2 * Copyright (c) 1982, 1986, 1988, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
36 #include "opt_param.h"
37 #include "opt_mbuf_stress_test.h"
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #include <sys/limits.h>
44 #include <sys/malloc.h>
46 #include <sys/sysctl.h>
47 #include <sys/domain.h>
48 #include <sys/protosw.h>
51 #include <security/mac/mac_framework.h>
57 #ifdef MBUF_STRESS_TEST
62 int m_defragrandomfailures;
66 * sysctl(8) exported objects
68 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD,
69 &max_linkhdr, 0, "Size of largest link layer header");
70 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD,
71 &max_protohdr, 0, "Size of largest protocol layer header");
72 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD,
73 &max_hdr, 0, "Size of largest link plus protocol header");
74 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RD,
75 &max_datalen, 0, "Minimum space left in mbuf after max_hdr");
76 #ifdef MBUF_STRESS_TEST
77 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
78 &m_defragpackets, 0, "");
79 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
80 &m_defragbytes, 0, "");
81 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
82 &m_defraguseless, 0, "");
83 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
84 &m_defragfailure, 0, "");
85 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
86 &m_defragrandomfailures, 0, "");
90 * Allocate a given length worth of mbufs and/or clusters (whatever fits
91 * best) and return a pointer to the top of the allocated chain. If an
92 * existing mbuf chain is provided, then we will append the new chain
93 * to the existing one but still return the top of the newly allocated
97 m_getm2(struct mbuf *m, int len, int how, short type, int flags)
99 struct mbuf *mb, *nm = NULL, *mtail = NULL;
101 KASSERT(len >= 0, ("%s: len is < 0", __func__));
103 /* Validate flags. */
104 flags &= (M_PKTHDR | M_EOR);
106 /* Packet header mbuf must be first in chain. */
107 if ((flags & M_PKTHDR) && m != NULL)
110 /* Loop and append maximum sized mbufs to the chain tail. */
113 mb = m_getjcl(how, type, (flags & M_PKTHDR),
115 else if (len >= MINCLSIZE)
116 mb = m_getcl(how, type, (flags & M_PKTHDR));
117 else if (flags & M_PKTHDR)
118 mb = m_gethdr(how, type);
120 mb = m_get(how, type);
122 /* Fail the whole operation if one mbuf can't be allocated. */
130 len -= (mb->m_flags & M_EXT) ? mb->m_ext.ext_size :
131 ((mb->m_flags & M_PKTHDR) ? MHLEN : MLEN);
137 flags &= ~M_PKTHDR; /* Only valid on the first mbuf. */
140 mtail->m_flags |= M_EOR; /* Only valid on the last mbuf. */
142 /* If mbuf was supplied, append new chain to the end of it. */
144 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next)
147 mtail->m_flags &= ~M_EOR;
155 * Free an entire chain of mbufs and associated external buffers, if
159 m_freem(struct mbuf *mb)
167 * Configure a provided mbuf to refer to the provided external storage
168 * buffer and setup a reference count for said buffer. If the setting
169 * up of the reference count fails, the M_EXT bit will not be set. If
170 * successfull, the M_EXT bit is set in the mbuf's flags.
173 * mb The existing mbuf to which to attach the provided buffer.
174 * buf The address of the provided external storage buffer.
175 * size The size of the provided buffer.
176 * freef A pointer to a routine that is responsible for freeing the
177 * provided external storage buffer.
178 * args A pointer to an argument structure (of any type) to be passed
179 * to the provided freef routine (may be NULL).
180 * flags Any other flags to be passed to the provided mbuf.
181 * type The type that the external storage buffer should be
188 m_extadd(struct mbuf *mb, caddr_t buf, u_int size,
189 void (*freef)(void *, void *), void *args, int flags, int type)
191 KASSERT(type != EXT_CLUSTER, ("%s: EXT_CLUSTER not allowed", __func__));
193 if (type != EXT_EXTREF)
194 mb->m_ext.ref_cnt = (u_int *)uma_zalloc(zone_ext_refcnt, M_NOWAIT);
195 if (mb->m_ext.ref_cnt != NULL) {
196 *(mb->m_ext.ref_cnt) = 1;
197 mb->m_flags |= (M_EXT | flags);
198 mb->m_ext.ext_buf = buf;
199 mb->m_data = mb->m_ext.ext_buf;
200 mb->m_ext.ext_size = size;
201 mb->m_ext.ext_free = freef;
202 mb->m_ext.ext_args = args;
203 mb->m_ext.ext_type = type;
208 * Non-directly-exported function to clean up after mbufs with M_EXT
209 * storage attached to them if the reference count hits 1.
