freebsd-dev/sys/kern/uipc_mbuf.c

1838 lines
45 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.
* 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
*/
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_mac.h"
#include "opt_param.h"
#include "opt_mbuf_stress_test.h"
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/limits.h>
#include <sys/lock.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>
#include <sys/uio.h>
#include <security/mac/mac_framework.h>
int max_linkhdr;
int max_protohdr;
int max_hdr;
int max_datalen;
#ifdef MBUF_STRESS_TEST
int m_defragpackets;
int m_defragbytes;
int m_defraguseless;
int m_defragfailure;
int m_defragrandomfailures;
#endif
2000-09-30 06:30:39 +00:00
/*
* sysctl(8) exported objects
*/
SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD,
&max_linkhdr, 0, "Size of largest link layer header");
SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD,
&max_protohdr, 0, "Size of largest protocol layer header");
SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD,
&max_hdr, 0, "Size of largest link plus protocol header");
SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RD,
&max_datalen, 0, "Minimum space left in mbuf after max_hdr");
#ifdef MBUF_STRESS_TEST
SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
&m_defragpackets, 0, "");
SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
&m_defragbytes, 0, "");
SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
&m_defraguseless, 0, "");
SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
&m_defragfailure, 0, "");
SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
&m_defragrandomfailures, 0, "");
#endif
1994-05-24 10:09:53 +00:00
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
/*
* Allocate a given length worth of mbufs and/or clusters (whatever fits
* best) and return a pointer to the top of the allocated chain. If an
* existing mbuf chain is provided, then we will append the new chain
* to the existing one but still return the top of the newly allocated
* chain.
*/
struct mbuf *
m_getm2(struct mbuf *m, int len, int how, short type, int flags)
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
{
struct mbuf *mb, *nm = NULL, *mtail = NULL;
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
KASSERT(len >= 0, ("%s: len is < 0", __func__));
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
/* Validate flags. */
flags &= (M_PKTHDR | M_EOR);
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
/* Packet header mbuf must be first in chain. */
if ((flags & M_PKTHDR) && m != NULL)
flags &= ~M_PKTHDR;
/* Loop and append maximum sized mbufs to the chain tail. */
while (len > 0) {
if (len > MCLBYTES)
mb = m_getjcl(how, type, (flags & M_PKTHDR),
MJUMPAGESIZE);
else if (len >= MINCLSIZE)
mb = m_getcl(how, type, (flags & M_PKTHDR));
else if (flags & M_PKTHDR)
mb = m_gethdr(how, type);
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
else
mb = m_get(how, type);
/* Fail the whole operation if one mbuf can't be allocated. */
if (mb == NULL) {
if (nm != NULL)
m_freem(nm);
return (NULL);
}
/* Book keeping. */
len -= (mb->m_flags & M_EXT) ? mb->m_ext.ext_size :
((mb->m_flags & M_PKTHDR) ? MHLEN : MLEN);
if (mtail != NULL)
mtail->m_next = mb;
else
nm = mb;
mtail = mb;
flags &= ~M_PKTHDR; /* Only valid on the first mbuf. */
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
}
if (flags & M_EOR)
mtail->m_flags |= M_EOR; /* Only valid on the last mbuf. */
/* If mbuf was supplied, append new chain to the end of it. */
if (m != NULL) {
for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next)
;
mtail->m_next = nm;
mtail->m_flags &= ~M_EOR;
} else
m = nm;
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
return (m);
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
}
/*
* Free an entire chain of mbufs and associated external buffers, if
* applicable.
*/
void
m_freem(struct mbuf *mb)
{
while (mb != NULL)
mb = m_free(mb);
}
/*-
* Configure a provided mbuf to refer to the provided external storage
* buffer and setup a reference count for said buffer. If the setting
* up of the reference count fails, the M_EXT bit will not be set. If
* successfull, the M_EXT bit is set in the mbuf's flags.
*
* Arguments:
* mb The existing mbuf to which to attach the provided buffer.
* buf The address of the provided external storage buffer.
* size The size of the provided buffer.
* freef A pointer to a routine that is responsible for freeing the
* provided external storage buffer.
* args A pointer to an argument structure (of any type) to be passed
* to the provided freef routine (may be NULL).
* flags Any other flags to be passed to the provided mbuf.
* type The type that the external storage buffer should be
* labeled with.
*
* Returns:
* Nothing.
*/
void
m_extadd(struct mbuf *mb, caddr_t buf, u_int size,
void (*freef)(void *, void *), void *args, int flags, int type)
{
KASSERT(type != EXT_CLUSTER, ("%s: EXT_CLUSTER not allowed", __func__));
if (type != EXT_EXTREF)
mb->m_ext.ref_cnt = (u_int *)uma_zalloc(zone_ext_refcnt, M_NOWAIT);
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
if (mb->m_ext.ref_cnt != NULL) {
*(mb->m_ext.ref_cnt) = 1;
mb->m_flags |= (M_EXT | flags);
mb->m_ext.ext_buf = buf;
mb->m_data = mb->m_ext.ext_buf;
mb->m_ext.ext_size = size;
mb->m_ext.ext_free = freef;
mb->m_ext.ext_args = args;
mb->m_ext.ext_type = type;
}
}
/*
* Non-directly-exported function to clean up after mbufs with M_EXT
* storage attached to them if the reference count hits 1.
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
*/
void
mb_free_ext(struct mbuf *m)
{
KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
/* Free attached storage if this mbuf is the only reference to it. */
if (*(m->m_ext.ref_cnt) == 1 ||
atomic_fetchadd_int(m->m_ext.ref_cnt, -1) == 1) {
switch (m->m_ext.ext_type) {
case EXT_PACKET: /* The packet zone is special. */
if (*(m->m_ext.ref_cnt) == 0)
*(m->m_ext.ref_cnt) = 1;
uma_zfree(zone_pack, m);
return; /* Job done. */
case EXT_CLUSTER:
uma_zfree(zone_clust, m->m_ext.ext_buf);
break;
case EXT_JUMBOP:
uma_zfree(zone_jumbop, m->m_ext.ext_buf);
break;
case EXT_JUMBO9:
uma_zfree(zone_jumbo9, m->m_ext.ext_buf);
break;
case EXT_JUMBO16:
uma_zfree(zone_jumbo16, m->m_ext.ext_buf);
break;
case EXT_SFBUF:
case EXT_NET_DRV:
case EXT_MOD_TYPE:
case EXT_DISPOSABLE:
*(m->m_ext.ref_cnt) = 0;
uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *,
m->m_ext.ref_cnt));
/* FALLTHROUGH */
case EXT_EXTREF:
KASSERT(m->m_ext.ext_free != NULL,
("%s: ext_free not set", __func__));
(*(m->m_ext.ext_free))(m->m_ext.ext_buf,
m->m_ext.ext_args);
break;
default:
KASSERT(m->m_ext.ext_type == 0,
("%s: unknown ext_type", __func__));
}
}
/*
* Free this mbuf back to the mbuf zone with all m_ext
* information purged.
*/
m->m_ext.ext_buf = NULL;
m->m_ext.ext_free = NULL;
m->m_ext.ext_args = NULL;
m->m_ext.ref_cnt = NULL;
m->m_ext.ext_size = 0;
m->m_ext.ext_type = 0;
m->m_flags &= ~M_EXT;
uma_zfree(zone_mbuf, m);
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
}
/*
* Attach the the cluster from *m to *n, set up m_ext in *n
* and bump the refcount of the cluster.
