freebsd-dev/sys/kern/uipc_mbuf.c

713 lines
17 KiB
C
Raw Normal View History

1994-05-24 10:09:53 +00:00
/*
* Copyright (c) 1982, 1986, 1988, 1991, 1993
* The Regents of the University of California. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by the University of
* California, Berkeley and its contributors.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94
1999-08-28 01:08:13 +00:00
* $FreeBSD$
1994-05-24 10:09:53 +00:00
*/
#include "opt_mac.h"
#include "opt_param.h"
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/mac.h>
#include <sys/malloc.h>
1994-05-24 10:09:53 +00:00
#include <sys/mbuf.h>
#include <sys/sysctl.h>
1994-05-24 10:09:53 +00:00
#include <sys/domain.h>
#include <sys/protosw.h>
int max_linkhdr;
int max_protohdr;
int max_hdr;
int max_datalen;
2000-09-30 06:30:39 +00:00
/*
* sysctl(8) exported objects
*/
SYSCTL_DECL(_kern_ipc);
SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW,
&max_linkhdr, 0, "");
SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW,
&max_protohdr, 0, "");
SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, "");
SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW,
&max_datalen, 0, "");
1994-05-24 10:09:53 +00:00
/*
* Copy mbuf pkthdr from "from" to "to".
* "from" must have M_PKTHDR set, and "to" must be empty.
* aux pointer will be moved to "to".
*/
void
m_copy_pkthdr(struct mbuf *to, struct mbuf *from)
{
#if 0
KASSERT(to->m_flags & M_PKTHDR,
("m_copy_pkthdr() called on non-header"));
#endif
#ifdef MAC
if (to->m_flags & M_PKTHDR)
mac_destroy_mbuf(to);
#endif
to->m_data = to->m_pktdat;
to->m_flags = from->m_flags & M_COPYFLAGS;
to->m_pkthdr = from->m_pkthdr;
#ifdef MAC
mac_init_mbuf(to, 1); /* XXXMAC no way to fail */
mac_create_mbuf_from_mbuf(from, to);
#endif
from->m_pkthdr.aux = NULL;
}
1994-05-24 10:09:53 +00:00
/*
* Lesser-used path for M_PREPEND:
* allocate new mbuf to prepend to chain,
* copy junk along.
*/
struct mbuf *
m_prepend(struct mbuf *m, int len, int how)
1994-05-24 10:09:53 +00:00
{
struct mbuf *mn;
MGET(mn, how, m->m_type);
if (mn == NULL) {
1994-05-24 10:09:53 +00:00
m_freem(m);
return (NULL);
1994-05-24 10:09:53 +00:00
}
if (m->m_flags & M_PKTHDR) {
M_COPY_PKTHDR(mn, m);
#ifdef MAC
mac_destroy_mbuf(m);
#endif
1994-05-24 10:09:53 +00:00
m->m_flags &= ~M_PKTHDR;
}
mn->m_next = m;
m = mn;
if (len < MHLEN)
MH_ALIGN(m, len);
m->m_len = len;
return (m);
}
/*
* Make a copy of an mbuf chain starting "off0" bytes from the beginning,
* continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf.
* The wait parameter is a choice of M_TRYWAIT/M_DONTWAIT from caller.
* Note that the copy is read-only, because clusters are not copied,
* only their reference counts are incremented.
1994-05-24 10:09:53 +00:00
*/
struct mbuf *
m_copym(struct mbuf *m, int off0, int len, int wait)
1994-05-24 10:09:53 +00:00
{
struct mbuf *n, **np;
int off = off0;
1994-05-24 10:09:53 +00:00
struct mbuf *top;
int copyhdr = 0;
KASSERT(off >= 0, ("m_copym, negative off %d", off));
KASSERT(len >= 0, ("m_copym, negative len %d", len));
1994-05-24 10:09:53 +00:00
if (off == 0 && m->m_flags & M_PKTHDR)
copyhdr = 1;
while (off > 0) {
KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
1994-05-24 10:09:53 +00:00
if (off < m->m_len)
break;
off -= m->m_len;
m = m->m_next;
}
np = &top;
top = 0;
while (len > 0) {
if (m == NULL) {
KASSERT(len == M_COPYALL,
("m_copym, length > size of mbuf chain"));
1994-05-24 10:09:53 +00:00
break;
}
MGET(n, wait, m->m_type);
*np = n;
if (n == NULL)
1994-05-24 10:09:53 +00:00
goto nospace;
if (copyhdr) {
M_COPY_PKTHDR(n, m);
if (len == M_COPYALL)
n->m_pkthdr.len -= off0;
else
n->m_pkthdr.len = len;
copyhdr = 0;
}
n->m_len = min(len, m->m_len - off);
if (m->m_flags & M_EXT) {
n->m_data = m->m_data + off;
n->m_ext = m->m_ext;
n->m_flags |= M_EXT;
Replace the mbuf external reference counting code with something that should be better. The old code counted references to mbuf clusters by using the offset of the cluster from the start of memory allocated for mbufs and clusters as an index into an array of chars, which did the reference counting. If the external storage was not a cluster then reference counting had to be done by the code using that external storage. NetBSD's system of linked lists of mbufs was cosidered, but Alfred felt it would have locking issues when the kernel was made more SMP friendly. The system implimented uses a pool of unions to track external storage. The union contains an int for counting the references and a pointer for forming a free list. The reference counts are incremented and decremented atomically and so should be SMP friendly. This system can track reference counts for any sort of external storage. Access to the reference counting stuff is now through macros defined in mbuf.h, so it should be easier to make changes to the system in the future. The possibility of storing the reference count in one of the referencing mbufs was considered, but was rejected 'cos it would often leave extra mbufs allocated. Storing the reference count in the cluster was also considered, but because the external storage may not be a cluster this isn't an option. The size of the pool of reference counters is available in the stats provided by "netstat -m". PR: 19866 Submitted by: Bosko Milekic <bmilekic@dsuper.net> Reviewed by: alfred (glanced at by others on -net)