212 mb_free_ext(struct mbuf *m)
216 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
217 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
221 * check if the header is embedded in the cluster
223 skipmbuf = (m->m_flags & M_NOFREE);
225 /* Free attached storage if this mbuf is the only reference to it. */
226 if (*(m->m_ext.ref_cnt) == 1 ||
227 atomic_fetchadd_int(m->m_ext.ref_cnt, -1) == 1) {
228 switch (m->m_ext.ext_type) {
229 case EXT_PACKET: /* The packet zone is special. */
230 if (*(m->m_ext.ref_cnt) == 0)
231 *(m->m_ext.ref_cnt) = 1;
232 uma_zfree(zone_pack, m);
233 return; /* Job done. */
235 uma_zfree(zone_clust, m->m_ext.ext_buf);
238 uma_zfree(zone_jumbop, m->m_ext.ext_buf);
241 uma_zfree(zone_jumbo9, m->m_ext.ext_buf);
244 uma_zfree(zone_jumbo16, m->m_ext.ext_buf);
250 *(m->m_ext.ref_cnt) = 0;
251 uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *,
255 KASSERT(m->m_ext.ext_free != NULL,
256 ("%s: ext_free not set", __func__));
257 (*(m->m_ext.ext_free))(m->m_ext.ext_buf,
261 KASSERT(m->m_ext.ext_type == 0,
262 ("%s: unknown ext_type", __func__));
269 * Free this mbuf back to the mbuf zone with all m_ext
270 * information purged.
272 m->m_ext.ext_buf = NULL;
273 m->m_ext.ext_free = NULL;
274 m->m_ext.ext_args = NULL;
275 m->m_ext.ref_cnt = NULL;
276 m->m_ext.ext_size = 0;
277 m->m_ext.ext_type = 0;
278 m->m_flags &= ~M_EXT;
279 uma_zfree(zone_mbuf, m);
283 * Attach the the cluster from *m to *n, set up m_ext in *n
284 * and bump the refcount of the cluster.
287 mb_dupcl(struct mbuf *n, struct mbuf *m)
289 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
290 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
291 KASSERT((n->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
293 if (*(m->m_ext.ref_cnt) == 1)
294 *(m->m_ext.ref_cnt) += 1;
296 atomic_add_int(m->m_ext.ref_cnt, 1);
297 n->m_ext.ext_buf = m->m_ext.ext_buf;
298 n->m_ext.ext_free = m->m_ext.ext_free;
299 n->m_ext.ext_args = m->m_ext.ext_args;
300 n->m_ext.ext_size = m->m_ext.ext_size;
301 n->m_ext.ref_cnt = m->m_ext.ref_cnt;
302 n->m_ext.ext_type = m->m_ext.ext_type;
307 * Clean up mbuf (chain) from any tags and packet headers.
308 * If "all" is set then the first mbuf in the chain will be
312 m_demote(struct mbuf *m0, int all)
316 for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) {
317 if (m->m_flags & M_PKTHDR) {
318 m_tag_delete_chain(m, NULL);
319 m->m_flags &= ~M_PKTHDR;
320 bzero(&m->m_pkthdr, sizeof(struct pkthdr));
322 if (m->m_type == MT_HEADER)
324 if (m != m0 && m->m_nextpkt != NULL)
326 m->m_flags = m->m_flags & (M_EXT|M_EOR|M_RDONLY|M_FREELIST);
331 * Sanity checks on mbuf (chain) for use in KASSERT() and general
333 * Returns 0 or panics when bad and 1 on all tests passed.
334 * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they
338 m_sanity(struct mbuf *m0, int sanitize)
345 #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m)
347 #define M_SANITY_ACTION(s) printf("mbuf %p: " s, m)
350 for (m = m0; m != NULL; m = m->m_next) {
352 * Basic pointer checks. If any of these fails then some
353 * unrelated kernel memory before or after us is trashed.
354 * No way to recover from that.
356 a = ((m->m_flags & M_EXT) ? m->m_ext.ext_buf :
357 ((m->m_flags & M_PKTHDR) ? (caddr_t)(&m->m_pktdat) :
358 (caddr_t)(&m->m_dat)) );
359 b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size :
360 ((m->m_flags & M_PKTHDR) ? MHLEN : MLEN)));
361 if ((caddr_t)m->m_data < a)
362 M_SANITY_ACTION("m_data outside mbuf data range left");
363 if ((caddr_t)m->m_data > b)
364 M_SANITY_ACTION("m_data outside mbuf data range right");
365 if ((caddr_t)m->m_data + m->m_len > b)
366 M_SANITY_ACTION("m_data + m_len exeeds mbuf space");
367 if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.header) {
368 if ((caddr_t)m->m_pkthdr.header < a ||
369 (caddr_t)m->m_pkthdr.header > b)
370 M_SANITY_ACTION("m_pkthdr.header outside mbuf data range");
373 /* m->m_nextpkt may only be set on first mbuf in chain. */
374 if (m != m0 && m->m_nextpkt != NULL) {
376 m_freem(m->m_nextpkt);
377 m->m_nextpkt = (struct mbuf *)0xDEADC0DE;
379 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf");
382 /* packet length (not mbuf length!) calculation */
383 if (m0->m_flags & M_PKTHDR)
386 /* m_tags may only be attached to first mbuf in chain. */
387 if (m != m0 && m->m_flags & M_PKTHDR &&
388 !SLIST_EMPTY(&m->m_pkthdr.tags)) {
390 m_tag_delete_chain(m, NULL);
391 /* put in 0xDEADC0DE perhaps? */
393 M_SANITY_ACTION("m_tags on in-chain mbuf");
396 /* M_PKTHDR may only be set on first mbuf in chain */
397 if (m != m0 && m->m_flags & M_PKTHDR) {
399 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
400 m->m_flags &= ~M_PKTHDR;
401 /* put in 0xDEADCODE and leave hdr flag in */
403 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf");
407 if (pktlen && pktlen != m->m_pkthdr.len) {
411 M_SANITY_ACTION("m_pkthdr.len != mbuf chain length");
415 #undef M_SANITY_ACTION
420 * "Move" mbuf pkthdr from "from" to "to".