*/
static void
mb_dupcl(struct mbuf *n, struct mbuf *m)
{
KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
KASSERT((n->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
if (*(m->m_ext.ref_cnt) == 1)
*(m->m_ext.ref_cnt) += 1;
else
atomic_add_int(m->m_ext.ref_cnt, 1);
n->m_ext.ext_buf = m->m_ext.ext_buf;
n->m_ext.ext_free = m->m_ext.ext_free;
n->m_ext.ext_args = m->m_ext.ext_args;
n->m_ext.ext_size = m->m_ext.ext_size;
n->m_ext.ref_cnt = m->m_ext.ref_cnt;
n->m_ext.ext_type = m->m_ext.ext_type;
n->m_flags |= M_EXT;
}
/*
* Clean up mbuf (chain) from any tags and packet headers.
* If "all" is set then the first mbuf in the chain will be
* cleaned too.
*/
void
m_demote(struct mbuf *m0, int all)
{
struct mbuf *m;
for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) {
if (m->m_flags & M_PKTHDR) {
m_tag_delete_chain(m, NULL);
m->m_flags &= ~M_PKTHDR;
bzero(&m->m_pkthdr, sizeof(struct pkthdr));
}
if (m->m_type == MT_HEADER)
m->m_type = MT_DATA;
if (m != m0 && m->m_nextpkt != NULL)
m->m_nextpkt = NULL;
m->m_flags = m->m_flags & (M_EXT|M_EOR|M_RDONLY|M_FREELIST);
}
}
/*
* Sanity checks on mbuf (chain) for use in KASSERT() and general
* debugging.
* Returns 0 or panics when bad and 1 on all tests passed.
* Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they
* blow up later.
*/
int
m_sanity(struct mbuf *m0, int sanitize)
{
struct mbuf *m;
caddr_t a, b;
int pktlen = 0;
#define M_SANITY_ACTION(s) return (0)
/* #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m) */
for (m = m0; m != NULL; m = m->m_next) {
/*
* Basic pointer checks. If any of these fails then some
* unrelated kernel memory before or after us is trashed.
* No way to recover from that.
*/
a = ((m->m_flags & M_EXT) ? m->m_ext.ext_buf :
((m->m_flags & M_PKTHDR) ? (caddr_t)(&m->m_pktdat) :
(caddr_t)(&m->m_dat)) );
b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size :
((m->m_flags & M_PKTHDR) ? MHLEN : MLEN)));
if ((caddr_t)m->m_data < a)
M_SANITY_ACTION("m_data outside mbuf data range left");
if ((caddr_t)m->m_data > b)
M_SANITY_ACTION("m_data outside mbuf data range right");
if ((caddr_t)m->m_data + m->m_len > b)
M_SANITY_ACTION("m_data + m_len exeeds mbuf space");
if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.header) {
if ((caddr_t)m->m_pkthdr.header < a ||
(caddr_t)m->m_pkthdr.header > b)
M_SANITY_ACTION("m_pkthdr.header outside mbuf data range");
}
/* m->m_nextpkt may only be set on first mbuf in chain. */
if (m != m0 && m->m_nextpkt != NULL) {
if (sanitize) {
m_freem(m->m_nextpkt);
m->m_nextpkt = (struct mbuf *)0xDEADC0DE;
} else
M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf");
}
/* correct type correlations. */
if (m->m_type == MT_HEADER && !(m->m_flags & M_PKTHDR)) {
if (sanitize)
m->m_type = MT_DATA;
else
M_SANITY_ACTION("MT_HEADER set but not M_PKTHDR");
}
/* packet length (not mbuf length!) calculation */
if (m0->m_flags & M_PKTHDR)
pktlen += m->m_len;
/* m_tags may only be attached to first mbuf in chain. */
if (m != m0 && m->m_flags & M_PKTHDR &&
!SLIST_EMPTY(&m->m_pkthdr.tags)) {
if (sanitize) {
m_tag_delete_chain(m, NULL);
/* put in 0xDEADC0DE perhaps? */
} else
M_SANITY_ACTION("m_tags on in-chain mbuf");
}
/* M_PKTHDR may only be set on first mbuf in chain */
if (m != m0 && m->m_flags & M_PKTHDR) {
if (sanitize) {
bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
m->m_flags &= ~M_PKTHDR;
/* put in 0xDEADCODE and leave hdr flag in */
} else
M_SANITY_ACTION("M_PKTHDR on in-chain mbuf");
}
}
m = m0;
if (pktlen && pktlen != m->m_pkthdr.len) {
if (sanitize)
m->m_pkthdr.len = 0;
else
M_SANITY_ACTION("m_pkthdr.len != mbuf chain length");
}
return 1;
#undef M_SANITY_ACTION
}
/*
* "Move" mbuf pkthdr from "from" to "to".
* "from" must have M_PKTHDR set, and "to" must be empty.
*/
void
m_move_pkthdr(struct mbuf *to, struct mbuf *from)
{
#if 0
/* see below for why these are not enabled */
M_ASSERTPKTHDR(to);
Move MAC label storage for mbufs into m_tags from the m_pkthdr structure, returning some additional room in the first mbuf in a chain, and avoiding feature-specific contents in the mbuf header. To do this: - Modify mbuf_to_label() to extract the tag, returning NULL if not found. - Introduce mac_init_mbuf_tag() which does most of the work mac_init_mbuf() used to do, except on an m_tag rather than an mbuf. - Scale back mac_init_mbuf() to perform m_tag allocation and invoke mac_init_mbuf_tag(). - Replace mac_destroy_mbuf() with mac_destroy_mbuf_tag(), since m_tag's are now GC'd deep in the m_tag/mbuf code rather than at a higher level when mbufs are directly free()'d. - Add mac_copy_mbuf_tag() to support m_copy_pkthdr() and related notions. - Generally change all references to mbuf labels so that they use mbuf_to_label() rather than &mbuf->m_pkthdr.label. This required no changes in the MAC policies (yay!). - Tweak mbuf release routines to not call mac_destroy_mbuf(), tag destruction takes care of it for us now. - Remove MAC magic from m_copy_pkthdr() and m_move_pkthdr() -- the existing m_tag support does all this for us. Note that we can no longer just zero the m_tag list on the target mbuf, rather, we have to delete the chain because m_tag's will already be hung off freshly allocated mbuf's. - Tweak m_tag copying routines so that if we're copying a MAC m_tag, we don't do a binary copy, rather, we initialize the new storage and do a deep copy of the label. - Remove use of MAC_FLAG_INITIALIZED in a few bizarre places having to do with mbuf header copies previously. - When an mbuf is copied in ip_input(), we no longer need to explicitly copy the label because it will get handled by the m_tag code now. - No longer any weird handling of MAC labels in if_loop.c during header copies. - Add MPC_LOADTIME_FLAG_LABELMBUFS flag to Biba, MLS, mac_test. In mac_test, handle the label==NULL case, since it can be dynamically loaded. In order to improve performance with this change, introduce the notion of "lazy MAC label allocation" -- only allocate m_tag storage for MAC labels if we're running with a policy that uses MAC labels on mbufs. Policies declare this intent by setting the MPC_LOADTIME_FLAG_LABELMBUFS flag in their load-time flags field during declaration. Note: this opens up the possibility of post-boot policy modules getting back NULL slot entries even though they have policy invariants of non-NULL slot entries, as the policy might have been loaded after the mbuf was allocated, leaving the mbuf without label storage. Policies that cannot handle this case must be declared as NOTLATE, or must be modified. - mac_labelmbufs holds the current cumulative status as to whether any policies require mbuf labeling or not. This is updated whenever the active policy set changes by the function mac_policy_updateflags(). The function iterates the list and checks whether any have the flag set. Write access to this variable is protected by the policy list; read access is currently not protected for performance reasons. This might change if it causes problems. - Add MAC_POLICY_LIST_ASSERT_EXCLUSIVE() to permit the flags update function to assert appropriate locks. - This makes allocation in mac_init_mbuf() conditional on the flag. Reviewed by: sam Obtained from: TrustedBSD Project Sponsored by: DARPA, Network Associates Laboratories
2003-04-14 20:39:06 +00:00
/* Note: with MAC, this may not be a good assertion. */
KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
("m_move_pkthdr: to has tags"));
#endif
#ifdef MAC
Move MAC label storage for mbufs into m_tags from the m_pkthdr structure, returning some additional room in the first mbuf in a chain, and avoiding feature-specific contents in the mbuf header. To do this: - Modify mbuf_to_label() to extract the tag, returning NULL if not found. - Introduce mac_init_mbuf_tag() which does most of the work mac_init_mbuf() used to do, except on an m_tag rather than an mbuf. - Scale back mac_init_mbuf() to perform m_tag allocation and invoke mac_init_mbuf_tag(). - Replace mac_destroy_mbuf() with mac_destroy_mbuf_tag(), since m_tag's are now GC'd deep in the m_tag/mbuf code rather than at a higher level when mbufs are directly free()'d. - Add mac_copy_mbuf_tag() to support m_copy_pkthdr() and related notions. - Generally change all references to mbuf labels so that they use mbuf_to_label() rather than &mbuf->m_pkthdr.label. This required no changes in the MAC policies (yay!). - Tweak mbuf release routines to not call mac_destroy_mbuf(), tag destruction takes care of it for us now. - Remove MAC magic from m_copy_pkthdr() and m_move_pkthdr() -- the existing m_tag support does all this for us. Note that we can no longer just zero the m_tag list on the target mbuf, rather, we have to delete the chain because m_tag's will already be hung off freshly allocated mbuf's. - Tweak m_tag copying routines so that if we're copying a MAC m_tag, we don't do a binary copy, rather, we initialize the new storage and do a deep copy of the label. - Remove use of MAC_FLAG_INITIALIZED in a few bizarre places having to do with mbuf header copies previously. - When an mbuf is copied in ip_input(), we no longer need to explicitly copy the label because it will get handled by the m_tag code now. - No longer any weird handling of MAC labels in if_loop.c during header copies. - Add MPC_LOADTIME_FLAG_LABELMBUFS flag to Biba, MLS, mac_test. In mac_test, handle the label==NULL case, since it can be dynamically loaded. In order to improve performance with this change, introduce the notion of "lazy MAC label allocation" -- only allocate m_tag storage for MAC labels if we're running with a policy that uses MAC labels on mbufs. Policies declare this intent by setting the MPC_LOADTIME_FLAG_LABELMBUFS flag in their load-time flags field during declaration. Note: this opens up the possibility of post-boot policy modules getting back NULL slot entries even though they have policy invariants of non-NULL slot entries, as the policy might have been loaded after the mbuf was allocated, leaving the mbuf without label storage. Policies that cannot handle this case must be declared as NOTLATE, or must be modified. - mac_labelmbufs holds the current cumulative status as to whether any policies require mbuf labeling or not. This is updated whenever the active policy set changes by the function mac_policy_updateflags(). The function iterates the list and checks whether any have the flag set. Write access to this variable is protected by the policy list; read access is currently not protected for performance reasons. This might change if it causes problems. - Add MAC_POLICY_LIST_ASSERT_EXCLUSIVE() to permit the flags update function to assert appropriate locks. - This makes allocation in mac_init_mbuf() conditional on the flag. Reviewed by: sam Obtained from: TrustedBSD Project Sponsored by: DARPA, Network Associates Laboratories
2003-04-14 20:39:06 +00:00
/*
* XXXMAC: It could be this should also occur for non-MAC?
*/
if (to->m_flags & M_PKTHDR)
Move MAC label storage for mbufs into m_tags from the m_pkthdr structure, returning some additional room in the first mbuf in a chain, and avoiding feature-specific contents in the mbuf header. To do this: - Modify mbuf_to_label() to extract the tag, returning NULL if not found. - Introduce mac_init_mbuf_tag() which does most of the work mac_init_mbuf() used to do, except on an m_tag rather than an mbuf. - Scale back mac_init_mbuf() to perform m_tag allocation and invoke mac_init_mbuf_tag(). - Replace mac_destroy_mbuf() with mac_destroy_mbuf_tag(), since m_tag's are now GC'd deep in the m_tag/mbuf code rather than at a higher level when mbufs are directly free()'d. - Add mac_copy_mbuf_tag() to support m_copy_pkthdr() and related notions. - Generally change all references to mbuf labels so that they use mbuf_to_label() rather than &mbuf->m_pkthdr.label. This required no changes in the MAC policies (yay!). - Tweak mbuf release routines to not call mac_destroy_mbuf(), tag destruction takes care of it for us now. - Remove MAC magic from m_copy_pkthdr() and m_move_pkthdr() -- the existing m_tag support does all this for us. Note that we can no longer just zero the m_tag list on the target mbuf, rather, we have to delete the chain because m_tag's will already be hung off freshly allocated mbuf's. - Tweak m_tag copying routines so that if we're copying a MAC m_tag, we don't do a binary copy, rather, we initialize the new storage and do a deep copy of the label. - Remove use of MAC_FLAG_INITIALIZED in a few bizarre places having to do with mbuf header copies previously. - When an mbuf is copied in ip_input(), we no longer need to explicitly copy the label because it will get handled by the m_tag code now. - No longer any weird handling of MAC labels in if_loop.c during header copies. - Add MPC_LOADTIME_FLAG_LABELMBUFS flag to Biba, MLS, mac_test. In mac_test, handle the label==NULL case, since it can be dynamically loaded. In order to improve performance with this change, introduce the notion of "lazy MAC label allocation" -- only allocate m_tag storage for MAC labels if we're running with a policy that uses MAC labels on mbufs. Policies declare this intent by setting the MPC_LOADTIME_FLAG_LABELMBUFS flag in their load-time flags field during declaration. Note: this opens up the possibility of post-boot policy modules getting back NULL slot entries even though they have policy invariants of non-NULL slot entries, as the policy might have been loaded after the mbuf was allocated, leaving the mbuf without label storage. Policies that cannot handle this case must be declared as NOTLATE, or must be modified. - mac_labelmbufs holds the current cumulative status as to whether any policies require mbuf labeling or not. This is updated whenever the active policy set changes by the function mac_policy_updateflags(). The function iterates the list and checks whether any have the flag set. Write access to this variable is protected by the policy list; read access is currently not protected for performance reasons. This might change if it causes problems. - Add MAC_POLICY_LIST_ASSERT_EXCLUSIVE() to permit the flags update function to assert appropriate locks. - This makes allocation in mac_init_mbuf() conditional on the flag. Reviewed by: sam Obtained from: TrustedBSD Project Sponsored by: DARPA, Network Associates Laboratories
2003-04-14 20:39:06 +00:00
m_tag_delete_chain(to, NULL);
#endif
to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
if ((to->m_flags & M_EXT) == 0)
to->m_data = to->m_pktdat;
to->m_pkthdr = from->m_pkthdr; /* especially tags */
SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */
from->m_flags &= ~M_PKTHDR;
}
/*
* Duplicate "from"'s mbuf pkthdr in "to".