2000-08-19 08:32:59 +00:00
MEXT_ADD_REF(m);
1994-05-24 10:09:53 +00:00
} else
bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
(unsigned)n->m_len);
if (len != M_COPYALL)
len -= n->m_len;
off = 0;
m = m->m_next;
np = &n->m_next;
}
Introduce numerous SMP friendly changes to the mbuf allocator. Namely, introduce a modified allocation mechanism for mbufs and mbuf clusters; one which can scale under SMP and which offers the possibility of resource reclamation to be implemented in the future. Notable advantages: o Reduce contention for SMP by offering per-CPU pools and locks. o Better use of data cache due to per-CPU pools. o Much less code cache pollution due to excessively large allocation macros. o Framework for `grouping' objects from same page together so as to be able to possibly free wired-down pages back to the system if they are no longer needed by the network stacks. Additional things changed with this addition: - Moved some mbuf specific declarations and initializations from sys/conf/param.c into mbuf-specific code where they belong. - m_getclr() has been renamed to m_get_clrd() because the old name is really confusing. m_getclr() HAS been preserved though and is defined to the new name. No tree sweep has been done "to change the interface," as the old name will continue to be supported and is not depracated. The change was merely done because m_getclr() sounds too much like "m_get a cluster." - TEMPORARILY disabled mbtypes statistics displaying in netstat(1) and systat(1) (see TODO below). - Fixed systat(1) to display number of "free mbufs" based on new per-CPU stat structures. - Fixed netstat(1) to display new per-CPU stats based on sysctl-exported per-CPU stat structures. All infos are fetched via sysctl. TODO (in order of priority): - Re-enable mbtypes statistics in both netstat(1) and systat(1) after introducing an SMP friendly way to collect the mbtypes stats under the already introduced per-CPU locks (i.e. hopefully don't use atomic() - it seems too costly for a mere stat update, especially when other locks are already present). - Optionally have systat(1) display not only "total free mbufs" but also "total free mbufs per CPU pool." - Fix minor length-fetching issues in netstat(1) related to recently re-enabled option to read mbuf stats from a core file. - Move reference counters at least for mbuf clusters into an unused portion of the cluster itself, to save space and need to allocate a counter. - Look into introducing resource freeing possibly from a kproc. Reviewed by (in parts): jlemon, jake, silby, terry Tested by: jlemon (Intel & Alpha), mjacob (Intel & Alpha) Preliminary performance measurements: jlemon (and me, obviously) URL: http://people.freebsd.org/~bmilekic/mb_alloc/
2001-06-22 06:35:32 +00:00
if (top == NULL)
mbstat.m_mcfail++; /* XXX: No consistency. */
1994-05-24 10:09:53 +00:00
return (top);
nospace:
m_freem(top);
Introduce numerous SMP friendly changes to the mbuf allocator. Namely, introduce a modified allocation mechanism for mbufs and mbuf clusters; one which can scale under SMP and which offers the possibility of resource reclamation to be implemented in the future. Notable advantages: o Reduce contention for SMP by offering per-CPU pools and locks. o Better use of data cache due to per-CPU pools. o Much less code cache pollution due to excessively large allocation macros. o Framework for `grouping' objects from same page together so as to be able to possibly free wired-down pages back to the system if they are no longer needed by the network stacks. Additional things changed with this addition: - Moved some mbuf specific declarations and initializations from sys/conf/param.c into mbuf-specific code where they belong. - m_getclr() has been renamed to m_get_clrd() because the old name is really confusing. m_getclr() HAS been preserved though and is defined to the new name. No tree sweep has been done "to change the interface," as the old name will continue to be supported and is not depracated. The change was merely done because m_getclr() sounds