR) {
333 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
334 m->m_flags &= ~M_PKTHDR;
335 /* put in 0xDEADCODE and leave hdr flag in */
337 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf");
341 if (pktlen && pktlen != m->m_pkthdr.len) {
345 M_SANITY_ACTION("m_pkthdr.len != mbuf chain length");
349 #undef M_SANITY_ACTION
353 * Non-inlined part of m_init().
356 m_pkthdr_init(struct mbuf *m, int how)
361 m->m_data = m->m_pktdat;
362 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
364 m->m_pkthdr.numa_domain = M_NODOM;
367 /* If the label init fails, fail the alloc */
368 error = mac_mbuf_init(m, how);
377 * "Move" mbuf pkthdr from "from" to "to".
378 * "from" must have M_PKTHDR set, and "to" must be empty.
381 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
385 /* see below for why these are not enabled */
387 /* Note: with MAC, this may not be a good assertion. */
388 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
389 ("m_move_pkthdr: to has tags"));
393 * XXXMAC: It could be this should also occur for non-MAC?
395 if (to->m_flags & M_PKTHDR)
396 m_tag_delete_chain(to, NULL);
398 to->m_flags = (from->m_flags & M_COPYFLAGS) |
399 (to->m_flags & (M_EXT | M_NOMAP));
400 if ((to->m_flags & M_EXT) == 0)
401 to->m_data = to->m_pktdat;
402 to->m_pkthdr = from->m_pkthdr; /* especially tags */
403 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
404 from->m_flags &= ~M_PKTHDR;
405 if (from->m_pkthdr.csum_flags & CSUM_SND_TAG) {
406 from->m_pkthdr.csum_flags &= ~CSUM_SND_TAG;
407 from->m_pkthdr.snd_tag = NULL;
412 * Duplicate "from"'s mbuf pkthdr in "to".
413 * "from" must have M_PKTHDR set, and "to" must be empty.
414 * In particular, this does a deep copy of the packet tags.
417 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how)
422 * The mbuf allocator only initializes the pkthdr
423 * when the mbuf is allocated with m_gethdr(). Many users
424 * (e.g. m_copy*, m_prepend) use m_get() and then
425 * smash the pkthdr as needed causing these
426 * assertions to trip. For now just disable them.
429 /* Note: with MAC, this may not be a good assertion. */
430 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
432 MBUF_CHECKSLEEP(how);
434 if (to->m_flags & M_PKTHDR)
435 m_tag_delete_chain(to, NULL);
437 to->m_flags = (from->m_flags & M_COPYFLAGS) |
438 (to->m_flags & (M_EXT | M_NOMAP));
439 if ((to->m_flags & M_EXT) == 0)
440 to->m_data = to->m_pktdat;
441 to->m_pkthdr = from->m_pkthdr;
442 if (from->m_pkthdr.csum_flags & CSUM_SND_TAG)
443 m_snd_tag_ref(from->m_pkthdr.snd_tag);
444 SLIST_INIT(&to->m_pkthdr.tags);
445 return (m_tag_copy_chain(to, from, how));
449 * Lesser-used path for M_PREPEND:
450 * allocate new mbuf to prepend to chain,
454 m_prepend(struct mbuf *m, int len, int how)
458 if (m->m_flags & M_PKTHDR)
459 mn = m_gethdr(how, m->m_type);
461 mn = m_get(how, m->m_type);
466 if (m->m_flags & M_PKTHDR)
467 m_move_pkthdr(mn, m);
477 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
478 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
479 * The wait parameter is a choice of M_WAITOK/M_NOWAIT from caller.
480 * Note that the copy is read-only, because clusters are not copied,
481 * only their reference counts are incremented.
484 m_copym(struct mbuf *m, int off0, int len, int wait)
486 struct mbuf *n, **np;
491 KASSERT(off >= 0, ("m_copym, negative off %d", off));
492 KASSERT(len >= 0, ("m_copym, negative len %d", len));
493 MBUF_CHECKSLEEP(wait);
494 if (off == 0 && m->m_flags & M_PKTHDR)
497 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
507 KASSERT(len == M_COPYALL,
508 ("m_copym, length > size of mbuf chain"));
512 n = m_gethdr(wait, m->m_type);
514 n = m_get(wait, m->m_type);
519 if (!m_dup_pkthdr(n, m, wait))
521 if (len == M_COPYALL)
522 n->m_pkthdr.len -= off0;
524 n->m_pkthdr.len = len;
527 n->m_len = min(len, m->m_len - off);
528 if (m->m_flags & M_EXT) {
529 n->m_data = m->m_data + off;
532 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
534 if (len != M_COPYALL)
548 * Copy an entire packet, including header (which must be present).
