1994-05-24 10:09:53 +00:00
|
|
|
/*-
|
|
|
|
* Copyright (c) 1993
|
|
|
|
* The Regents of the University of California. All rights reserved.
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
* Modifications/enhancements:
|
|
|
|
* Copyright (c) 1995 John S. Dyson. All rights reserved.
|
1994-05-24 10:09:53 +00:00
|
|
|
*
|
|
|
|
* 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.
|
|
|
|
*
|
|
|
|
* @(#)vfs_cluster.c 8.7 (Berkeley) 2/13/94
|
|
|
|
*/
|
|
|
|
|
2003-06-11 00:56:59 +00:00
|
|
|
#include <sys/cdefs.h>
|
|
|
|
__FBSDID("$FreeBSD$");
|
|
|
|
|
1998-01-31 07:23:16 +00:00
|
|
|
#include "opt_debug_cluster.h"
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
#include <sys/param.h>
|
1994-05-25 09:21:21 +00:00
|
|
|
#include <sys/systm.h>
|
1998-11-15 14:11:06 +00:00
|
|
|
#include <sys/kernel.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
#include <sys/proc.h>
|
2000-05-05 09:59:14 +00:00
|
|
|
#include <sys/bio.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
#include <sys/buf.h>
|
|
|
|
#include <sys/vnode.h>
|
1998-11-13 01:01:44 +00:00
|
|
|
#include <sys/malloc.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
#include <sys/mount.h>
|
|
|
|
#include <sys/resourcevar.h>
|
Implement a low-memory deadlock solution.
Removed most of the hacks that were trying to deal with low-memory
situations prior to now.
The new code is based on the concept that I/O must be able to function in
a low memory situation. All major modules related to I/O (except
networking) have been adjusted to allow allocation out of the system
reserve memory pool. These modules now detect a low memory situation but
rather then block they instead continue to operate, then return resources
to the memory pool instead of cache them or leave them wired.
Code has been added to stall in a low-memory situation prior to a vnode
being locked.
Thus situations where a process blocks in a low-memory condition while
holding a locked vnode have been reduced to near nothing. Not only will
I/O continue to operate, but many prior deadlock conditions simply no
longer exist.
Implement a number of VFS/BIO fixes
(found by Ian): in biodone(), bogus-page replacement code, the loop
was not properly incrementing loop variables prior to a continue
statement. We do not believe this code can be hit anyway but we
aren't taking any chances. We'll turn the whole section into a
panic (as it already is in brelse()) after the release is rolled.
In biodone(), the foff calculation was incorrectly
clamped to the iosize, causing the wrong foff to be calculated
for pages in the case of an I/O error or biodone() called without
initiating I/O. The problem always caused a panic before. Now it
doesn't. The problem is mainly an issue with NFS.
Fixed casts for ~PAGE_MASK. This code worked properly before only
because the calculations use signed arithmatic. Better to properly
extend PAGE_MASK first before inverting it for the 64 bit masking
op.
In brelse(), the bogus_page fixup code was improperly throwing
away the original contents of 'm' when it did the j-loop to
fix the bogus pages. The result was that it would potentially
invalidate parts of the *WRONG* page(!), leading to corruption.
There may still be cases where a background bitmap write is
being duplicated, causing potential corruption. We have identified
a potentially serious bug related to this but the fix is still TBD.
So instead this patch contains a KASSERT to detect the problem
and panic the machine rather then continue to corrupt the filesystem.
The problem does not occur very often.. it is very hard to
reproduce, and it may or may not be the cause of the corruption
people have reported.
Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>)
Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
|
|
|
#include <sys/vmmeter.h>
|
1995-02-22 09:39:22 +00:00
|
|
|
#include <vm/vm.h>
|
1995-09-03 19:56:15 +00:00
|
|
|
#include <vm/vm_object.h>
|
|
|
|
#include <vm/vm_page.h>
|
1999-07-04 00:31:17 +00:00
|
|
|
#include <sys/sysctl.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
|
1996-12-29 02:45:28 +00:00
|
|
|
#if defined(CLUSTERDEBUG)
|
|
|
|
static int rcluster= 0;
|
2002-03-05 15:38:49 +00:00
|
|
|
SYSCTL_INT(_debug, OID_AUTO, rcluster, CTLFLAG_RW, &rcluster, 0,
|
|
|
|
"Debug VFS clustering code");
|
1996-12-29 02:45:28 +00:00
|
|
|
#endif
|
|
|
|
|
1998-11-13 01:01:44 +00:00
|
|
|
static MALLOC_DEFINE(M_SEGMENT, "cluster_save buffer", "cluster_save buffer");
|
|
|
|
|
1995-12-22 16:06:46 +00:00
|
|
|
static struct cluster_save *
|
2002-03-19 21:25:46 +00:00
|
|
|
cluster_collectbufs(struct vnode *vp, struct buf *last_bp);
|
1995-12-22 16:06:46 +00:00
|
|
|
static struct buf *
|
2002-03-19 21:25:46 +00:00
|
|
|
cluster_rbuild(struct vnode *vp, u_quad_t filesize, daddr_t lbn,
|
2002-05-14 11:09:43 +00:00
|
|
|
daddr_t blkno, long size, int run, struct buf *fbp);
|
1994-05-24 10:09:53 +00:00
|
|
|
|
1999-07-04 00:31:17 +00:00
|
|
|
static int write_behind = 1;
|
2002-03-05 15:38:49 +00:00
|
|
|
SYSCTL_INT(_vfs, OID_AUTO, write_behind, CTLFLAG_RW, &write_behind, 0,
|
|
|
|
"Cluster write-behind; 0: disable, 1: enable, 2: backed off");
|
1999-07-04 00:31:17 +00:00
|
|
|
|
2003-03-13 06:17:59 +00:00
|
|
|
static int read_max = 8;
|
2003-03-11 06:14:03 +00:00
|
|
|
SYSCTL_INT(_vfs, OID_AUTO, read_max, CTLFLAG_RW, &read_max, 0,
|
|
|
|
"Cluster read-ahead max block count");
|
|
|
|
|
2002-03-05 15:38:49 +00:00
|
|
|
/* Page expended to mark partially backed buffers */
|
1995-12-22 16:06:46 +00:00
|
|
|
extern vm_page_t bogus_page;
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
2002-03-05 15:38:49 +00:00
|
|
|
/*
|
|
|
|
* Number of physical bufs (pbufs) this subsystem is allowed.
|
|
|
|
* Manipulated by vm_pager.c
|
|
|
|
*/
|
1999-01-21 08:29:12 +00:00
|
|
|
extern int cluster_pbuf_freecnt;
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
2002-03-05 15:38:49 +00:00
|
|
|
* Read data to a buf, including read-ahead if we find this to be beneficial.
|
|
|
|
* cluster_read replaces bread.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1994-05-25 09:21:21 +00:00
|
|
|
int
|
1996-12-29 02:45:28 +00:00
|
|
|
cluster_read(vp, filesize, lblkno, size, cred, totread, seqcount, bpp)
|
1994-05-24 10:09:53 +00:00
|
|
|
struct vnode *vp;
|
|
|
|
u_quad_t filesize;
|
|
|
|
daddr_t lblkno;
|
|
|
|
long size;
|
|
|
|
struct ucred *cred;
|
1996-12-29 02:45:28 +00:00
|
|
|
long totread;
|
|
|
|
int seqcount;
|
1994-05-24 10:09:53 +00:00
|
|
|
struct buf **bpp;
|
|
|
|
{
|
1996-12-29 02:45:28 +00:00
|
|
|
struct buf *bp, *rbp, *reqbp;
|
2002-05-14 11:09:43 +00:00
|
|
|
daddr_t blkno, origblkno;
|
1996-12-29 02:45:28 +00:00
|
|
|
int maxra, racluster;
|
2003-03-11 06:14:03 +00:00
|
|
|
int error, ncontig;
|
|
|
|
int i;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
|
|
|
error = 0;
|
1996-12-29 02:45:28 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Try to limit the amount of read-ahead by a few
|
|
|
|
* ad-hoc parameters. This needs work!!!
|
|
|
|
*/
|
1999-09-29 20:05:33 +00:00
|
|
|
racluster = vp->v_mount->mnt_iosize_max / size;
|
2003-03-11 06:14:03 +00:00
|
|
|
maxra = seqcount;
|
|
|
|
maxra = min(read_max, maxra);
|
|
|
|
maxra = min(nbuf/8, maxra);
|
|
|
|
if (((u_quad_t)(lblkno + maxra + 1) * size) > filesize)
|
|
|
|
maxra = (filesize / size) - lblkno;
|
1996-12-29 02:45:28 +00:00
|
|
|
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
/*
|
|
|
|
* get the requested block
|
|
|
|
*/
|
2003-03-04 00:04:44 +00:00
|
|
|
*bpp = reqbp = bp = getblk(vp, lblkno, size, 0, 0, 0);
|
1996-12-29 02:45:28 +00:00
|
|
|
origblkno = lblkno;
|
1995-12-11 04:58:34 +00:00
|
|
|
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
/*
|
|
|
|
* if it is in the cache, then check to see if the reads have been
|
|
|
|
* sequential. If they have, then try some read-ahead, otherwise
|
|
|
|
* back-off on prospective read-aheads.