* "from" must have M_PKTHDR set, and "to" must be empty.
* In particular, this does a deep copy of the packet tags.
*/
int
m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how)
{
#if 0
/*
* The mbuf allocator only initializes the pkthdr
* when the mbuf is allocated with MGETHDR. Many users
* (e.g. m_copy*, m_prepend) use MGET and then
* smash the pkthdr as needed causing these
* assertions to trip. For now just disable them.
*/
M_ASSERTPKTHDR(to);
Move MAC label storage for mbufs into m_tags from the m_pkthdr structure, returning some additional room in the first mbuf in a chain, and avoiding feature-specific contents in the mbuf header. To do this: - Modify mbuf_to_label() to extract the tag, returning NULL if not found. - Introduce mac_init_mbuf_tag() which does most of the work mac_init_mbuf() used to do, except on an m_tag rather than an mbuf. - Scale back mac_init_mbuf() to perform m_tag allocation and invoke mac_init_mbuf_tag(). - Replace mac_destroy_mbuf() with mac_destroy_mbuf_tag(), since m_tag's are now GC'd deep in the m_tag/mbuf code rather than at a higher level when mbufs are directly free()'d. - Add mac_copy_mbuf_tag() to support m_copy_pkthdr() and related notions. - Generally change all references to mbuf labels so that they use mbuf_to_label() rather than &mbuf->m_pkthdr.label. This required no changes in the MAC policies (yay!). - Tweak mbuf release routines to not call mac_destroy_mbuf(), tag destruction takes care of it for us now. - Remove MAC magic from m_copy_pkthdr() and m_move_pkthdr() -- the existing m_tag support does all this for us. Note that we can no longer just zero the m_tag list on the target mbuf, rather, we have to delete the chain because m_tag's will already be hung off freshly allocated mbuf's. - Tweak m_tag copying routines so that if we're copying a MAC m_tag, we don't do a binary copy, rather, we initialize the new storage and do a deep copy of the label. - Remove use of MAC_FLAG_INITIALIZED in a few bizarre places having to do with mbuf header copies previously. - When an mbuf is copied in ip_input(), we no longer need to explicitly copy the label because it will get handled by the m_tag code now. - No longer any weird handling of MAC labels in if_loop.c during header copies. - Add MPC_LOADTIME_FLAG_LABELMBUFS flag to Biba, MLS, mac_test. In mac_test, handle the label==NULL case, since it can be dynamically loaded. In order to improve performance with this change, introduce the notion of "lazy MAC label allocation" -- only allocate m_tag storage for MAC labels if we're running with a policy that uses MAC labels on mbufs. Policies declare this intent by setting the MPC_LOADTIME_FLAG_LABELMBUFS flag in their load-time flags field during declaration. Note: this opens up the possibility of post-boot policy modules getting back NULL slot entries even though they have policy invariants of non-NULL slot entries, as the policy might have been loaded after the mbuf was allocated, leaving the mbuf without label storage. Policies that cannot handle this case must be declared as NOTLATE, or must be modified. - mac_labelmbufs holds the current cumulative status as to whether any policies require mbuf labeling or not. This is updated whenever the active policy set changes by the function mac_policy_updateflags(). The function iterates the list and checks whether any have the flag set. Write access to this variable is protected by the policy list; read access is currently not protected for performance reasons. This might change if it causes problems. - Add MAC_POLICY_LIST_ASSERT_EXCLUSIVE() to permit the flags update function to assert appropriate locks. - This makes allocation in mac_init_mbuf() conditional on the flag. Reviewed by: sam Obtained from: TrustedBSD Project Sponsored by: DARPA, Network Associates Laboratories
2003-04-14 20:39:06 +00:00
/* Note: with MAC, this may not be a good assertion. */
KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
#endif
MBUF_CHECKSLEEP(how);
#ifdef MAC
if (to->m_flags & M_PKTHDR)
Move MAC label storage for mbufs into m_tags from the m_pkthdr structure, returning some additional room in the first mbuf in a chain, and avoiding feature-specific contents in the mbuf header. To do this: - Modify mbuf_to_label() to extract the tag, returning NULL if not found. - Introduce mac_init_mbuf_tag() which does most of the work mac_init_mbuf() used to do, except on an m_tag rather than an mbuf. - Scale back mac_init_mbuf() to perform m_tag allocation and invoke mac_init_mbuf_tag(). - Replace mac_destroy_mbuf() with mac_destroy_mbuf_tag(), since m_tag's are now GC'd deep in the m_tag/mbuf code rather than at a higher level when mbufs are directly free()'d. - Add mac_copy_mbuf_tag() to support m_copy_pkthdr() and related notions. - Generally change all references to mbuf labels so that they use mbuf_to_label() rather than &mbuf->m_pkthdr.label. This required no changes in the MAC policies (yay!). - Tweak mbuf release routines to not call mac_destroy_mbuf(), tag destruction takes care of it for us now. - Remove MAC magic from m_copy_pkthdr() and m_move_pkthdr() -- the existing m_tag support does all this for us. Note that we can no longer just zero the m_tag list on the target mbuf, rather, we have to delete the chain because m_tag's will already be hung off freshly allocated mbuf's. - Tweak m_tag copying routines so that if we're copying a MAC m_tag, we don't do a binary copy, rather, we initialize the new storage and do a deep copy of the label. - Remove use of MAC_FLAG_INITIALIZED in a few bizarre places having to do with mbuf header copies previously. - When an mbuf is copied in ip_input(), we no longer need to explicitly copy the label because it will get handled by the m_tag code now. - No longer any weird handling of MAC labels in if_loop.c during header copies. - Add MPC_LOADTIME_FLAG_LABELMBUFS flag to Biba, MLS, mac_test. In mac_test, handle the label==NULL case, since it can be dynamically loaded. In order to improve performance with this change, introduce the notion of "lazy MAC label allocation" -- only allocate m_tag storage for MAC labels if we're running with a policy that uses MAC labels on mbufs. Policies declare this intent by setting the MPC_LOADTIME_FLAG_LABELMBUFS flag in their load-time flags field during declaration. Note: this opens up the possibility of post-boot policy modules getting back NULL slot entries even though they have policy invariants of non-NULL slot entries, as the policy might have been loaded after the mbuf was allocated, leaving the mbuf without label storage. Policies that cannot handle this case must be declared as NOTLATE, or must be modified. - mac_labelmbufs holds the current cumulative status as to whether any policies require mbuf labeling or not. This is updated whenever the active policy set changes by the function mac_policy_updateflags(). The function iterates the list and checks whether any have the flag set. Write access to this variable is protected by the policy list; read access is currently not protected for performance reasons. This might change if it causes problems. - Add MAC_POLICY_LIST_ASSERT_EXCLUSIVE() to permit the flags update function to assert appropriate locks. - This makes allocation in mac_init_mbuf() conditional on the flag. Reviewed by: sam Obtained from: TrustedBSD Project Sponsored by: DARPA, Network Associates Laboratories