549 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
550 * Note that the copy is read-only, because clusters are not copied,
551 * only their reference counts are incremented.
552 * Preserve alignment of the first mbuf so if the creator has left
553 * some room at the beginning (e.g. for inserting protocol headers)
554 * the copies still have the room available.
557 m_copypacket(struct mbuf *m, int how)
559 struct mbuf *top, *n, *o;
561 MBUF_CHECKSLEEP(how);
562 n = m_get(how, m->m_type);
567 if (!m_dup_pkthdr(n, m, how))
570 if (m->m_flags & M_EXT) {
571 n->m_data = m->m_data;
574 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
575 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
580 o = m_get(how, m->m_type);
588 if (m->m_flags & M_EXT) {
589 n->m_data = m->m_data;
592 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
604 m_copyfromunmapped(const struct mbuf *m, int off, int len, caddr_t cp)
610 KASSERT(off >= 0, ("m_copyfromunmapped: negative off %d", off));
611 KASSERT(len >= 0, ("m_copyfromunmapped: negative len %d", len));
612 KASSERT(off < m->m_len,
613 ("m_copyfromunmapped: len exceeds mbuf length"));
618 uio.uio_segflg = UIO_SYSSPACE;
621 uio.uio_rw = UIO_READ;
622 error = m_unmappedtouio(m, off, &uio, len);
623 KASSERT(error == 0, ("m_unmappedtouio failed: off %d, len %d", off,
628 * Copy data from an mbuf chain starting "off" bytes from the beginning,
629 * continuing for "len" bytes, into the indicated buffer.
632 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
636 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
637 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
639 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
646 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
647 count = min(m->m_len - off, len);
648 if ((m->m_flags & M_NOMAP) != 0)
649 m_copyfromunmapped(m, off, count, cp);
651 bcopy(mtod(m, caddr_t) + off, cp, count);
660 * Copy a packet header mbuf chain into a completely new chain, including
661 * copying any mbuf clusters. Use this instead of m_copypacket() when
662 * you need a writable copy of an mbuf chain.
665 m_dup(const struct mbuf *m, int how)
667 struct mbuf **p, *top = NULL;
668 int remain, moff, nsize;
670 MBUF_CHECKSLEEP(how);
676 /* While there's more data, get a new mbuf, tack it on, and fill it */
677 remain = m->m_pkthdr.len;
680 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
683 /* Get the next new mbuf */
684 if (remain >= MINCLSIZE) {
685 n = m_getcl(how, m->m_type, 0);
688 n = m_get(how, m->m_type);
694 if (top == NULL) { /* First one, must be PKTHDR */
695 if (!m_dup_pkthdr(n, m, how)) {
699 if ((n->m_flags & M_EXT) == 0)
701 n->m_flags &= ~M_RDONLY;
705 /* Link it into the new chain */
709 /* Copy data from original mbuf(s) into new mbuf */
710 while (n->m_len < nsize && m != NULL) {
711 int chunk = min(nsize - n->m_len, m->m_len - moff);
713 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
717 if (moff == m->m_len) {
723 /* Check correct total mbuf length */
724 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
725 ("%s: bogus m_pkthdr.len", __func__));
735 * Concatenate mbuf chain n to m.
736 * Both chains must be of the same type (e.g. MT_DATA).
737 * Any m_pkthdr is not updated.
740 m_cat(struct mbuf *m, struct mbuf *n)
745 if (!M_WRITABLE(m) ||
746 (n->m_flags & M_NOMAP) != 0 ||
747 M_TRAILINGSPACE(m) < n->m_len) {
748 /* just join the two chains */
752 /* splat the data from one into the other */
753 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
755 m->m_len += n->m_len;
761 * Concatenate two pkthdr mbuf chains.