421 * "from" must have M_PKTHDR set, and "to" must be empty.
424 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
428 /* see below for why these are not enabled */
430 /* Note: with MAC, this may not be a good assertion. */
431 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
432 ("m_move_pkthdr: to has tags"));
436 * XXXMAC: It could be this should also occur for non-MAC?
438 if (to->m_flags & M_PKTHDR)
439 m_tag_delete_chain(to, NULL);
441 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
442 if ((to->m_flags & M_EXT) == 0)
443 to->m_data = to->m_pktdat;
444 to->m_pkthdr = from->m_pkthdr; /* especially tags */
445 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
446 from->m_flags &= ~M_PKTHDR;
450 * Duplicate "from"'s mbuf pkthdr in "to".
451 * "from" must have M_PKTHDR set, and "to" must be empty.
452 * In particular, this does a deep copy of the packet tags.
455 m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how)
460 * The mbuf allocator only initializes the pkthdr
461 * when the mbuf is allocated with MGETHDR. Many users
462 * (e.g. m_copy*, m_prepend) use MGET and then
463 * smash the pkthdr as needed causing these
464 * assertions to trip. For now just disable them.
467 /* Note: with MAC, this may not be a good assertion. */
468 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
470 MBUF_CHECKSLEEP(how);
472 if (to->m_flags & M_PKTHDR)
473 m_tag_delete_chain(to, NULL);
475 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
476 if ((to->m_flags & M_EXT) == 0)
477 to->m_data = to->m_pktdat;
478 to->m_pkthdr = from->m_pkthdr;
479 SLIST_INIT(&to->m_pkthdr.tags);
480 return (m_tag_copy_chain(to, from, MBTOM(how)));
484 * Lesser-used path for M_PREPEND:
485 * allocate new mbuf to prepend to chain,
489 m_prepend(struct mbuf *m, int len, int how)
493 if (m->m_flags & M_PKTHDR)
494 MGETHDR(mn, how, m->m_type);
496 MGET(mn, how, m->m_type);
501 if (m->m_flags & M_PKTHDR)
502 M_MOVE_PKTHDR(mn, m);
505 if(m->m_flags & M_PKTHDR) {
517 * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
518 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
519 * The wait parameter is a choice of M_TRYWAIT/M_DONTWAIT from caller.
520 * Note that the copy is read-only, because clusters are not copied,
521 * only their reference counts are incremented.
524 m_copym(struct mbuf *m, int off0, int len, int wait)
526 struct mbuf *n, **np;
531 KASSERT(off >= 0, ("m_copym, negative off %d", off));
532 KASSERT(len >= 0, ("m_copym, negative len %d", len));
533 MBUF_CHECKSLEEP(wait);
534 if (off == 0 && m->m_flags & M_PKTHDR)
537 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
547 KASSERT(len == M_COPYALL,
548 ("m_copym, length > size of mbuf chain"));
552 MGETHDR(n, wait, m->m_type);
554 MGET(n, wait, m->m_type);
559 if (!m_dup_pkthdr(n, m, wait))
561 if (len == M_COPYALL)
562 n->m_pkthdr.len -= off0;
564 n->m_pkthdr.len = len;
567 n->m_len = min(len, m->m_len - off);
568 if (m->m_flags & M_EXT) {
569 n->m_data = m->m_data + off;
572 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
574 if (len != M_COPYALL)
581 mbstat.m_mcfail++; /* XXX: No consistency. */
586 mbstat.m_mcfail++; /* XXX: No consistency. */
591 * Returns mbuf chain with new head for the prepending case.
592 * Copies from mbuf (chain) n from off for len to mbuf (chain) m
593 * either prepending or appending the data.
594 * The resulting mbuf (chain) m is fully writeable.
595 * m is destination (is made writeable)
596 * n is source, off is offset in source, len is len from offset
597 * dir, 0 append, 1 prepend
598 * how, wait or nowait
602 m_bcopyxxx(void *s, void *t, u_int len)
604 bcopy(s, t, (size_t)len);
609 m_copymdata(struct mbuf *m, struct mbuf *n, int off, int len,
612 struct mbuf *mm, *x, *z, *prev = NULL;
617 KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source"));
618 KASSERT(off >= 0, ("m_copymdata, negative off %d", off));
619 KASSERT(len >= 0, ("m_copymdata, negative len %d", len));
620 KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep));
629 for (z = n; z != NULL; z = z->m_next)
631 if (len == M_COPYALL)
633 if (off + len > nlen || len < 1)
636 if (!M_WRITABLE(mm)) {
637 /* XXX: Use proper m_xxx function instead. */
638 x = m_getcl(how, MT_DATA, mm->m_flags);
641 bcopy(mm->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size);
642 p = x->m_ext.ext_buf + (mm->m_data - mm->m_ext.ext_buf);
652 * Append/prepend the data. Allocating mbufs as necessary.