2003-04-14 20:39:06 +00:00
m_tag_delete_chain(to, NULL);
#endif
to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
if ((to->m_flags & M_EXT) == 0)
to->m_data = to->m_pktdat;
to->m_pkthdr = from->m_pkthdr;
SLIST_INIT(&to->m_pkthdr.tags);
return (m_tag_copy_chain(to, from, MBTOM(how)));
}
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;
if (m->m_flags & M_PKTHDR)
MGETHDR(mn, how, m->m_type);
else
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
}
Move MAC label storage for mbufs into m_tags from the m_pkthdr structure, returning some additional room in the first mbuf in a chain, and avoiding feature-specific contents in the mbuf header. To do this: - Modify mbuf_to_label() to extract the tag, returning NULL if not found. - Introduce mac_init_mbuf_tag() which does most of the work mac_init_mbuf() used to do, except on an m_tag rather than an mbuf. - Scale back mac_init_mbuf() to perform m_tag allocation and invoke mac_init_mbuf_tag(). - Replace mac_destroy_mbuf() with mac_destroy_mbuf_tag(), since m_tag's are now GC'd deep in the m_tag/mbuf code rather than at a higher level when mbufs are directly free()'d. - Add mac_copy_mbuf_tag() to support m_copy_pkthdr() and related notions. - Generally change all references to mbuf labels so that they use mbuf_to_label() rather than &mbuf->m_pkthdr.label. This required no changes in the MAC policies (yay!). - Tweak mbuf release routines to not call mac_destroy_mbuf(), tag destruction takes care of it for us now. - Remove MAC magic from m_copy_pkthdr() and m_move_pkthdr() -- the existing m_tag support does all this for us. Note that we can no longer just zero the m_tag list on the target mbuf, rather, we have to delete the chain because m_tag's will already be hung off freshly allocated mbuf's. - Tweak m_tag copying routines so that if we're copying a MAC m_tag, we don't do a binary copy, rather, we initialize the new storage and do a deep copy of the label. - Remove use of MAC_FLAG_INITIALIZED in a few bizarre places having to do with mbuf header copies previously. - When an mbuf is copied in ip_input(), we no longer need to explicitly copy the label because it will get handled by the m_tag code now. - No longer any weird handling of MAC labels in if_loop.c during header copies. - Add MPC_LOADTIME_FLAG_LABELMBUFS flag to Biba, MLS, mac_test. In mac_test, handle the label==NULL case, since it can be dynamically loaded. In order to improve performance with this change, introduce the notion of "lazy MAC label allocation" -- only allocate m_tag storage for MAC labels if we're running with a policy that uses MAC labels on mbufs. Policies declare this intent by setting the MPC_LOADTIME_FLAG_LABELMBUFS flag in their load-time flags field during declaration. Note: this opens up the possibility of post-boot policy modules getting back NULL slot entries even though they have policy invariants of non-NULL slot entries, as the policy might have been loaded after the mbuf was allocated, leaving the mbuf without label storage. Policies that cannot handle this case must be declared as NOTLATE, or must be modified. - mac_labelmbufs holds the current cumulative status as to whether any policies require mbuf labeling or not. This is updated whenever the active policy set changes by the function mac_policy_updateflags(). The function iterates the list and checks whether any have the flag set. Write access to this variable is protected by the policy list; read access is currently not protected for performance reasons. This might change if it causes problems. - Add MAC_POLICY_LIST_ASSERT_EXCLUSIVE() to permit the flags update function to assert appropriate locks. - This makes allocation in mac_init_mbuf() conditional on the flag. Reviewed by: sam Obtained from: TrustedBSD Project Sponsored by: DARPA, Network Associates Laboratories
2003-04-14 20:39:06 +00:00
if (m->m_flags & M_PKTHDR)
M_MOVE_PKTHDR(mn, m);
1994-05-24 10:09:53 +00:00
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));
MBUF_CHECKSLEEP(wait);
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;
}
if (copyhdr)
MGETHDR(n, wait, m->m_type);
else
MGET(n, wait, m->m_type);
1994-05-24 10:09:53 +00:00
*np = n;
if (n == NULL)
1994-05-24 10:09:53 +00:00
goto nospace;
if (copyhdr) {
if (!m_dup_pkthdr(n, m, wait))
goto nospace;
1994-05-24 10:09:53 +00:00
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;
mb_dupcl(n, m);
1994-05-24 10:09:53 +00:00
} else
bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
(u_int)n->m_len);
1994-05-24 10:09:53 +00:00
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
}
/*
* Returns mbuf chain with new head for the prepending case.
* Copies from mbuf (chain) n from off for len to mbuf (chain) m
* either prepending or appending the data.
* The resulting mbuf (chain) m is fully writeable.
* m is destination (is made writeable)
* n is source, off is offset in source, len is len from offset
* dir, 0 append, 1 prepend
* how, wait or nowait
*/
static int
m_bcopyxxx(void *s, void *t, u_int len)
{
bcopy(s, t, (size_t)len);
return 0;
}
struct mbuf *
m_copymdata(struct mbuf *m, struct mbuf *n, int off, int len,
int prep, int how)
{
struct mbuf *mm, *x, *z, *prev = NULL;
caddr_t p;
int i, nlen = 0;
caddr_t buf[MLEN];
KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source"));
KASSERT(off >= 0, ("m_copymdata, negative off %d", off));
KASSERT(len >= 0, ("m_copymdata, negative len %d", len));
KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep));
mm = m;
if (!prep) {
while(mm->m_next) {
prev = mm;
mm = mm->m_next;
}
}
for (z = n; z != NULL; z = z->m_next)
nlen += z->m_len;
if (len == M_COPYALL)
len = nlen - off;
if (off + len > nlen || len < 1)
return NULL;
if (!M_WRITABLE(mm)) {
/* XXX: Use proper m_xxx function instead. */
x = m_getcl(how, MT_DATA, mm->m_flags);
if (x == NULL)
return NULL;
bcopy(mm->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size);
p = x->m_ext.ext_buf + (mm->m_data - mm->m_ext.ext_buf);
x->m_data = p;
mm->m_next = NULL;
if (mm != m)
prev->m_next = x;
m_free(mm);
mm = x;
}
/*
* Append/prepend the data. Allocating mbufs as necessary.