764 m_catpkt(struct mbuf *m, struct mbuf *n)
770 m->m_pkthdr.len += n->m_pkthdr.len;
777 m_adj(struct mbuf *mp, int req_len)
783 if ((m = mp) == NULL)
789 while (m != NULL && len > 0) {
790 if (m->m_len <= len) {
800 if (mp->m_flags & M_PKTHDR)
801 mp->m_pkthdr.len -= (req_len - len);
804 * Trim from tail. Scan the mbuf chain,
805 * calculating its length and finding the last mbuf.
806 * If the adjustment only affects this mbuf, then just
807 * adjust and return. Otherwise, rescan and truncate
808 * after the remaining size.
814 if (m->m_next == (struct mbuf *)0)
818 if (m->m_len >= len) {
820 if (mp->m_flags & M_PKTHDR)
821 mp->m_pkthdr.len -= len;
828 * Correct length for chain is "count".
829 * Find the mbuf with last data, adjust its length,
830 * and toss data from remaining mbufs on chain.
833 if (m->m_flags & M_PKTHDR)
834 m->m_pkthdr.len = count;
835 for (; m; m = m->m_next) {
836 if (m->m_len >= count) {
838 if (m->m_next != NULL) {
850 * Rearange an mbuf chain so that len bytes are contiguous
851 * and in the data area of an mbuf (so that mtod will work
852 * for a structure of size len). Returns the resulting
853 * mbuf chain on success, frees it and returns null on failure.
854 * If there is room, it will add up to max_protohdr-len extra bytes to the
855 * contiguous region in an attempt to avoid being called next time.
858 m_pullup(struct mbuf *n, int len)
864 KASSERT((n->m_flags & M_NOMAP) == 0,
865 ("%s: unmapped mbuf %p", __func__, n));
868 * If first mbuf has no cluster, and has room for len bytes
869 * without shifting current data, pullup into it,
870 * otherwise allocate a new mbuf to prepend to the chain.
872 if ((n->m_flags & M_EXT) == 0 &&
873 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
882 m = m_get(M_NOWAIT, n->m_type);
885 if (n->m_flags & M_PKTHDR)
888 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
890 count = min(min(max(len, max_protohdr), space), n->m_len);
891 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
901 } while (len > 0 && n);
914 * Like m_pullup(), except a new mbuf is always allocated, and we allow
915 * the amount of empty space before the data in the new mbuf to be specified
916 * (in the event that the caller expects to prepend later).
919 m_copyup(struct mbuf *n, int len, int dstoff)
924 if (len > (MHLEN - dstoff))
926 m = m_get(M_NOWAIT, n->m_type);
929 if (n->m_flags & M_PKTHDR)
932 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
934 count = min(min(max(len, max_protohdr), space), n->m_len);
935 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
945 } while (len > 0 && n);
958 * Partition an mbuf chain in two pieces, returning the tail --
959 * all but the first len0 bytes. In case of failure, it returns NULL and
960 * attempts to restore the chain to its original state.
962 * Note that the resulting mbufs might be read-only, because the new
963 * mbuf can end up sharing an mbuf cluster with the original mbuf if
964 * the "breaking point" happens to lie within a cluster mbuf. Use the
965 * M_WRITABLE() macro to check for this case.
968 m_split(struct mbuf *m0, int len0, int wait)
971 u_int len = len0, remain;
973 MBUF_CHECKSLEEP(wait);
974 for (m = m0; m && len > m->m_len; m = m->m_next)
978 remain = m->m_len - len;
979 if (m0->m_flags & M_PKTHDR && remain == 0) {
980 n = m_gethdr(wait, m0->m_type);
983 n->m_next = m->m_next;
985 if (m0->m_pkthdr.csum_flags & CSUM_SND_TAG) {
986 n->m_pkthdr.snd_tag =
987 m_snd_tag_ref(m0->m_pkthdr.snd_tag);
988 n->m_pkthdr.csum_flags |= CSUM_SND_TAG;
990 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
991 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
992 m0->m_pkthdr.len = len0;
994 } else if (m0->m_flags & M_PKTHDR) {
995 n = m_gethdr(wait, m0->m_type);
998 if (m0->m_pkthdr.csum_flags & CSUM_SND_TAG) {
999 n->m_pkthdr.snd_tag =
1000 m_snd_tag_ref(m0->m_pkthdr.snd_tag);
1001 n->m_pkthdr.csum_flags |= CSUM_SND_TAG;
1003 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1004 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1005 m0->m_pkthdr.len = len0;
1006 if (m->m_flags & M_EXT)
1008 if (remain > MHLEN) {
1009 /* m can't be the lead packet */
1011 n->m_next = m_split(m, len, wait);
1012 if (n->m_next == NULL) {
1021 } else if (remain == 0) {
1026 n = m_get(wait, m->m_type);
1032 if (m->m_flags & M_EXT) {
1033 n->m_data = m->m_data + len;
1036 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1040 n->m_next = m->m_next;
1045 * Routine to copy from device local memory into mbufs.