too much like "m_get a cluster." - TEMPORARILY disabled mbtypes statistics displaying in netstat(1) and systat(1) (see TODO below). - Fixed systat(1) to display number of "free mbufs" based on new per-CPU stat structures. - Fixed netstat(1) to display new per-CPU stats based on sysctl-exported per-CPU stat structures. All infos are fetched via sysctl. TODO (in order of priority): - Re-enable mbtypes statistics in both netstat(1) and systat(1) after introducing an SMP friendly way to collect the mbtypes stats under the already introduced per-CPU locks (i.e. hopefully don't use atomic() - it seems too costly for a mere stat update, especially when other locks are already present). - Optionally have systat(1) display not only "total free mbufs" but also "total free mbufs per CPU pool." - Fix minor length-fetching issues in netstat(1) related to recently re-enabled option to read mbuf stats from a core file. - Move reference counters at least for mbuf clusters into an unused portion of the cluster itself, to save space and need to allocate a counter. - Look into introducing resource freeing possibly from a kproc. Reviewed by (in parts): jlemon, jake, silby, terry Tested by: jlemon (Intel & Alpha), mjacob (Intel & Alpha) Preliminary performance measurements: jlemon (and me, obviously) URL: http://people.freebsd.org/~bmilekic/mb_alloc/
2001-06-22 06:35:32 +00:00
mbstat.m_mcfail++; /* XXX: No consistency. */
return (NULL);
1994-05-24 10:09:53 +00:00
}
/*
* Copy an entire packet, including header (which must be present).
* An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
* Note that the copy is read-only, because clusters are not copied,
* only their reference counts are incremented.
* Preserve alignment of the first mbuf so if the creator has left
* some room at the beginning (e.g. for inserting protocol headers)
* the copies still have the room available.
*/
struct mbuf *
m_copypacket(struct mbuf *m, int how)
{
struct mbuf *top, *n, *o;
MGET(n, how, m->m_type);
top = n;
if (n == NULL)
goto nospace;
M_COPY_PKTHDR(n, m);
n->m_len = m->m_len;
if (m->m_flags & M_EXT) {
n->m_data = m->m_data;
n->m_ext = m->m_ext;
n->m_flags |= M_EXT;
Replace the mbuf external reference counting code with something that should be better. The old code counted references to mbuf clusters by using the offset of the cluster from the start of memory allocated for mbufs and clusters as an index into an array of chars, which did the reference counting. If the external storage was not a cluster then reference counting had to be done by the code using that external storage. NetBSD's system of linked lists of mbufs was cosidered, but Alfred felt it would have locking issues when the kernel was made more SMP friendly. The system implimented uses a pool of unions to track external storage. The union contains an int for counting the references and a pointer for forming a free list. The reference counts are incremented and decremented atomically and so should be SMP friendly. This system can track reference counts for any sort of external storage. Access to the reference counting stuff is now through macros defined in mbuf.h, so it should be easier to make changes to the system in the future. The possibility of storing the reference count in one of the referencing mbufs was considered, but was rejected 'cos it would often leave extra mbufs allocated. Storing the reference count in the cluster was also considered, but because the external storage may not be a cluster this isn't an option. The size of the pool of reference counters is available in the stats provided by "netstat -m". PR: 19866 Submitted by: Bosko Milekic <bmilekic@dsuper.net> Reviewed by: alfred (glanced at by others on -net)
2000-08-19 08:32:59 +00:00
MEXT_ADD_REF(m);
} else {
n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
}
m = m->m_next;
while (m) {
MGET(o, how, m->m_type);
if (o == NULL)
goto nospace;
n->m_next = o;
n = n->m_next;
n->m_len = m->m_len;
if (m->m_flags & M_EXT) {
n->m_data = m->m_data;
n->m_ext = m->m_ext;
n->m_flags |= M_EXT;