*/
/* Shortcut if enough free space in first/last mbuf. */
if (!prep && M_TRAILINGSPACE(mm) >= len) {
m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) +
mm->m_len);
mm->m_len += len;
mm->m_pkthdr.len += len;
return m;
}
if (prep && M_LEADINGSPACE(mm) >= len) {
mm->m_data = mtod(mm, caddr_t) - len;
m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t));
mm->m_len += len;
mm->m_pkthdr.len += len;
return mm;
}
/* Expand first/last mbuf to cluster if possible. */
if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) {
bcopy(mm->m_data, &buf, mm->m_len);
m_clget(mm, how);
if (!(mm->m_flags & M_EXT))
return NULL;
bcopy(&buf, mm->m_ext.ext_buf, mm->m_len);
mm->m_data = mm->m_ext.ext_buf;
mm->m_pkthdr.header = NULL;
}
if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) {
bcopy(mm->m_data, &buf, mm->m_len);
m_clget(mm, how);
if (!(mm->m_flags & M_EXT))
return NULL;
bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf +
mm->m_ext.ext_size - mm->m_len, mm->m_len);
mm->m_data = (caddr_t)mm->m_ext.ext_buf +
mm->m_ext.ext_size - mm->m_len;
mm->m_pkthdr.header = NULL;
}
/* Append/prepend as many mbuf (clusters) as necessary to fit len. */
if (!prep && len > M_TRAILINGSPACE(mm)) {
if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA))
return NULL;
}
if (prep && len > M_LEADINGSPACE(mm)) {
if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA)))
return NULL;
i = 0;
for (x = z; x != NULL; x = x->m_next) {
i += x->m_flags & M_EXT ? x->m_ext.ext_size :
(x->m_flags & M_PKTHDR ? MHLEN : MLEN);
if (!x->m_next)
break;
}
z->m_data += i - len;
m_move_pkthdr(mm, z);
x->m_next = mm;
mm = z;
}
/* Seek to start position in source mbuf. Optimization for long chains. */
while (off > 0) {
if (off < n->m_len)
break;
off -= n->m_len;
n = n->m_next;
}
/* Copy data into target mbuf. */
z = mm;
while (len > 0) {
KASSERT(z != NULL, ("m_copymdata, falling off target edge"));
i = M_TRAILINGSPACE(z);
m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len);
z->m_len += i;
/* fixup pkthdr.len if necessary */
if ((prep ? mm : m)->m_flags & M_PKTHDR)
(prep ? mm : m)->m_pkthdr.len += i;
off += i;
len -= i;
z = z->m_next;
}
return (prep ? mm : m);
}
/*
* 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;
MBUF_CHECKSLEEP(how);
MGET(n, how, m->m_type);
top = n;
if (n == NULL)
goto nospace;
if (!m_dup_pkthdr(n, m, how))
goto nospace;
n->m_len = m->m_len;
if (m->m_flags & M_EXT) {
n->m_data = m->m_data;
mb_dupcl(n, 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;
mb_dupcl(n, 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
{
u_int 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;
MBUF_CHECKSLEEP(how);
/* Sanity check */
if (m == NULL)
return (NULL);
M_ASSERTPKTHDR(m);
/* 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 */
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
if (remain >= MINCLSIZE) {
n = m_getcl(how, m->m_type, 0);
nsize = MCLBYTES;
} else {
n = m_get(how, m->m_type);
nsize = MLEN;
}
if (n == NULL)
goto nospace;
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
if (top == NULL) { /* First one, must be PKTHDR */
if (!m_dup_pkthdr(n, m, how)) {
m_free(n);
goto nospace;
}
if ((n->m_flags & M_EXT) == 0)
nsize = MHLEN;
}
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;
if (m->m_next != NULL) {
m_freem(m->m_next);
m->m_next = NULL;
}
1994-05-24 10:09:53 +00:00
break;
}
count -= m->m_len;
}
}
}
/*
* 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_MOVE_PKTHDR(m, n);
1994-05-24 10:09:53 +00:00
}
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,
(u_int)count);
1994-05-24 10:09:53 +00:00
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
}
/*
* Like m_pullup(), except a new mbuf is always allocated, and we allow
* the amount of empty space before the data in the new mbuf to be specified
* (in the event that the caller expects to prepend later).
*/
int MSFail;
struct mbuf *
m_copyup(struct mbuf *n, int len, int dstoff)
{
struct mbuf *m;
int count, space;
if (len > (MHLEN - dstoff))
goto bad;
MGET(m, M_DONTWAIT, n->m_type);
if (m == NULL)
goto bad;
m->m_len = 0;
if (n->m_flags & M_PKTHDR)
M_MOVE_PKTHDR(m, n);
m->m_data += dstoff;
space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
do {
count = min(min(max(len, max_protohdr), space), n->m_len);
memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
(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);
MSFail++;
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;
u_int len = len0, remain;
1994-05-24 10:09:53 +00:00
MBUF_CHECKSLEEP(wait);
1994-05-24 10:09:53 +00:00
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_data = m->m_data + len;
mb_dupcl(n, m);
1994-05-24 10:09:53 +00:00
} 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;
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
struct mbuf *top = NULL, **mp = &top;
int len;
if (off < 0 || off > MHLEN)
return (NULL);
1994-05-24 10:09:53 +00:00
while (totlen > 0) {
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
if (top == NULL) { /* First one, must be PKTHDR */
if (totlen + off >= MINCLSIZE) {
m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
len = MCLBYTES;
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
} else {
m = m_gethdr(M_DONTWAIT, MT_DATA);
len = MHLEN;
/* Place initial small packet/header at end of mbuf */
if (m && totlen + off + max_linkhdr <= MLEN) {
m->m_data += max_linkhdr;
len -= max_linkhdr;
}
}
if (m == NULL)
return NULL;
m->m_pkthdr.rcvif = ifp;
m->m_pkthdr.len = totlen;
1994-05-24 10:09:53 +00:00
} else {
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
if (totlen + off >= MINCLSIZE) {
m = m_getcl(M_DONTWAIT, MT_DATA, 0);
len = MCLBYTES;
} else {
m = m_get(M_DONTWAIT, MT_DATA);
len = MLEN;
}
if (m == NULL) {
m_freem(top);
return NULL;
}
}
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), (u_int)len);
1994-05-24 10:09:53 +00:00
else
bcopy(buf, mtod(m, caddr_t), (u_int)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, c_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) {
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
n = m_get(M_DONTWAIT, m->m_type);
if (n == NULL)
1994-10-04 06:50:01 +00:00
goto out;
Bring in mbuma to replace mballoc. mbuma is an Mbuf & Cluster allocator built on top of a number of extensions to the UMA framework, all included herein. Extensions to UMA worth noting: - Better layering between slab <-> zone caches; introduce Keg structure which splits off slab cache away from the zone structure and allows multiple zones to be stacked on top of a single Keg (single type of slab cache); perhaps we should look into defining a subset API on top of the Keg for special use by malloc(9), for example. - UMA_ZONE_REFCNT zones can now be added, and reference counters automagically allocated for them within the end of the associated slab structures. uma_find_refcnt() does a kextract to fetch the slab struct reference from the underlying page, and lookup the corresponding refcnt. mbuma things worth noting: - integrates mbuf & cluster allocations with extended UMA and provides caches for commonly-allocated items; defines several zones (two primary, one secondary) and two kegs. - change up certain code paths that always used to do: m_get() + m_clget() to instead just use m_getcl() and try to take advantage of the newly defined secondary Packet zone. - netstat(1) and systat(1) quickly hacked up to do basic stat reporting but additional stats work needs to be done once some other details within UMA have been taken care of and it becomes clearer to how stats will work within the modified framework. From the user perspective, one implication is that the NMBCLUSTERS compile-time option is no longer used. The maximum number of clusters is still capped off according to maxusers, but it can be made unlimited by setting the kern.ipc.nmbclusters boot-time tunable to zero. Work should be done to write an appropriate sysctl handler allowing dynamic tuning of kern.ipc.nmbclusters at runtime. Additional things worth noting/known issues (READ): - One report of 'ips' (ServeRAID) driver acting really slow in conjunction with mbuma. Need more data. Latest report is that ips is equally sucking with and without mbuma. - Giant leak in NFS code sometimes occurs, can't reproduce but currently analyzing; brueffer is able to reproduce but THIS IS NOT an mbuma-specific problem and currently occurs even WITHOUT mbuma. - Issues in network locking: there is at least one code path in the rip code where one or more locks are acquired and we end up in m_prepend() with M_WAITOK, which causes WITNESS to whine from within UMA. Current temporary solution: force all UMA allocations to be M_NOWAIT from within UMA for now to avoid deadlocks unless WITNESS is defined and we can determine with certainty that we're not holding any locks when we're M_WAITOK. - I've seen at least one weird socketbuffer empty-but- mbuf-still-attached panic. I don't believe this to be related to mbuma but please keep your eyes open, turn on debugging, and capture crash dumps. This change removes more code than it adds. A paper is available detailing the change and considering various performance issues, it was presented at BSDCan2004: http://www.unixdaemons.com/~bmilekic/netbuf_bmilekic.pdf Please read the paper for Future Work and implementation details, as well as credits. Testing and Debugging: rwatson, brueffer, Ketrien I. Saihr-Kesenchedra, ... Reviewed by: Lots of people (for different parts)
2004-05-31 21:46:06 +00:00
bzero(mtod(n, caddr_t), MLEN);
1994-10-04 06:50:01 +00:00
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), (u_int)mlen);
1994-10-04 06:50:01 +00:00
cp += mlen;
len -= mlen;
mlen += off;
off = 0;
totlen += mlen;
if (len == 0)
break;
if (m->m_next == NULL) {
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;
}
/*
* Append the specified data to the indicated mbuf chain,
* Extend the mbuf chain if the new data does not fit in
* existing space.