1046 * Note that `off' argument is offset into first mbuf of target chain from
1047 * which to begin copying the data to.
1050 m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
1051 void (*copy)(char *from, caddr_t to, u_int len))
1054 struct mbuf *top = NULL, **mp = ⊤
1057 if (off < 0 || off > MHLEN)
1060 while (totlen > 0) {
1061 if (top == NULL) { /* First one, must be PKTHDR */
1062 if (totlen + off >= MINCLSIZE) {
1063 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1066 m = m_gethdr(M_NOWAIT, MT_DATA);
1069 /* Place initial small packet/header at end of mbuf */
1070 if (m && totlen + off + max_linkhdr <= MHLEN) {
1071 m->m_data += max_linkhdr;
1077 m->m_pkthdr.rcvif = ifp;
1078 m->m_pkthdr.len = totlen;
1080 if (totlen + off >= MINCLSIZE) {
1081 m = m_getcl(M_NOWAIT, MT_DATA, 0);
1084 m = m_get(M_NOWAIT, MT_DATA);
1097 m->m_len = len = min(totlen, len);
1099 copy(buf, mtod(m, caddr_t), (u_int)len);
1101 bcopy(buf, mtod(m, caddr_t), (u_int)len);
1111 * Copy data from a buffer back into the indicated mbuf chain,
1112 * starting "off" bytes from the beginning, extending the mbuf
1113 * chain if necessary.
1116 m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
1119 struct mbuf *m = m0, *n;
1124 while (off > (mlen = m->m_len)) {
1127 if (m->m_next == NULL) {
1128 n = m_get(M_NOWAIT, m->m_type);
1131 bzero(mtod(n, caddr_t), MLEN);
1132 n->m_len = min(MLEN, len + off);
1138 if (m->m_next == NULL && (len > m->m_len - off)) {
1139 m->m_len += min(len - (m->m_len - off),
1140 M_TRAILINGSPACE(m));
1142 mlen = min (m->m_len - off, len);
1143 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
1151 if (m->m_next == NULL) {
1152 n = m_get(M_NOWAIT, m->m_type);
1155 n->m_len = min(MLEN, len);
1160 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1161 m->m_pkthdr.len = totlen;
1165 * Append the specified data to the indicated mbuf chain,
1166 * Extend the mbuf chain if the new data does not fit in
1169 * Return 1 if able to complete the job; otherwise 0.
1172 m_append(struct mbuf *m0, int len, c_caddr_t cp)
1175 int remainder, space;
1177 for (m = m0; m->m_next != NULL; m = m->m_next)
1180 space = M_TRAILINGSPACE(m);
1183 * Copy into available space.
1185 if (space > remainder)
1187 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1189 cp += space, remainder -= space;
1191 while (remainder > 0) {
1193 * Allocate a new mbuf; could check space
1194 * and allocate a cluster instead.
1196 n = m_get(M_NOWAIT, m->m_type);
1199 n->m_len = min(MLEN, remainder);
1200 bcopy(cp, mtod(n, caddr_t), n->m_len);
1201 cp += n->m_len, remainder -= n->m_len;
1205 if (m0->m_flags & M_PKTHDR)
1206 m0->m_pkthdr.len += len - remainder;
1207 return (remainder == 0);
1211 * Apply function f to the data in an mbuf chain starting "off" bytes from
1212 * the beginning, continuing for "len" bytes.
1215 m_apply(struct mbuf *m, int off, int len,
1216 int (*f)(void *, void *, u_int), void *arg)
1221 KASSERT(off >= 0, ("m_apply, negative off %d", off));
1222 KASSERT(len >= 0, ("m_apply, negative len %d", len));
1224 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1231 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1232 count = min(m->m_len - off, len);
1233 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1244 * Return a pointer to mbuf/offset of location in mbuf chain.