Replace the mbuf external reference counting code with something that should be better. The old code counted references to mbuf clusters by using the offset of the cluster from the start of memory allocated for mbufs and clusters as an index into an array of chars, which did the reference counting. If the external storage was not a cluster then reference counting had to be done by the code using that external storage. NetBSD's system of linked lists of mbufs was cosidered, but Alfred felt it would have locking issues when the kernel was made more SMP friendly. The system implimented uses a pool of unions to track external storage. The union contains an int for counting the references and a pointer for forming a free list. The reference counts are incremented and decremented atomically and so should be SMP friendly. This system can track reference counts for any sort of external storage. Access to the reference counting stuff is now through macros defined in mbuf.h, so it should be easier to make changes to the system in the future. The possibility of storing the reference count in one of the referencing mbufs was considered, but was rejected 'cos it would often leave extra mbufs allocated. Storing the reference count in the cluster was also considered, but because the external storage may not be a cluster this isn't an option. The size of the pool of reference counters is available in the stats provided by "netstat -m". PR: 19866 Submitted by: Bosko Milekic <bmilekic@dsuper.net> Reviewed by: alfred (glanced at by others on -net)
2000-08-19 08:32:59 +00:00
MEXT_ADD_REF(m);
} else {
bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
}
m = m->m_next;
}
return top;
nospace:
m_freem(top);
Introduce numerous SMP friendly changes to the mbuf allocator. Namely, introduce a modified allocation mechanism for mbufs and mbuf clusters; one which can scale under SMP and which offers the possibility of resource reclamation to be implemented in the future. Notable advantages: o Reduce contention for SMP by offering per-CPU pools and locks. o Better use of data cache due to per-CPU pools. o Much less code cache pollution due to excessively large allocation macros. o Framework for `grouping' objects from same page together so as to be able to possibly free wired-down pages back to the system if they are no longer needed by the network stacks. Additional things changed with this addition: - Moved some mbuf specific declarations and initializations from sys/conf/param.c into mbuf-specific code where they belong. - m_getclr() has been renamed to m_get_clrd() because the old name is really confusing. m_getclr() HAS been preserved though and is defined to the new name. No tree sweep has been done "to change the interface," as the old name will continue to be supported and is not depracated. The change was merely done because m_getclr() sounds too much like "m_get a cluster." - TEMPORARILY disabled mbtypes statistics displaying in netstat(1) and systat(1) (see TODO below). - Fixed systat(1) to display number of "free mbufs" based on new per-CPU stat structures. - Fixed netstat(1) to display new per-CPU stats based on sysctl-exported per-CPU stat structures. All infos are fetched via sysctl. TODO (in order of priority): - Re-enable mbtypes statistics in both netstat(1) and systat(1) after introducing an SMP friendly way to collect the mbtypes stats under the already introduced per-CPU locks (i.e. hopefully don't use atomic() - it seems too costly for a mere stat update, especially when other locks are already present). - Optionally have systat(1) display not only "total free mbufs" but also "total free mbufs per CPU pool." - Fix minor length-fetching issues in netstat(1) related to recently re-enabled option to read mbuf stats from a core file. - Move reference counters at least for mbuf clusters into an unused portion of the cluster itself, to save space and need to allocate a counter. - Look into introducing resource freeing possibly from a kproc. Reviewed by (in parts): jlemon, jake, silby, terry Tested by: jlemon (Intel & Alpha), mjacob (Intel & Alpha) Preliminary performance measurements: jlemon (and me, obviously) URL: http://people.freebsd.org/~bmilekic/mb_alloc/
2001-06-22 06:35:32 +00:00
mbstat.m_mcfail++; /* XXX: No consistency. */
return (NULL);
}
1994-05-24 10:09:53 +00:00
/*
* Copy data from an mbuf chain starting "off" bytes from the beginning,
* continuing for "len" bytes, into the indicated buffer.
*/
void
m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
1994-05-24 10:09:53 +00:00
{
unsigned count;
1994-05-24 10:09:53 +00:00
KASSERT(off >= 0, ("m_copydata, negative off %d", off));
KASSERT(len >= 0, ("m_copydata, negative len %d", len));
1994-05-24 10:09:53 +00:00
while (off > 0) {
KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
1994-05-24 10:09:53 +00:00
if (off < m->m_len)
break;
off -= m->m_len;
m = m->m_next;
}
while (len > 0) {
KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
1994-05-24 10:09:53 +00:00
count = min(m->m_len - off, len);
bcopy(mtod(m, caddr_t) + off, cp, count);
len -= count;
cp += count;
off = 0;
m = m->m_next;
}
}
/*
* Copy a packet header mbuf chain into a completely new chain, including
* copying any mbuf clusters. Use this instead of m_copypacket() when
* you need a writable copy of an mbuf chain.