*
* Return 1 if able to complete the job; otherwise 0.
*/
int
m_append(struct mbuf *m0, int len, c_caddr_t cp)
{
struct mbuf *m, *n;
int remainder, space;
for (m = m0; m->m_next != NULL; m = m->m_next)
;
remainder = len;
space = M_TRAILINGSPACE(m);
if (space > 0) {
/*
* Copy into available space.
*/
if (space > remainder)
space = remainder;
bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
m->m_len += space;
cp += space, remainder -= space;
}
while (remainder > 0) {
/*
* Allocate a new mbuf; could check space
* and allocate a cluster instead.
*/
n = m_get(M_DONTWAIT, m->m_type);
if (n == NULL)
break;
n->m_len = min(MLEN, remainder);
bcopy(cp, mtod(n, caddr_t), n->m_len);
cp += n->m_len, remainder -= n->m_len;
m->m_next = n;
m = n;
}
if (m0->m_flags & M_PKTHDR)
m0->m_pkthdr.len += len - remainder;
return (remainder == 0);
}
/*
* Apply function f to the data in an mbuf chain starting "off" bytes from
* the beginning, continuing for "len" bytes.
*/
int
m_apply(struct mbuf *m, int off, int len,
int (*f)(void *, void *, u_int), void *arg)
{
u_int count;
int rval;
KASSERT(off >= 0, ("m_apply, negative off %d", off));
KASSERT(len >= 0, ("m_apply, negative len %d", len));
while (off > 0) {
KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
if (off < m->m_len)
break;
off -= m->m_len;
m = m->m_next;
}
while (len > 0) {
KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
count = min(m->m_len - off, len);
rval = (*f)(arg, mtod(m, caddr_t) + off, count);
if (rval)
return (rval);
len -= count;
off = 0;
m = m->m_next;
}
return (0);
}
/*
* Return a pointer to mbuf/offset of location in mbuf chain.
*/
struct mbuf *
m_getptr(struct mbuf *m, int loc, int *off)
{
while (loc >= 0) {
/* Normal end of search. */
if (m->m_len > loc) {
*off = loc;
return (m);
} else {
loc -= m->m_len;
if (m->m_next == NULL) {
if (loc == 0) {
/* Point at the end of valid data. */
*off = m->m_len;
return (m);
}
return (NULL);
}
m = m->m_next;
}
}
return (NULL);
}
void
m_print(const struct mbuf *m, int maxlen)
{
int len;
int pdata;
const struct mbuf *m2;
if (m->m_flags & M_PKTHDR)
len = m->m_pkthdr.len;
else
len = -1;
m2 = m;
while (m2 != NULL && (len == -1 || len)) {
pdata = m2->m_len;
if (maxlen != -1 && pdata > maxlen)
pdata = maxlen;
printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
m2->m_next, m2->m_flags, "\20\20freelist\17skipfw"
"\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
"\3eor\2pkthdr\1ext", pdata ? "" : "\n");
if (pdata)
printf(", %*D\n", pdata, (u_char *)m2->m_data, "-");
if (len != -1)
len -= m2->m_len;
m2 = m2->m_next;
}
if (len > 0)
printf("%d bytes unaccounted for.\n", len);
return;
}
u_int
m_fixhdr(struct mbuf *m0)
{
u_int len;
len = m_length(m0, NULL);
m0->m_pkthdr.len = len;
return (len);
}
u_int
m_length(struct mbuf *m0, struct mbuf **last)
{
struct mbuf *m;
u_int len;
len = 0;
for (m = m0; m != NULL; m = m->m_next) {
len += m->m_len;
if (m->m_next == NULL)
break;
}
if (last != NULL)
*last = m;
return (len);
}
/*
* Defragment a mbuf chain, returning the shortest possible
* chain of mbufs and clusters. If allocation fails and
* this cannot be completed, NULL will be returned, but
* the passed in chain will be unchanged. Upon success,
* the original chain will be freed, and the new chain
* will be returned.
*
* If a non-packet header is passed in, the original
* mbuf (chain?) will be returned unharmed.
*/
struct mbuf *
m_defrag(struct mbuf *m0, int how)
{
struct mbuf *m_new = NULL, *m_final = NULL;
int progress = 0, length;
MBUF_CHECKSLEEP(how);
if (!(m0->m_flags & M_PKTHDR))
return (m0);
m_fixhdr(m0); /* Needed sanity check */
#ifdef MBUF_STRESS_TEST
if (m_defragrandomfailures) {
int temp = arc4random() & 0xff;
if (temp == 0xba)
goto nospace;
}
#endif
if (m0->m_pkthdr.len > MHLEN)
m_final = m_getcl(how, MT_DATA, M_PKTHDR);
else
m_final = m_gethdr(how, MT_DATA);
if (m_final == NULL)
goto nospace;
if (m_dup_pkthdr(m_final, m0, how) == 0)
goto nospace;
m_new = m_final;
while (progress < m0->m_pkthdr.len) {
length = m0->m_pkthdr.len - progress;
if (length > MCLBYTES)
length = MCLBYTES;
if (m_new == NULL) {
if (length > MLEN)
m_new = m_getcl(how, MT_DATA, 0);
else
m_new = m_get(how, MT_DATA);
if (m_new == NULL)
goto nospace;
}
m_copydata(m0, progress, length, mtod(m_new, caddr_t));
progress += length;
m_new->m_len = length;
if (m_new != m_final)
m_cat(m_final, m_new);
m_new = NULL;
}
#ifdef MBUF_STRESS_TEST
if (m0->m_next == NULL)
m_defraguseless++;
#endif
m_freem(m0);
m0 = m_final;
#ifdef MBUF_STRESS_TEST
m_defragpackets++;
m_defragbytes += m0->m_pkthdr.len;
#endif
return (m0);
nospace:
#ifdef MBUF_STRESS_TEST
m_defragfailure++;
#endif
if (m_final)
m_freem(m_final);
return (NULL);
}
#ifdef MBUF_STRESS_TEST
/*
* Fragment an mbuf chain. There's no reason you'd ever want to do
* this in normal usage, but it's great for stress testing various
* mbuf consumers.
*
* If fragmentation is not possible, the original chain will be
* returned.