1247 m_getptr(struct mbuf *m, int loc, int *off)
1251 /* Normal end of search. */
1252 if (m->m_len > loc) {
1257 if (m->m_next == NULL) {
1259 /* Point at the end of valid data. */
1272 m_print(const struct mbuf *m, int maxlen)
1276 const struct mbuf *m2;
1279 printf("mbuf: %p\n", m);
1283 if (m->m_flags & M_PKTHDR)
1284 len = m->m_pkthdr.len;
1288 while (m2 != NULL && (len == -1 || len)) {
1290 if (maxlen != -1 && pdata > maxlen)
1292 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
1293 m2->m_next, m2->m_flags, "\20\20freelist\17skipfw"
1294 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
1295 "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
1297 printf(", %*D\n", pdata, (u_char *)m2->m_data, "-");
1303 printf("%d bytes unaccounted for.\n", len);
1308 m_fixhdr(struct mbuf *m0)
1312 len = m_length(m0, NULL);
1313 m0->m_pkthdr.len = len;
1318 m_length(struct mbuf *m0, struct mbuf **last)
1324 for (m = m0; m != NULL; m = m->m_next) {
1326 if (m->m_next == NULL)
1335 * Defragment a mbuf chain, returning the shortest possible
1336 * chain of mbufs and clusters. If allocation fails and
1337 * this cannot be completed, NULL will be returned, but
1338 * the passed in chain will be unchanged. Upon success,
1339 * the original chain will be freed, and the new chain
1342 * If a non-packet header is passed in, the original
1343 * mbuf (chain?) will be returned unharmed.
1346 m_defrag(struct mbuf *m0, int how)
1348 struct mbuf *m_new = NULL, *m_final = NULL;
1349 int progress = 0, length;
1351 MBUF_CHECKSLEEP(how);
1352 if (!(m0->m_flags & M_PKTHDR))
1355 m_fixhdr(m0); /* Needed sanity check */
1357 #ifdef MBUF_STRESS_TEST
1358 if (m_defragrandomfailures) {
1359 int temp = arc4random() & 0xff;
1365 if (m0->m_pkthdr.len > MHLEN)
1366 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1368 m_final = m_gethdr(how, MT_DATA);
1370 if (m_final == NULL)
1373 if (m_dup_pkthdr(m_final, m0, how) == 0)
1378 while (progress < m0->m_pkthdr.len) {
1379 length = m0->m_pkthdr.len - progress;
1380 if (length > MCLBYTES)
1383 if (m_new == NULL) {
1385 m_new = m_getcl(how, MT_DATA, 0);
1387 m_new = m_get(how, MT_DATA);
1392 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1394 m_new->m_len = length;
1395 if (m_new != m_final)
1396 m_cat(m_final, m_new);
1399 #ifdef MBUF_STRESS_TEST
1400 if (m0->m_next == NULL)
1405 #ifdef MBUF_STRESS_TEST
1407 m_defragbytes += m0->m_pkthdr.len;
1411 #ifdef MBUF_STRESS_TEST
1420 * Return the number of fragments an mbuf will use. This is usually
1421 * used as a proxy for the number of scatter/gather elements needed by
1422 * a DMA engine to access an mbuf. In general mapped mbufs are
1423 * assumed to be backed by physically contiguous buffers that only
1424 * need a single fragment. Unmapped mbufs, on the other hand, can
1425 * span disjoint physical pages.
1428 frags_per_mbuf(struct mbuf *m)
1430 struct mbuf_ext_pgs *ext_pgs;
1433 if ((m->m_flags & M_NOMAP) == 0)
1437 * The header and trailer are counted as a single fragment
1438 * each when present.
1440 * XXX: This overestimates the number of fragments by assuming
1441 * all the backing physical pages are disjoint.
1443 ext_pgs = &m->m_ext_pgs;
1445 if (ext_pgs->hdr_len != 0)
1447 frags += ext_pgs->npgs;
1448 if (ext_pgs->trail_len != 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)
1631 struct mbuf_ext_pgs *ext_pgs;
1634 MBUF_EXT_PGS_ASSERT(m);
1635 ext_pgs = &m->m_ext_pgs;
1636 for (int i = 0; i < ext_pgs->npgs; i++) {
1637 pg = PHYS_TO_VM_PAGE(m->m_epg_pa[i]);
1638 vm_page_unwire_noq(pg);
1643 static struct mbuf *
1644 m_uiotombuf_nomap(struct uio *uio, int how, int len, int maxseg, int flags)
1646 struct mbuf *m, *mb, *prev;
1647 struct mbuf_ext_pgs *pgs;
1648 vm_page_t pg_array[MBUF_PEXT_MAX_PGS];
1649 int error, length, i, needed;
1651 int pflags = malloc2vm_flags(how) | VM_ALLOC_NOOBJ | VM_ALLOC_NODUMP |
1655 * len can be zero or an arbitrary large value bound by
1656 * the total data supplied by the uio.