654 /* Shortcut if enough free space in first/last mbuf. */
655 if (!prep && M_TRAILINGSPACE(mm) >= len) {
656 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) +
659 mm->m_pkthdr.len += len;
662 if (prep && M_LEADINGSPACE(mm) >= len) {
663 mm->m_data = mtod(mm, caddr_t) - len;
664 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t));
666 mm->m_pkthdr.len += len;
670 /* Expand first/last mbuf to cluster if possible. */
671 if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) {
672 bcopy(mm->m_data, &buf, mm->m_len);
674 if (!(mm->m_flags & M_EXT))
676 bcopy(&buf, mm->m_ext.ext_buf, mm->m_len);
677 mm->m_data = mm->m_ext.ext_buf;
678 mm->m_pkthdr.header = NULL;
680 if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) {
681 bcopy(mm->m_data, &buf, mm->m_len);
683 if (!(mm->m_flags & M_EXT))
685 bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf +
686 mm->m_ext.ext_size - mm->m_len, mm->m_len);
687 mm->m_data = (caddr_t)mm->m_ext.ext_buf +
688 mm->m_ext.ext_size - mm->m_len;
689 mm->m_pkthdr.header = NULL;
692 /* Append/prepend as many mbuf (clusters) as necessary to fit len. */
693 if (!prep && len > M_TRAILINGSPACE(mm)) {
694 if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA))
697 if (prep && len > M_LEADINGSPACE(mm)) {
698 if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA)))
701 for (x = z; x != NULL; x = x->m_next) {
702 i += x->m_flags & M_EXT ? x->m_ext.ext_size :
703 (x->m_flags & M_PKTHDR ? MHLEN : MLEN);
707 z->m_data += i - len;
708 m_move_pkthdr(mm, z);
713 /* Seek to start position in source mbuf. Optimization for long chains. */
721 /* Copy data into target mbuf. */
724 KASSERT(z != NULL, ("m_copymdata, falling off target edge"));
725 i = M_TRAILINGSPACE(z);
726 m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len);
728 /* fixup pkthdr.len if necessary */
729 if ((prep ? mm : m)->m_flags & M_PKTHDR)
730 (prep ? mm : m)->m_pkthdr.len += i;
735 return (prep ? mm : m);
739 * Copy an entire packet, including header (which must be present).
740 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
741 * Note that the copy is read-only, because clusters are not copied,
742 * only their reference counts are incremented.
743 * Preserve alignment of the first mbuf so if the creator has left
744 * some room at the beginning (e.g. for inserting protocol headers)
745 * the copies still have the room available.
748 m_copypacket(struct mbuf *m, int how)
750 struct mbuf *top, *n, *o;
752 MBUF_CHECKSLEEP(how);
753 MGET(n, how, m->m_type);
758 if (!m_dup_pkthdr(n, m, how))
761 if (m->m_flags & M_EXT) {
762 n->m_data = m->m_data;
765 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
766 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
771 MGET(o, how, m->m_type);
779 if (m->m_flags & M_EXT) {
780 n->m_data = m->m_data;
783 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
791 mbstat.m_mcfail++; /* XXX: No consistency. */
796 * Copy data from an mbuf chain starting "off" bytes from the beginning,
797 * continuing for "len" bytes, into the indicated buffer.
800 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
804 KASSERT(off >= 0, ("m_copydata, negative off %d", off));
805 KASSERT(len >= 0, ("m_copydata, negative len %d", len));
807 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
814 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
815 count = min(m->m_len - off, len);
816 bcopy(mtod(m, caddr_t) + off, cp, count);
825 * Copy a packet header mbuf chain into a completely new chain, including
826 * copying any mbuf clusters. Use this instead of m_copypacket() when
827 * you need a writable copy of an mbuf chain.
830 m_dup(struct mbuf *m, int how)
832 struct mbuf **p, *top = NULL;
833 int remain, moff, nsize;
835 MBUF_CHECKSLEEP(how);
841 /* While there's more data, get a new mbuf, tack it on, and fill it */
842 remain = m->m_pkthdr.len;
845 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
848 /* Get the next new mbuf */
849 if (remain >= MINCLSIZE) {
850 n = m_getcl(how, m->m_type, 0);
853 n = m_get(how, m->m_type);
859 if (top == NULL) { /* First one, must be PKTHDR */
860 if (!m_dup_pkthdr(n, m, how)) {
864 if ((n->m_flags & M_EXT) == 0)
869 /* Link it into the new chain */
873 /* Copy data from original mbuf(s) into new mbuf */
874 while (n->m_len < nsize && m != NULL) {
875 int chunk = min(nsize - n->m_len, m->m_len - moff);
877 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
881 if (moff == m->m_len) {
887 /* Check correct total mbuf length */
888 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
889 ("%s: bogus m_pkthdr.len", __func__));
895 mbstat.m_mcfail++; /* XXX: No consistency. */
900 * Concatenate mbuf chain n to m.
901 * Both chains must be of the same type (e.g. MT_DATA).
902 * Any m_pkthdr is not updated.