*
* Possible length values:
* 0 no fragmentation will occur
* > 0 each fragment will be of the specified length
* -1 each fragment will be the same random value in length
* -2 each fragment's length will be entirely random
* (Random values range from 1 to 256)
*/
struct mbuf *
m_fragment(struct mbuf *m0, int how, int length)
{
struct mbuf *m_new = NULL, *m_final = NULL;
int progress = 0;
if (!(m0->m_flags & M_PKTHDR))
return (m0);
if ((length == 0) || (length < -2))
return (m0);
m_fixhdr(m0); /* Needed sanity check */
m_final = m_getcl(how, MT_DATA, M_PKTHDR);
if (m_final == NULL)
goto nospace;
2003-12-25 01:17:27 +00:00
if (m_dup_pkthdr(m_final, m0, how) == 0)
goto nospace;
m_new = m_final;
if (length == -1)
length = 1 + (arc4random() & 255);
while (progress < m0->m_pkthdr.len) {
int fraglen;
if (length > 0)
fraglen = length;
else
fraglen = 1 + (arc4random() & 255);
if (fraglen > m0->m_pkthdr.len - progress)
fraglen = m0->m_pkthdr.len - progress;
if (fraglen > MCLBYTES)
fraglen = MCLBYTES;
if (m_new == NULL) {
m_new = m_getcl(how, MT_DATA, 0);
if (m_new == NULL)
goto nospace;
}
m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t));
progress += fraglen;
m_new->m_len = fraglen;
if (m_new != m_final)
m_cat(m_final, m_new);
m_new = NULL;
}
m_freem(m0);
m0 = m_final;
return (m0);
nospace:
if (m_final)
m_freem(m_final);
/* Return the original chain on failure */
return (m0);
}
#endif
/*
* Copy the contents of uio into a properly sized mbuf chain.
*/
struct mbuf *
m_uiotombuf(struct uio *uio, int how, int len, int align, int flags)
{
struct mbuf *m, *mb;
int error, length, total;
int progress = 0;
/*
* len can be zero or an arbitrary large value bound by
* the total data supplied by the uio.
*/
if (len > 0)
total = min(uio->uio_resid, len);
else
total = uio->uio_resid;
/*
* The smallest unit returned by m_getm2() is a single mbuf
* with pkthdr. We can't align past it. Align align itself.
*/
if (align)
align &= ~(sizeof(long) - 1);
if (align >= MHLEN)
return (NULL);
/* Give us all or nothing. */
m = m_getm2(NULL, total + align, how, MT_DATA, flags);
if (m == NULL)
return (NULL);
m->m_data += align;
/* Fill all mbufs with uio data and update header information. */
for (mb = m; mb != NULL; mb = mb->m_next) {
length = min(M_TRAILINGSPACE(mb), total - progress);
error = uiomove(mtod(mb, void *), length, uio);
if (error) {
m_freem(m);
return (NULL);
}
mb->m_len = length;
progress += length;
if (flags & M_PKTHDR)
m->m_pkthdr.len += length;
}
KASSERT(progress == total, ("%s: progress != total", __func__));
return (m);
}
/*
* Set the m_data pointer of a newly-allocated mbuf
* to place an object of the specified size at the
* end of the mbuf, longword aligned.
*/
void
m_align(struct mbuf *m, int len)
{
int adjust;
if (m->m_flags & M_EXT)
adjust = m->m_ext.ext_size - len;
else if (m->m_flags & M_PKTHDR)
adjust = MHLEN - len;
else
adjust = MLEN - len;
m->m_data += adjust &~ (sizeof(long)-1);
}
/*
* Create a writable copy of the mbuf chain. While doing this
* we compact the chain with a goal of producing a chain with
* at most two mbufs. The second mbuf in this chain is likely
* to be a cluster. The primary purpose of this work is to create
* a writable packet for encryption, compression, etc. The
* secondary goal is to linearize the data so the data can be
* passed to crypto hardware in the most efficient manner possible.
*/
struct mbuf *
m_unshare(struct mbuf *m0, int how)
{
struct mbuf *m, *mprev;
struct mbuf *n, *mfirst, *mlast;
int len, off;
mprev = NULL;
for (m = m0; m != NULL; m = mprev->m_next) {
/*
* Regular mbufs are ignored unless there's a cluster
* in front of it that we can use to coalesce. We do
* the latter mainly so later clusters can be coalesced
* also w/o having to handle them specially (i.e. convert
* mbuf+cluster -> cluster). This optimization is heavily
* influenced by the assumption that we're running over
* Ethernet where MCLBYTES is large enough that the max
* packet size will permit lots of coalescing into a
* single cluster. This in turn permits efficient
* crypto operations, especially when using hardware.
*/
if ((m->m_flags & M_EXT) == 0) {
if (mprev && (mprev->m_flags & M_EXT) &&
m->m_len <= M_TRAILINGSPACE(mprev)) {
/* XXX: this ignores mbuf types */
memcpy(mtod(mprev, caddr_t) + mprev->m_len,
mtod(m, caddr_t), m->m_len);
mprev->m_len += m->m_len;
mprev->m_next = m->m_next; /* unlink from chain */
m_free(m); /* reclaim mbuf */
#if 0
newipsecstat.ips_mbcoalesced++;
#endif
} else {
mprev = m;
}
continue;
}
/*
* Writable mbufs are left alone (for now).
*/
if (M_WRITABLE(m)) {
mprev = m;
continue;
}
/*
* Not writable, replace with a copy or coalesce with
* the previous mbuf if possible (since we have to copy
* it anyway, we try to reduce the number of mbufs and
* clusters so that future work is easier).
*/
KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags));
/* NB: we only coalesce into a cluster or larger */
if (mprev != NULL && (mprev->m_flags & M_EXT) &&
m->m_len <= M_TRAILINGSPACE(mprev)) {
/* XXX: this ignores mbuf types */
memcpy(mtod(mprev, caddr_t) + mprev->m_len,
mtod(m, caddr_t), m->m_len);
mprev->m_len += m->m_len;
mprev->m_next = m->m_next; /* unlink from chain */
m_free(m); /* reclaim mbuf */
#if 0
newipsecstat.ips_clcoalesced++;
#endif
continue;
}
/*
* Allocate new space to hold the copy...
*/
/* XXX why can M_PKTHDR be set past the first mbuf? */
if (mprev == NULL && (m->m_flags & M_PKTHDR)) {
/*
* NB: if a packet header is present we must
* allocate the mbuf separately from any cluster
* because M_MOVE_PKTHDR will smash the data
* pointer and drop the M_EXT marker.
*/
MGETHDR(n, how, m->m_type);
if (n == NULL) {
m_freem(m0);
return (NULL);
}
M_MOVE_PKTHDR(n, m);
MCLGET(n, how);
if ((n->m_flags & M_EXT) == 0) {
m_free(n);
m_freem(m0);
return (NULL);
}
} else {
n = m_getcl(how, m->m_type, m->m_flags);
if (n == NULL) {
m_freem(m0);
return (NULL);
}
}
/*
* ... and copy the data. We deal with jumbo mbufs
* (i.e. m_len > MCLBYTES) by splitting them into
* clusters. We could just malloc a buffer and make
* it external but too many device drivers don't know
* how to break up the non-contiguous memory when
* doing DMA.
*/
len = m->m_len;
off = 0;
mfirst = n;
mlast = NULL;
for (;;) {
int cc = min(len, MCLBYTES);
memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc);
n->m_len = cc;
if (mlast != NULL)
mlast->m_next = n;
mlast = n;
#if 0
newipsecstat.ips_clcopied++;
#endif
len -= cc;
if (len <= 0)
break;
off += cc;
n = m_getcl(how, m->m_type, m->m_flags);
if (n == NULL) {
m_freem(mfirst);
m_freem(m0);
return (NULL);
}
}
n->m_next = m->m_next;
if (mprev == NULL)
m0 = mfirst; /* new head of chain */
else
mprev->m_next = mfirst; /* replace old mbuf */
m_free(m); /* release old mbuf */
mprev = mfirst;
}
return (m0);
}