*/
struct mbuf *
m_dup(struct mbuf *m, int how)
{
struct mbuf **p, *top = NULL;
int remain, moff, nsize;
/* Sanity check */
if (m == NULL)
return (NULL);
KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __func__));
/* While there's more data, get a new mbuf, tack it on, and fill it */
remain = m->m_pkthdr.len;
moff = 0;
p = &top;
while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */
struct mbuf *n;
/* Get the next new mbuf */
MGET(n, how, m->m_type);
if (n == NULL)
goto nospace;
if (top == NULL) { /* first one, must be PKTHDR */
M_COPY_PKTHDR(n, m);
nsize = MHLEN;
} else /* not the first one */
nsize = MLEN;
if (remain >= MINCLSIZE) {
MCLGET(n, how);
if ((n->m_flags & M_EXT) == 0) {
(void)m_free(n);
goto nospace;
}
nsize = MCLBYTES;
}
n->m_len = 0;
/* Link it into the new chain */
*p = n;
p = &n->m_next;
/* Copy data from original mbuf(s) into new mbuf */
while (n->m_len < nsize && m != NULL) {
int chunk = min(nsize - n->m_len, m->m_len - moff);
bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
moff += chunk;
n->m_len += chunk;
remain -= chunk;
if (moff == m->m_len) {
m = m->m_next;
moff = 0;
}
}
/* Check correct total mbuf length */
KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
("%s: bogus m_pkthdr.len", __func__));
}
return (top);
nospace:
m_freem(top);
Introduce numerous SMP friendly changes to the mbuf allocator. Namely, introduce a modified allocation mechanism for mbufs and mbuf clusters; one which can scale under SMP and which offers the possibility of resource reclamation to be implemented in the future. Notable advantages: o Reduce contention for SMP by offering per-CPU pools and locks. o Better use of data cache due to per-CPU pools. o Much less code cache pollution due to excessively large allocation macros. o Framework for `grouping' objects from same page together so as to be able to possibly free wired-down pages back to the system if they are no longer needed by the network stacks. Additional things changed with this addition: - Moved some mbuf specific declarations and initializations from sys/conf/param.c into mbuf-specific code where they belong. - m_getclr() has been renamed to m_get_clrd() because the old name is really confusing. m_getclr() HAS been preserved though and is defined to the new name. No tree sweep has been done "to change the interface," as the old name will continue to be supported and is not depracated. The change was merely done because m_getclr() sounds too much like "m_get a cluster." - TEMPORARILY disabled mbtypes statistics displaying in netstat(1) and systat(1) (see TODO below). - Fixed systat(1) to display number of "free mbufs" based on new per-CPU stat structures. - Fixed netstat(1) to display new per-CPU stats based on sysctl-exported per-CPU stat structures. All infos are fetched via sysctl. TODO (in order of priority): - Re-enable mbtypes statistics in both netstat(1) and systat(1) after introducing an SMP friendly way to collect the mbtypes stats under the already introduced per-CPU locks (i.e. hopefully don't use atomic() - it seems too costly for a mere stat update, especially when other locks are already present). - Optionally have systat(1) display not only "total free mbufs" but also "total free mbufs per CPU pool." - Fix minor length-fetching issues in netstat(1) related to recently re-enabled option to read mbuf stats from a core file. - Move reference counters at least for mbuf clusters into an unused portion of the cluster itself, to save space and need to allocate a counter. - Look into introducing resource freeing possibly from a kproc. Reviewed by (in parts): jlemon, jake, silby, terry Tested by: jlemon (Intel & Alpha), mjacob (Intel & Alpha) Preliminary performance measurements: jlemon (and me, obviously) URL: http://people.freebsd.org/~bmilekic/mb_alloc/
2001-06-22 06:35:32 +00:00
mbstat.m_mcfail++; /* XXX: No consistency. */
return (NULL);
}
1994-05-24 10:09:53 +00:00
/*
* Concatenate mbuf chain n to m.
* Both chains must be of the same type (e.g. MT_DATA).
* Any m_pkthdr is not updated.
*/
void
m_cat(struct mbuf *m, struct mbuf *n)
1994-05-24 10:09:53 +00:00
{
while (m->m_next)
m = m->m_next;
while (n) {
if (m->m_flags & M_EXT ||
m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
/* just join the two chains */
m->m_next = n;
return;
}
/* splat the data from one into the other */
bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
(u_int)n->m_len);
m->m_len += n->m_len;
n = m_free(n);
}
}
void
m_adj(struct mbuf *mp, int req_len)
1994-05-24 10:09:53 +00:00
{
int len = req_len;
struct mbuf *m;
int count;
1994-05-24 10:09:53 +00:00
if ((m = mp) == NULL)
return;
if (len >= 0) {
/*
* Trim from head.
*/
while (m != NULL && len > 0) {
if (m->m_len <= len) {
len -= m->m_len;
m->m_len = 0;
m = m->m_next;
} else {
m->m_len -= len;
m->m_data += len;
len = 0;
}
}
m = mp;
if (mp->m_flags & M_PKTHDR)
m->m_pkthdr.len -= (req_len - len);
} else {
/*
* Trim from tail. Scan the mbuf chain,
* calculating its length and finding the last mbuf.
* If the adjustment only affects this mbuf, then just
* adjust and return. Otherwise, rescan and truncate
* after the remaining size.
*/
len = -len;
count = 0;
for (;;) {
count += m->m_len;
if (m->m_next == (struct mbuf *)0)
break;
m = m->m_next;
}
if (m->m_len >= len) {
m->m_len -= len;
if (mp->m_flags & M_PKTHDR)
mp->m_pkthdr.len -= len;
return;
}
count -= len;
if (count < 0)
count = 0;
/*
* Correct length for chain is "count".