1659 total = MIN(uio->uio_resid, len);
1661 total = uio->uio_resid;
1664 maxseg = MBUF_PEXT_MAX_PGS * PAGE_SIZE;
1667 * Allocate the pages
1670 MPASS((flags & M_PKTHDR) == 0);
1672 mb = mb_alloc_ext_pgs(how, mb_free_mext_pgs);
1680 pgs = &mb->m_ext_pgs;
1681 pgs->flags = EPG_FLAG_ANON;
1682 needed = length = MIN(maxseg, total);
1683 for (i = 0; needed > 0; i++, needed -= PAGE_SIZE) {
1685 pg_array[i] = vm_page_alloc(NULL, 0, pflags);
1686 if (pg_array[i] == NULL) {
1687 if (how & M_NOWAIT) {
1694 pg_array[i]->flags &= ~PG_ZERO;
1695 mb->m_epg_pa[i] = VM_PAGE_TO_PHYS(pg_array[i]);
1698 pgs->last_pg_len = length - PAGE_SIZE * (pgs->npgs - 1);
1699 MBUF_EXT_PGS_ASSERT_SANITY(mb);
1701 error = uiomove_fromphys(pg_array, 0, length, uio);
1705 mb->m_ext.ext_size += PAGE_SIZE * pgs->npgs;
1706 if (flags & M_PKTHDR)
1707 m->m_pkthdr.len += length;
1717 * Copy the contents of uio into a properly sized mbuf chain.
1720 m_uiotombuf(struct uio *uio, int how, int len, int align, int flags)
1722 struct mbuf *m, *mb;
1727 if (flags & M_NOMAP)
1728 return (m_uiotombuf_nomap(uio, how, len, align, flags));
1731 * len can be zero or an arbitrary large value bound by
1732 * the total data supplied by the uio.
1735 total = (uio->uio_resid < len) ? uio->uio_resid : len;
1737 total = uio->uio_resid;
1740 * The smallest unit returned by m_getm2() is a single mbuf
1741 * with pkthdr. We can't align past it.
1747 * Give us the full allocation or nothing.
1748 * If len is zero return the smallest empty mbuf.
1750 m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags);
1755 /* Fill all mbufs with uio data and update header information. */
1756 for (mb = m; mb != NULL; mb = mb->m_next) {
1757 length = min(M_TRAILINGSPACE(mb), total - progress);
1759 error = uiomove(mtod(mb, void *), length, uio);
1767 if (flags & M_PKTHDR)
1768 m->m_pkthdr.len += length;
1770 KASSERT(progress == total, ("%s: progress != total", __func__));
1776 * Copy data from an unmapped mbuf into a uio limited by len if set.
1779 m_unmappedtouio(const struct mbuf *m, int m_off, struct uio *uio, int len)
1781 struct mbuf_ext_pgs *ext_pgs;
1783 int error, i, off, pglen, pgoff, seglen, segoff;
1785 MBUF_EXT_PGS_ASSERT(m);
1786 ext_pgs = __DECONST(void *, &m->m_ext_pgs);
1789 /* Skip over any data removed from the front. */
1790 off = mtod(m, vm_offset_t);
1793 if (ext_pgs->hdr_len != 0) {
1794 if (off >= ext_pgs->hdr_len) {
1795 off -= ext_pgs->hdr_len;
1797 seglen = ext_pgs->hdr_len - off;
1799 seglen = min(seglen, len);
1802 error = uiomove(__DECONST(void *,
1803 &m->m_epg_hdr[segoff]), seglen, uio);
1806 pgoff = ext_pgs->first_pg_off;
1807 for (i = 0; i < ext_pgs->npgs && error == 0 && len > 0; i++) {
1808 pglen = m_epg_pagelen(m, i, pgoff);
1814 seglen = pglen - off;
1815 segoff = pgoff + off;
1817 seglen = min(seglen, len);
1819 pg = PHYS_TO_VM_PAGE(m->m_epg_pa[i]);
1820 error = uiomove_fromphys(&pg, segoff, seglen, uio);
1823 if (len != 0 && error == 0) {
1824 KASSERT((off + len) <= ext_pgs->trail_len,
1825 ("off + len > trail (%d + %d > %d, m_off = %d)", off, len,
1826 ext_pgs->trail_len, m_off));
1827 error = uiomove(__DECONST(void *, &m->m_epg_trail[off]),
1834 * Copy an mbuf chain into a uio limited by len if set.