905 m_cat(struct mbuf *m, struct mbuf *n)
910 if (m->m_flags & M_EXT ||
911 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
912 /* just join the two chains */
916 /* splat the data from one into the other */
917 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
919 m->m_len += n->m_len;
925 m_adj(struct mbuf *mp, int req_len)
931 if ((m = mp) == NULL)
937 while (m != NULL && len > 0) {
938 if (m->m_len <= len) {
949 if (mp->m_flags & M_PKTHDR)
950 m->m_pkthdr.len -= (req_len - len);
953 * Trim from tail. Scan the mbuf chain,
954 * calculating its length and finding the last mbuf.
955 * If the adjustment only affects this mbuf, then just
956 * adjust and return. Otherwise, rescan and truncate
957 * after the remaining size.
963 if (m->m_next == (struct mbuf *)0)
967 if (m->m_len >= len) {
969 if (mp->m_flags & M_PKTHDR)
970 mp->m_pkthdr.len -= len;
977 * Correct length for chain is "count".
978 * Find the mbuf with last data, adjust its length,
979 * and toss data from remaining mbufs on chain.
982 if (m->m_flags & M_PKTHDR)
983 m->m_pkthdr.len = count;
984 for (; m; m = m->m_next) {
985 if (m->m_len >= count) {
987 if (m->m_next != NULL) {
999 * Rearange an mbuf chain so that len bytes are contiguous
1000 * and in the data area of an mbuf (so that mtod and dtom
1001 * will work for a structure of size len). Returns the resulting
1002 * mbuf chain on success, frees it and returns null on failure.
1003 * If there is room, it will add up to max_protohdr-len extra bytes to the
1004 * contiguous region in an attempt to avoid being called next time.
1007 m_pullup(struct mbuf *n, int len)
1014 * If first mbuf has no cluster, and has room for len bytes
1015 * without shifting current data, pullup into it,
1016 * otherwise allocate a new mbuf to prepend to the chain.
1018 if ((n->m_flags & M_EXT) == 0 &&
1019 n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
1020 if (n->m_len >= len)
1028 MGET(m, M_DONTWAIT, n->m_type);
1032 if (n->m_flags & M_PKTHDR)
1033 M_MOVE_PKTHDR(m, n);
1035 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1037 count = min(min(max(len, max_protohdr), space), n->m_len);
1038 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
1048 } while (len > 0 && n);
1057 mbstat.m_mpfail++; /* XXX: No consistency. */
1062 * Like m_pullup(), except a new mbuf is always allocated, and we allow
1063 * the amount of empty space before the data in the new mbuf to be specified
1064 * (in the event that the caller expects to prepend later).
1069 m_copyup(struct mbuf *n, int len, int dstoff)
1074 if (len > (MHLEN - dstoff))
1076 MGET(m, M_DONTWAIT, n->m_type);
1080 if (n->m_flags & M_PKTHDR)
1081 M_MOVE_PKTHDR(m, n);
1082 m->m_data += dstoff;
1083 space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
1085 count = min(min(max(len, max_protohdr), space), n->m_len);
1086 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
1096 } while (len > 0 && n);
1110 * Partition an mbuf chain in two pieces, returning the tail --
1111 * all but the first len0 bytes. In case of failure, it returns NULL and
1112 * attempts to restore the chain to its original state.
1114 * Note that the resulting mbufs might be read-only, because the new
1115 * mbuf can end up sharing an mbuf cluster with the original mbuf if
1116 * the "breaking point" happens to lie within a cluster mbuf. Use the
1117 * M_WRITABLE() macro to check for this case.
1120 m_split(struct mbuf *m0, int len0, int wait)
1123 u_int len = len0, remain;
1125 MBUF_CHECKSLEEP(wait);
1126 for (m = m0; m && len > m->m_len; m = m->m_next)
1130 remain = m->m_len - len;
1131 if (m0->m_flags & M_PKTHDR) {
1132 MGETHDR(n, wait, m0->m_type);
1135 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
1136 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
1137 m0->m_pkthdr.len = len0;
1138 if (m->m_flags & M_EXT)
1140 if (remain > MHLEN) {
1141 /* m can't be the lead packet */
1143 n->m_next = m_split(m, len, wait);
1144 if (n->m_next == NULL) {
1152 MH_ALIGN(n, remain);
1153 } else if (remain == 0) {
1158 MGET(n, wait, m->m_type);
1164 if (m->m_flags & M_EXT) {
1165 n->m_data = m->m_data + len;
1168 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
1172 n->m_next = m->m_next;
1177 * Routine to copy from device local memory into mbufs.
1178 * Note that `off' argument is offset into first mbuf of target chain from
1179 * which to begin copying the data to.