* Find the mbuf with last data, adjust its length,
* and toss data from remaining mbufs on chain.
*/
m = mp;
if (m->m_flags & M_PKTHDR)
m->m_pkthdr.len = count;
for (; m; m = m->m_next) {
if (m->m_len >= count) {
m->m_len = count;
break;
}
count -= m->m_len;
}
while (m->m_next)
(m = m->m_next) ->m_len = 0;
1994-05-24 10:09:53 +00:00
}
}
/*
* Rearange an mbuf chain so that len bytes are contiguous
* and in the data area of an mbuf (so that mtod and dtom
* will work for a structure of size len). Returns the resulting
* mbuf chain on success, frees it and returns null on failure.
* If there is room, it will add up to max_protohdr-len extra bytes to the
* contiguous region in an attempt to avoid being called next time.
*/
struct mbuf *
m_pullup(struct mbuf *n, int len)
1994-05-24 10:09:53 +00:00
{
struct mbuf *m;
int count;
1994-05-24 10:09:53 +00:00
int space;
/*
* If first mbuf has no cluster, and has room for len bytes
* without shifting current data, pullup into it,
* otherwise allocate a new mbuf to prepend to the chain.
*/
if ((n->m_flags & M_EXT) == 0 &&
n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
if (n->m_len >= len)
return (n);
m = n;
n = n->m_next;
len -= m->m_len;
} else {
if (len > MHLEN)
goto bad;
MGET(m, M_DONTWAIT, n->m_type);
if (m == NULL)
1994-05-24 10:09:53 +00:00
goto bad;
m->m_len = 0;
if (n->m_flags & M_PKTHDR) {
M_COPY_PKTHDR(m, n);
n->m_flags &= ~M_PKTHDR;
}
}
space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
do {
count = min(min(max(len, max_protohdr), space), n->m_len);
bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
(unsigned)count);
len -= count;
m->m_len += count;
n->m_len -= count;
space -= count;
if (n->m_len)
n->m_data += count;
else
n = m_free(n);
} while (len > 0 && n);
if (len > 0) {
(void) m_free(m);
goto bad;
}
m->m_next = n;
return (m);
bad:
m_freem(n);
Introduce numerous SMP friendly changes to the mbuf allocator. Namely, introduce a modified allocation mechanism for mbufs and mbuf clusters; one which can scale under SMP and which offers the possibility of resource reclamation to be implemented in the future. Notable advantages: o Reduce contention for SMP by offering per-CPU pools and locks. o Better use of data cache due to per-CPU pools. o Much less code cache pollution due to excessively large allocation macros. o Framework for `grouping' objects from same page together so as to be able to possibly free wired-down pages back to the system if they are no longer needed by the network stacks. Additional things changed with this addition: - Moved some mbuf specific declarations and initializations from sys/conf/param.c into mbuf-specific code where they belong. - m_getclr() has been renamed to m_get_clrd() because the old name is really confusing. m_getclr() HAS been preserved though and is defined to the new name. No tree sweep has been done "to change the interface," as the old name will continue to be supported and is not depracated. The change was merely done because m_getclr() sounds too much like "m_get a cluster." - TEMPORARILY disabled mbtypes statistics displaying in netstat(1) and systat(1) (see TODO below). - Fixed systat(1) to display number of "free mbufs" based on new per-CPU stat structures. - Fixed netstat(1) to display new per-CPU stats based on sysctl-exported per-CPU stat structures. All infos are fetched via sysctl. TODO (in order of priority): - Re-enable mbtypes statistics in both netstat(1) and systat(1) after introducing an SMP friendly way to collect the mbtypes stats under the already introduced per-CPU locks (i.e. hopefully don't use atomic() - it seems too costly for a mere stat update, especially when other locks are already present). - Optionally have systat(1) display not only "total free mbufs" but also "total free mbufs per CPU pool." - Fix minor length-fetching issues in netstat(1) related to recently re-enabled option to read mbuf stats from a core file. - Move reference counters at least for mbuf clusters into an unused portion of the cluster itself, to save space and need to allocate a counter. - Look into introducing resource freeing possibly from a kproc. Reviewed by (in parts): jlemon, jake, silby, terry Tested by: jlemon (Intel & Alpha), mjacob (Intel & Alpha) Preliminary performance measurements: jlemon (and me, obviously) URL: http://people.freebsd.org/~bmilekic/mb_alloc/
2001-06-22 06:35:32 +00:00
mbstat.m_mpfail++; /* XXX: No consistency. */
return (NULL);
1994-05-24 10:09:53 +00:00
}
/*
* Partition an mbuf chain in two pieces, returning the tail --
* all but the first len0 bytes. In case of failure, it returns NULL and
* attempts to restore the chain to its original state.
*
* Note that the resulting mbufs might be read-only, because the new
* mbuf can end up sharing an mbuf cluster with the original mbuf if
* the "breaking point" happens to lie within a cluster mbuf. Use the
* M_WRITABLE() macro to check for this case.