1837 m_mbuftouio(struct uio *uio, const struct mbuf *m, int len)
1839 int error, length, total;
1843 total = min(uio->uio_resid, len);
1845 total = uio->uio_resid;
1847 /* Fill the uio with data from the mbufs. */
1848 for (; m != NULL; m = m->m_next) {
1849 length = min(m->m_len, total - progress);
1851 if ((m->m_flags & M_NOMAP) != 0)
1852 error = m_unmappedtouio(m, 0, uio, length);
1854 error = uiomove(mtod(m, void *), length, uio);
1865 * Create a writable copy of the mbuf chain. While doing this
1866 * we compact the chain with a goal of producing a chain with
1867 * at most two mbufs. The second mbuf in this chain is likely
1868 * to be a cluster. The primary purpose of this work is to create
1869 * a writable packet for encryption, compression, etc. The
1870 * secondary goal is to linearize the data so the data can be
1871 * passed to crypto hardware in the most efficient manner possible.
1874 m_unshare(struct mbuf *m0, int how)
1876 struct mbuf *m, *mprev;
1877 struct mbuf *n, *mfirst, *mlast;
1881 for (m = m0; m != NULL; m = mprev->m_next) {
1883 * Regular mbufs are ignored unless there's a cluster
1884 * in front of it that we can use to coalesce. We do
1885 * the latter mainly so later clusters can be coalesced
1886 * also w/o having to handle them specially (i.e. convert
1887 * mbuf+cluster -> cluster). This optimization is heavily
1888 * influenced by the assumption that we're running over
1889 * Ethernet where MCLBYTES is large enough that the max
1890 * packet size will permit lots of coalescing into a
1891 * single cluster. This in turn permits efficient
1892 * crypto operations, especially when using hardware.
1894 if ((m->m_flags & M_EXT) == 0) {
1895 if (mprev && (mprev->m_flags & M_EXT) &&
1896 m->m_len <= M_TRAILINGSPACE(mprev)) {
1897 /* XXX: this ignores mbuf types */
1898 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1899 mtod(m, caddr_t), m->m_len);
1900 mprev->m_len += m->m_len;
1901 mprev->m_next = m->m_next; /* unlink from chain */
1902 m_free(m); /* reclaim mbuf */
1909 * Writable mbufs are left alone (for now).
1911 if (M_WRITABLE(m)) {
1917 * Not writable, replace with a copy or coalesce with
1918 * the previous mbuf if possible (since we have to copy
1919 * it anyway, we try to reduce the number of mbufs and
1920 * clusters so that future work is easier).
1922 KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags));
1923 /* NB: we only coalesce into a cluster or larger */
1924 if (mprev != NULL && (mprev->m_flags & M_EXT) &&
1925 m->m_len <= M_TRAILINGSPACE(mprev)) {
1926 /* XXX: this ignores mbuf types */
1927 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1928 mtod(m, caddr_t), m->m_len);
1929 mprev->m_len += m->m_len;
1930 mprev->m_next = m->m_next; /* unlink from chain */
1931 m_free(m); /* reclaim mbuf */
1936 * Allocate new space to hold the copy and copy the data.