1182 m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
1183 void (*copy)(char *from, caddr_t to, u_int len))
1186 struct mbuf *top = NULL, **mp = ⊤
1189 if (off < 0 || off > MHLEN)
1192 while (totlen > 0) {
1193 if (top == NULL) { /* First one, must be PKTHDR */
1194 if (totlen + off >= MINCLSIZE) {
1195 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
1198 m = m_gethdr(M_DONTWAIT, MT_DATA);
1201 /* Place initial small packet/header at end of mbuf */
1202 if (m && totlen + off + max_linkhdr <= MLEN) {
1203 m->m_data += max_linkhdr;
1209 m->m_pkthdr.rcvif = ifp;
1210 m->m_pkthdr.len = totlen;
1212 if (totlen + off >= MINCLSIZE) {
1213 m = m_getcl(M_DONTWAIT, MT_DATA, 0);
1216 m = m_get(M_DONTWAIT, MT_DATA);
1229 m->m_len = len = min(totlen, len);
1231 copy(buf, mtod(m, caddr_t), (u_int)len);
1233 bcopy(buf, mtod(m, caddr_t), (u_int)len);
1243 * Copy data from a buffer back into the indicated mbuf chain,
1244 * starting "off" bytes from the beginning, extending the mbuf
1245 * chain if necessary.
1248 m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
1251 struct mbuf *m = m0, *n;
1256 while (off > (mlen = m->m_len)) {
1259 if (m->m_next == NULL) {
1260 n = m_get(M_DONTWAIT, m->m_type);
1263 bzero(mtod(n, caddr_t), MLEN);
1264 n->m_len = min(MLEN, len + off);
1270 if (m->m_next == NULL && (len > m->m_len - off)) {
1271 m->m_len += min(len - (m->m_len - off),
1272 M_TRAILINGSPACE(m));
1274 mlen = min (m->m_len - off, len);
1275 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
1283 if (m->m_next == NULL) {
1284 n = m_get(M_DONTWAIT, m->m_type);
1287 n->m_len = min(MLEN, len);
1292 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
1293 m->m_pkthdr.len = totlen;
1297 * Append the specified data to the indicated mbuf chain,
1298 * Extend the mbuf chain if the new data does not fit in
1301 * Return 1 if able to complete the job; otherwise 0.
1304 m_append(struct mbuf *m0, int len, c_caddr_t cp)
1307 int remainder, space;
1309 for (m = m0; m->m_next != NULL; m = m->m_next)
1312 space = M_TRAILINGSPACE(m);
1315 * Copy into available space.
1317 if (space > remainder)
1319 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1321 cp += space, remainder -= space;
1323 while (remainder > 0) {
1325 * Allocate a new mbuf; could check space
1326 * and allocate a cluster instead.
1328 n = m_get(M_DONTWAIT, m->m_type);
1331 n->m_len = min(MLEN, remainder);
1332 bcopy(cp, mtod(n, caddr_t), n->m_len);
1333 cp += n->m_len, remainder -= n->m_len;
1337 if (m0->m_flags & M_PKTHDR)
1338 m0->m_pkthdr.len += len - remainder;
1339 return (remainder == 0);
1343 * Apply function f to the data in an mbuf chain starting "off" bytes from
1344 * the beginning, continuing for "len" bytes.
1347 m_apply(struct mbuf *m, int off, int len,
1348 int (*f)(void *, void *, u_int), void *arg)
1353 KASSERT(off >= 0, ("m_apply, negative off %d", off));
1354 KASSERT(len >= 0, ("m_apply, negative len %d", len));
1356 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1363 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
1364 count = min(m->m_len - off, len);
1365 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
1376 * Return a pointer to mbuf/offset of location in mbuf chain.
1379 m_getptr(struct mbuf *m, int loc, int *off)
1383 /* Normal end of search. */
1384 if (m->m_len > loc) {
1389 if (m->m_next == NULL) {
1391 /* Point at the end of valid data. */
1404 m_print(const struct mbuf *m, int maxlen)
1408 const struct mbuf *m2;
1410 if (m->m_flags & M_PKTHDR)
1411 len = m->m_pkthdr.len;
1415 while (m2 != NULL && (len == -1 || len)) {
1417 if (maxlen != -1 && pdata > maxlen)
1419 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
1420 m2->m_next, m2->m_flags, "\20\20freelist\17skipfw"
1421 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
1422 "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
1424 printf(", %*D\n", pdata, (u_char *)m2->m_data, "-");
1430 printf("%d bytes unaccounted for.\n", len);
1435 m_fixhdr(struct mbuf *m0)
1439 len = m_length(m0, NULL);
1440 m0->m_pkthdr.len = len;
1445 m_length(struct mbuf *m0, struct mbuf **last)
1451 for (m = m0; m != NULL; m = m->m_next) {
1453 if (m->m_next == NULL)
1462 * Defragment a mbuf chain, returning the shortest possible
1463 * chain of mbufs and clusters. If allocation fails and
1464 * this cannot be completed, NULL will be returned, but
1465 * the passed in chain will be unchanged. Upon success,
1466 * the original chain will be freed, and the new chain
1469 * If a non-packet header is passed in, the original
1470 * mbuf (chain?) will be returned unharmed.