1994-05-24 10:09:53 +00:00
*/
struct mbuf *
m_split(struct mbuf *m0, int len0, int wait)
1994-05-24 10:09:53 +00:00
{
struct mbuf *m, *n;
1994-05-24 10:09:53 +00:00
unsigned len = len0, remain;
for (m = m0; m && len > m->m_len; m = m->m_next)
len -= m->m_len;
if (m == NULL)
return (NULL);
1994-05-24 10:09:53 +00:00
remain = m->m_len - len;
if (m0->m_flags & M_PKTHDR) {
MGETHDR(n, wait, m0->m_type);
if (n == NULL)
return (NULL);
1994-05-24 10:09:53 +00:00
n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
n->m_pkthdr.len = m0->m_pkthdr.len - len0;
m0->m_pkthdr.len = len0;
if (m->m_flags & M_EXT)
goto extpacket;
if (remain > MHLEN) {
/* m can't be the lead packet */
MH_ALIGN(n, 0);
n->m_next = m_split(m, len, wait);
if (n->m_next == NULL) {
1994-05-24 10:09:53 +00:00
(void) m_free(n);
return (NULL);
} else {
n->m_len = 0;
1994-05-24 10:09:53 +00:00
return (n);
}
1994-05-24 10:09:53 +00:00
} else
MH_ALIGN(n, remain);
} else if (remain == 0) {
n = m->m_next;
m->m_next = NULL;
1994-05-24 10:09:53 +00:00
return (n);
} else {
MGET(n, wait, m->m_type);
if (n == NULL)
return (NULL);
1994-05-24 10:09:53 +00:00
M_ALIGN(n, remain);
}
extpacket:
if (m->m_flags & M_EXT) {
n->m_flags |= M_EXT;
n->m_ext = m->m_ext;
Replace the mbuf external reference counting code with something that should be better. The old code counted references to mbuf clusters by using the offset of the cluster from the start of memory allocated for mbufs and clusters as an index into an array of chars, which did the reference counting. If the external storage was not a cluster then reference counting had to be done by the code using that external storage. NetBSD's system of linked lists of mbufs was cosidered, but Alfred felt it would have locking issues when the kernel was made more SMP friendly. The system implimented uses a pool of unions to track external storage. The union contains an int for counting the references and a pointer for forming a free list. The reference counts are incremented and decremented atomically and so should be SMP friendly. This system can track reference counts for any sort of external storage. Access to the reference counting stuff is now through macros defined in mbuf.h, so it should be easier to make changes to the system in the future. The possibility of storing the reference count in one of the referencing mbufs was considered, but was rejected 'cos it would often leave extra mbufs allocated. Storing the reference count in the cluster was also considered, but because the external storage may not be a cluster this isn't an option. The size of the pool of reference counters is available in the stats provided by "netstat -m". PR: 19866 Submitted by: Bosko Milekic <bmilekic@dsuper.net> Reviewed by: alfred (glanced at by others on -net)
2000-08-19 08:32:59 +00:00
MEXT_ADD_REF(m);
1994-05-24 10:09:53 +00:00
n->m_data = m->m_data + len;
} else {
bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
}
n->m_len = remain;
m->m_len = len;
n->m_next = m->m_next;
m->m_next = NULL;
1994-05-24 10:09:53 +00:00
return (n);
}
/*
* Routine to copy from device local memory into mbufs.
* Note that `off' argument is offset into first mbuf of target chain from
* which to begin copying the data to.
1994-05-24 10:09:53 +00:00
*/
struct mbuf *
m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
void (*copy)(char *from, caddr_t to, u_int len))
1994-05-24 10:09:53 +00:00
{
struct mbuf *m;
1994-05-24 10:09:53 +00:00
struct mbuf *top = 0, **mp = &top;
int len;
if (off < 0 || off > MHLEN)
return (NULL);
1994-05-24 10:09:53 +00:00
MGETHDR(m, M_DONTWAIT, MT_DATA);
if (m == NULL)
return (NULL);
1994-05-24 10:09:53 +00:00
m->m_pkthdr.rcvif = ifp;
m->m_pkthdr.len = totlen;
len = MHLEN;
1994-05-24 10:09:53 +00:00
while (totlen > 0) {
if (top) {
MGET(m, M_DONTWAIT, MT_DATA);
if (m == NULL) {
1994-05-24 10:09:53 +00:00
m_freem(top);
return (NULL);
1994-05-24 10:09:53 +00:00
}
len = MLEN;
1994-05-24 10:09:53 +00:00
}
if (totlen + off >= MINCLSIZE) {
1994-05-24 10:09:53 +00:00
MCLGET(m, M_DONTWAIT);
if (m->m_flags & M_EXT)
len = MCLBYTES;
1994-05-24 10:09:53 +00:00
} else {
/*
* Place initial small packet/header at end of mbuf.