1937 * We deal with jumbo mbufs (i.e. m_len > MCLBYTES) by
1938 * splitting them into clusters. We could just malloc a
1939 * buffer and make it external but too many device drivers
1940 * don't know how to break up the non-contiguous memory when
1943 n = m_getcl(how, m->m_type, m->m_flags & M_COPYFLAGS);
1948 if (m->m_flags & M_PKTHDR) {
1949 KASSERT(mprev == NULL, ("%s: m0 %p, m %p has M_PKTHDR",
1951 m_move_pkthdr(n, m);
1958 int cc = min(len, MCLBYTES);
1959 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc);
1965 newipsecstat.ips_clcopied++;
1973 n = m_getcl(how, m->m_type, m->m_flags & M_COPYFLAGS);
1980 n->m_next = m->m_next;
1982 m0 = mfirst; /* new head of chain */
1984 mprev->m_next = mfirst; /* replace old mbuf */
1985 m_free(m); /* release old mbuf */
1991 #ifdef MBUF_PROFILING
1993 #define MP_BUCKETS 32 /* don't just change this as things may overflow.*/
1994 struct mbufprofile {
1995 uintmax_t wasted[MP_BUCKETS];
1996 uintmax_t used[MP_BUCKETS];
1997 uintmax_t segments[MP_BUCKETS];
2000 #define MP_MAXDIGITS 21 /* strlen("16,000,000,000,000,000,000") == 21 */
2001 #define MP_NUMLINES 6
2002 #define MP_NUMSPERLINE 16
2003 #define MP_EXTRABYTES 64 /* > strlen("used:\nwasted:\nsegments:\n") */
2004 /* work out max space needed and add a bit of spare space too */
2005 #define MP_MAXLINE ((MP_MAXDIGITS+1) * MP_NUMSPERLINE)
2006 #define MP_BUFSIZE ((MP_MAXLINE * MP_NUMLINES) + 1 + MP_EXTRABYTES)
2008 char mbprofbuf[MP_BUFSIZE];
2011 m_profile(struct mbuf *m)
2020 if (m->m_flags & M_EXT) {
2021 wasted += MHLEN - sizeof(m->m_ext) +
2022 m->m_ext.ext_size - m->m_len;
2024 if (m->m_flags & M_PKTHDR)
2025 wasted += MHLEN - m->m_len;
2027 wasted += MLEN - m->m_len;
2031 /* be paranoid.. it helps */
2032 if (segments > MP_BUCKETS - 1)
2033 segments = MP_BUCKETS - 1;
2036 if (wasted > 100000)
2038 /* store in the appropriate bucket */
2039 /* don't bother locking. if it's slightly off, so what? */
2040 mbprof.segments[segments]++;
2041 mbprof.used[fls(used)]++;
2042 mbprof.wasted[fls(wasted)]++;
2046 mbprof_textify(void)
2052 p = &mbprof.wasted[0];
2054 offset = snprintf(c, MP_MAXLINE + 10,
2056 "%ju %ju %ju %ju %ju %ju %ju %ju "
2057 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2058 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2059 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2061 p = &mbprof.wasted[16];
2063 offset = snprintf(c, MP_MAXLINE,
2064 "%ju %ju %ju %ju %ju %ju %ju %ju "
2065 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2066 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2067 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2069 p = &mbprof.used[0];
2071 offset = snprintf(c, MP_MAXLINE + 10,
2073 "%ju %ju %ju %ju %ju %ju %ju %ju "
2074 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2075 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2076 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2078 p = &mbprof.used[16];
2080 offset = snprintf(c, MP_MAXLINE,
2081 "%ju %ju %ju %ju %ju %ju %ju %ju "
2082 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2083 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2084 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2086 p = &mbprof.segments[0];
2088 offset = snprintf(c, MP_MAXLINE + 10,
2090 "%ju %ju %ju %ju %ju %ju %ju %ju "
2091 "%ju %ju %ju %ju %ju %ju %ju %ju\n",
2092 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2093 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2095 p = &mbprof.segments[16];
2097 offset = snprintf(c, MP_MAXLINE,
2098 "%ju %ju %ju %ju %ju %ju %ju %ju "
2099 "%ju %ju %ju %ju %ju %ju %ju %jju",
2100 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
2101 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
2106 mbprof_handler(SYSCTL_HANDLER_ARGS)
2111 error = SYSCTL_OUT(req, mbprofbuf, strlen(mbprofbuf) + 1);
2116 mbprof_clr_handler(SYSCTL_HANDLER_ARGS)
2121 error = sysctl_handle_int(oidp, &clear, 0, req);
2122 if (error || !req->newptr)
2126 bzero(&mbprof, sizeof(mbprof));
2132 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofile,
2133 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, NULL, 0,
2134 mbprof_handler, "A",
2135 "mbuf profiling statistics");
2137 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofileclr,
2138 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, NULL, 0,
2139 mbprof_clr_handler, "I",
2140 "clear mbuf profiling statistics");