1473 m_defrag(struct mbuf *m0, int how)
1475 struct mbuf *m_new = NULL, *m_final = NULL;
1476 int progress = 0, length;
1478 MBUF_CHECKSLEEP(how);
1479 if (!(m0->m_flags & M_PKTHDR))
1482 m_fixhdr(m0); /* Needed sanity check */
1484 #ifdef MBUF_STRESS_TEST
1485 if (m_defragrandomfailures) {
1486 int temp = arc4random() & 0xff;
1492 if (m0->m_pkthdr.len > MHLEN)
1493 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1495 m_final = m_gethdr(how, MT_DATA);
1497 if (m_final == NULL)
1500 if (m_dup_pkthdr(m_final, m0, how) == 0)
1505 while (progress < m0->m_pkthdr.len) {
1506 length = m0->m_pkthdr.len - progress;
1507 if (length > MCLBYTES)
1510 if (m_new == NULL) {
1512 m_new = m_getcl(how, MT_DATA, 0);
1514 m_new = m_get(how, MT_DATA);
1519 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
1521 m_new->m_len = length;
1522 if (m_new != m_final)
1523 m_cat(m_final, m_new);
1526 #ifdef MBUF_STRESS_TEST
1527 if (m0->m_next == NULL)
1532 #ifdef MBUF_STRESS_TEST
1534 m_defragbytes += m0->m_pkthdr.len;
1538 #ifdef MBUF_STRESS_TEST
1547 * Defragment an mbuf chain, returning at most maxfrags separate
1548 * mbufs+clusters. If this is not possible NULL is returned and
1549 * the original mbuf chain is left in it's present (potentially
1550 * modified) state. We use two techniques: collapsing consecutive
1551 * mbufs and replacing consecutive mbufs by a cluster.
1553 * NB: this should really be named m_defrag but that name is taken
1556 m_collapse(struct mbuf *m0, int how, int maxfrags)
1558 struct mbuf *m, *n, *n2, **prev;
1562 * Calculate the current number of frags.
1565 for (m = m0; m != NULL; m = m->m_next)
1568 * First, try to collapse mbufs. Note that we always collapse
1569 * towards the front so we don't need to deal with moving the
1570 * pkthdr. This may be suboptimal if the first mbuf has much
1571 * less data than the following.
1579 if ((m->m_flags & M_RDONLY) == 0 &&
1580 n->m_len < M_TRAILINGSPACE(m)) {
1581 bcopy(mtod(n, void *), mtod(m, char *) + m->m_len,
1583 m->m_len += n->m_len;
1584 m->m_next = n->m_next;
1586 if (--curfrags <= maxfrags)
1591 KASSERT(maxfrags > 1,
1592 ("maxfrags %u, but normal collapse failed", maxfrags));
1594 * Collapse consecutive mbufs to a cluster.
1596 prev = &m0->m_next; /* NB: not the first mbuf */
1597 while ((n = *prev) != NULL) {
1598 if ((n2 = n->m_next) != NULL &&
1599 n->m_len + n2->m_len < MCLBYTES) {
1600 m = m_getcl(how, MT_DATA, 0);
1603 bcopy(mtod(n, void *), mtod(m, void *), n->m_len);
1604 bcopy(mtod(n2, void *), mtod(m, char *) + n->m_len,
1606 m->m_len = n->m_len + n2->m_len;
1607 m->m_next = n2->m_next;
1611 if (--curfrags <= maxfrags) /* +1 cl -2 mbufs */
1614 * Still not there, try the normal collapse
1615 * again before we allocate another cluster.
1622 * No place where we can collapse to a cluster; punt.
1623 * This can occur if, for example, you request 2 frags
1624 * but the packet requires that both be clusters (we
1625 * never reallocate the first mbuf to avoid moving the
1632 #ifdef MBUF_STRESS_TEST
1635 * Fragment an mbuf chain. There's no reason you'd ever want to do
1636 * this in normal usage, but it's great for stress testing various
1639 * If fragmentation is not possible, the original chain will be
1642 * Possible length values:
1643 * 0 no fragmentation will occur
1644 * > 0 each fragment will be of the specified length
1645 * -1 each fragment will be the same random value in length
1646 * -2 each fragment's length will be entirely random
1647 * (Random values range from 1 to 256)
1650 m_fragment(struct mbuf *m0, int how, int length)
1652 struct mbuf *m_new = NULL, *m_final = NULL;
1655 if (!(m0->m_flags & M_PKTHDR))
1658 if ((length == 0) || (length < -2))
1661 m_fixhdr(m0); /* Needed sanity check */
1663 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
1665 if (m_final == NULL)
1668 if (m_dup_pkthdr(m_final, m0, how) == 0)
1674 length = 1 + (arc4random() & 255);
1676 while (progress < m0->m_pkthdr.len) {
1682 fraglen = 1 + (arc4random() & 255);
1683 if (fraglen > m0->m_pkthdr.len - progress)
1684 fraglen = m0->m_pkthdr.len - progress;
1686 if (fraglen > MCLBYTES)
1689 if (m_new == NULL) {
1690 m_new = m_getcl(how, MT_DATA, 0);
1695 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t));
1696 progress += fraglen;
1697 m_new->m_len = fraglen;
1698 if (m_new != m_final)
1699 m_cat(m_final, m_new);
1708 /* Return the original chain on failure */
1715 * Copy the contents of uio into a properly sized mbuf chain.
1718 m_uiotombuf(struct uio *uio, int how, int len, int align, int flags)
1720 struct mbuf *m, *mb;
1721 int error, length, total;
1725 * len can be zero or an arbitrary large value bound by
1726 * the total data supplied by the uio.