*/
if (top == NULL && totlen + off + max_linkhdr <= len) {
m->m_data += max_linkhdr;
len -= max_linkhdr;
}
}
if (off) {
m->m_data += off;
len -= off;
off = 0;
1994-05-24 10:09:53 +00:00
}
m->m_len = len = min(totlen, len);
1994-05-24 10:09:53 +00:00
if (copy)
copy(buf, mtod(m, caddr_t), (unsigned)len);
1994-05-24 10:09:53 +00:00
else
bcopy(buf, mtod(m, caddr_t), (unsigned)len);
buf += len;
1994-05-24 10:09:53 +00:00
*mp = m;
mp = &m->m_next;
totlen -= len;
}
return (top);
}
1994-10-04 06:50:01 +00:00
/*
* Copy data from a buffer back into the indicated mbuf chain,
* starting "off" bytes from the beginning, extending the mbuf
* chain if necessary.
*/
void
m_copyback(struct mbuf *m0, int off, int len, caddr_t cp)
1994-10-04 06:50:01 +00:00
{
int mlen;
struct mbuf *m = m0, *n;
1994-10-04 06:50:01 +00:00
int totlen = 0;
if (m0 == NULL)
1994-10-04 06:50:01 +00:00
return;
while (off > (mlen = m->m_len)) {
off -= mlen;
totlen += mlen;
if (m->m_next == NULL) {
Introduce numerous SMP friendly changes to the mbuf allocator. Namely, introduce a modified allocation mechanism for mbufs and mbuf clusters; one which can scale under SMP and which offers the possibility of resource reclamation to be implemented in the future. Notable advantages: o Reduce contention for SMP by offering per-CPU pools and locks. o Better use of data cache due to per-CPU pools. o Much less code cache pollution due to excessively large allocation macros. o Framework for `grouping' objects from same page together so as to be able to possibly free wired-down pages back to the system if they are no longer needed by the network stacks. Additional things changed with this addition: - Moved some mbuf specific declarations and initializations from sys/conf/param.c into mbuf-specific code where they belong. - m_getclr() has been renamed to m_get_clrd() because the old name is really confusing. m_getclr() HAS been preserved though and is defined to the new name. No tree sweep has been done "to change the interface," as the old name will continue to be supported and is not depracated. The change was merely done because m_getclr() sounds too much like "m_get a cluster." - TEMPORARILY disabled mbtypes statistics displaying in netstat(1) and systat(1) (see TODO below). - Fixed systat(1) to display number of "free mbufs" based on new per-CPU stat structures. - Fixed netstat(1) to display new per-CPU stats based on sysctl-exported per-CPU stat structures. All infos are fetched via sysctl. TODO (in order of priority): - Re-enable mbtypes statistics in both netstat(1) and systat(1) after introducing an SMP friendly way to collect the mbtypes stats under the already introduced per-CPU locks (i.e. hopefully don't use atomic() - it seems too costly for a mere stat update, especially when other locks are already present). - Optionally have systat(1) display not only "total free mbufs" but also "total free mbufs per CPU pool." - Fix minor length-fetching issues in netstat(1) related to recently re-enabled option to read mbuf stats from a core file. - Move reference counters at least for mbuf clusters into an unused portion of the cluster itself, to save space and need to allocate a counter. - Look into introducing resource freeing possibly from a kproc. Reviewed by (in parts): jlemon, jake, silby, terry Tested by: jlemon (Intel & Alpha), mjacob (Intel & Alpha) Preliminary performance measurements: jlemon (and me, obviously) URL: http://people.freebsd.org/~bmilekic/mb_alloc/
2001-06-22 06:35:32 +00:00
n = m_get_clrd(M_DONTWAIT, m->m_type);
if (n == NULL)
1994-10-04 06:50:01 +00:00
goto out;
n->m_len = min(MLEN, len + off);
m->m_next = n;
}
m = m->m_next;
}
while (len > 0) {
mlen = min (m->m_len - off, len);
bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen);
cp += mlen;
len -= mlen;
mlen += off;
off = 0;
totlen += mlen;
if (len == 0)
break;
if (m->m_next == NULL) {
1994-10-04 06:50:01 +00:00
n = m_get(M_DONTWAIT, m->m_type);
if (n == NULL)
1994-10-04 06:50:01 +00:00
break;
n->m_len = min(MLEN, len);
m->m_next = n;
}
m = m->m_next;
}
out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
m->m_pkthdr.len = totlen;
}
void
m_print(const struct mbuf *m)
{
int len;
const struct mbuf *m2;
len = m->m_pkthdr.len;
m2 = m;
while (len) {
printf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-");
len -= m2->m_len;
m2 = m2->m_next;
}
return;
}