1729 total = min(uio->uio_resid, len);
1731 total = uio->uio_resid;
1734 * The smallest unit returned by m_getm2() is a single mbuf
1735 * with pkthdr. We can't align past it.
1741 * Give us the full allocation or nothing.
1742 * If len is zero return the smallest empty mbuf.
1744 m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags);
1749 /* Fill all mbufs with uio data and update header information. */
1750 for (mb = m; mb != NULL; mb = mb->m_next) {
1751 length = min(M_TRAILINGSPACE(mb), total - progress);
1753 error = uiomove(mtod(mb, void *), length, uio);
1761 if (flags & M_PKTHDR)
1762 m->m_pkthdr.len += length;
1764 KASSERT(progress == total, ("%s: progress != total", __func__));
1770 * Set the m_data pointer of a newly-allocated mbuf
1771 * to place an object of the specified size at the
1772 * end of the mbuf, longword aligned.
1775 m_align(struct mbuf *m, int len)
1779 if (m->m_flags & M_EXT)
1780 adjust = m->m_ext.ext_size - len;
1781 else if (m->m_flags & M_PKTHDR)
1782 adjust = MHLEN - len;
1784 adjust = MLEN - len;
1785 m->m_data += adjust &~ (sizeof(long)-1);
1789 * Create a writable copy of the mbuf chain. While doing this
1790 * we compact the chain with a goal of producing a chain with
1791 * at most two mbufs. The second mbuf in this chain is likely
1792 * to be a cluster. The primary purpose of this work is to create
1793 * a writable packet for encryption, compression, etc. The
1794 * secondary goal is to linearize the data so the data can be
1795 * passed to crypto hardware in the most efficient manner possible.
1798 m_unshare(struct mbuf *m0, int how)
1800 struct mbuf *m, *mprev;
1801 struct mbuf *n, *mfirst, *mlast;
1805 for (m = m0; m != NULL; m = mprev->m_next) {
1807 * Regular mbufs are ignored unless there's a cluster
1808 * in front of it that we can use to coalesce. We do
1809 * the latter mainly so later clusters can be coalesced
1810 * also w/o having to handle them specially (i.e. convert
1811 * mbuf+cluster -> cluster). This optimization is heavily
1812 * influenced by the assumption that we're running over
1813 * Ethernet where MCLBYTES is large enough that the max
1814 * packet size will permit lots of coalescing into a
1815 * single cluster. This in turn permits efficient
1816 * crypto operations, especially when using hardware.
1818 if ((m->m_flags & M_EXT) == 0) {
1819 if (mprev && (mprev->m_flags & M_EXT) &&
1820 m->m_len <= M_TRAILINGSPACE(mprev)) {
1821 /* XXX: this ignores mbuf types */
1822 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1823 mtod(m, caddr_t), m->m_len);
1824 mprev->m_len += m->m_len;
1825 mprev->m_next = m->m_next; /* unlink from chain */
1826 m_free(m); /* reclaim mbuf */
1828 newipsecstat.ips_mbcoalesced++;
1836 * Writable mbufs are left alone (for now).
1838 if (M_WRITABLE(m)) {
1844 * Not writable, replace with a copy or coalesce with
1845 * the previous mbuf if possible (since we have to copy
1846 * it anyway, we try to reduce the number of mbufs and
1847 * clusters so that future work is easier).
1849 KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags));
1850 /* NB: we only coalesce into a cluster or larger */
1851 if (mprev != NULL && (mprev->m_flags & M_EXT) &&
1852 m->m_len <= M_TRAILINGSPACE(mprev)) {
1853 /* XXX: this ignores mbuf types */
1854 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
1855 mtod(m, caddr_t), m->m_len);
1856 mprev->m_len += m->m_len;
1857 mprev->m_next = m->m_next; /* unlink from chain */
1858 m_free(m); /* reclaim mbuf */
1860 newipsecstat.ips_clcoalesced++;
1866 * Allocate new space to hold the copy...
1868 /* XXX why can M_PKTHDR be set past the first mbuf? */
1869 if (mprev == NULL && (m->m_flags & M_PKTHDR)) {
1871 * NB: if a packet header is present we must
1872 * allocate the mbuf separately from any cluster
1873 * because M_MOVE_PKTHDR will smash the data
1874 * pointer and drop the M_EXT marker.
1876 MGETHDR(n, how, m->m_type);
1881 M_MOVE_PKTHDR(n, m);
1883 if ((n->m_flags & M_EXT) == 0) {
1889 n = m_getcl(how, m->m_type, m->m_flags);
1896 * ... and copy the data. We deal with jumbo mbufs
1897 * (i.e. m_len > MCLBYTES) by splitting them into
1898 * clusters. We could just malloc a buffer and make
1899 * it external but too many device drivers don't know
1900 * how to break up the non-contiguous memory when
1908 int cc = min(len, MCLBYTES);
1909 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc);
1915 newipsecstat.ips_clcopied++;
1923 n = m_getcl(how, m->m_type, m->m_flags);
1930 n->m_next = m->m_next;
1932 m0 = mfirst; /* new head of chain */
1934 mprev->m_next = mfirst; /* replace old mbuf */
1935 m_free(m); /* release old mbuf */