|
|
|
|
*/
|
1994-05-24 10:09:53 +00:00
|
|
|
if (bp->b_flags & B_CACHE) {
|
1996-12-29 02:45:28 +00:00
|
|
|
if (!seqcount) {
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
return 0;
|
1996-12-29 02:45:28 +00:00
|
|
|
} else if ((bp->b_flags & B_RAM) == 0) {
|
|
|
|
return 0;
|
|
|
|
} else {
|
|
|
|
int s;
|
|
|
|
bp->b_flags &= ~B_RAM;
|
|
|
|
/*
|
|
|
|
* We do the spl here so that there is no window
|
|
|
|
* between the incore and the b_usecount increment
|
|
|
|
* below. We opt to keep the spl out of the loop
|
|
|
|
* for efficiency.
|
|
|
|
*/
|
|
|
|
s = splbio();
|
2002-09-25 02:12:32 +00:00
|
|
|
VI_LOCK(vp);
|
1999-06-26 02:47:16 +00:00
|
|
|
for (i = 1; i < maxra; i++) {
|
2002-07-10 17:02:32 +00:00
|
|
|
/*
|
|
|
|
* Stop if the buffer does not exist or it
|
|
|
|
* is invalid (about to go away?)
|
|
|
|
*/
|
2004-10-22 08:47:20 +00:00
|
|
|
rbp = gbincore(&vp->v_bufobj, lblkno+i);
|
2003-03-11 06:14:03 +00:00
|
|
|
if (rbp == NULL || (rbp->b_flags & B_INVAL))
|
1996-12-29 02:45:28 +00:00
|
|
|
break;
|
|
|
|
|
|
|
|
/*
|
1999-07-08 06:06:00 +00:00
|
|
|
* Set another read-ahead mark so we know
|
|
|
|
* to check again.
|
1996-12-29 02:45:28 +00:00
|
|
|
*/
|
|
|
|
if (((i % racluster) == (racluster - 1)) ||
|
|
|
|
(i == (maxra - 1)))
|
2003-03-11 06:14:03 +00:00
|
|
|
rbp->b_flags |= B_RAM;
|
1996-12-29 02:45:28 +00:00
|
|
|
}
|
2002-09-25 02:12:32 +00:00
|
|
|
VI_UNLOCK(vp);
|
1996-12-29 02:45:28 +00:00
|
|
|
splx(s);
|
|
|
|
if (i >= maxra) {
|
1995-01-24 10:00:46 +00:00
|
|
|
return 0;
|
1995-09-03 19:56:15 +00:00
|
|
|
}
|
1996-12-29 02:45:28 +00:00
|
|
|
lblkno += i;
|
|
|
|
}
|
|
|
|
reqbp = bp = NULL;
|
2003-03-04 21:35:28 +00:00
|
|
|
/*
|
|
|
|
* If it isn't in the cache, then get a chunk from
|
|
|
|
* disk if sequential, otherwise just get the block.
|
|
|
|
*/
|
1996-12-29 02:45:28 +00:00
|
|
|
} else {
|
1999-01-10 01:58:29 +00:00
|
|
|
off_t firstread = bp->b_offset;
|
2003-03-04 21:35:28 +00:00
|
|
|
int nblks;
|
1999-01-10 01:58:29 +00:00
|
|
|
|
1999-01-08 17:31:30 +00:00
|
|
|
KASSERT(bp->b_offset != NOOFFSET,
|
1999-01-10 01:58:29 +00:00
|
|
|
("cluster_read: no buffer offset"));
|
2003-03-04 21:35:28 +00:00
|
|
|
|
2003-03-11 06:14:03 +00:00
|
|
|
ncontig = 0;
|
2003-03-04 21:35:28 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Compute the total number of blocks that we should read
|
|
|
|
* synchronously.
|
|
|
|
*/
|
1996-12-29 02:45:28 +00:00
|
|
|
if (firstread + totread > filesize)
|
|
|
|
totread = filesize - firstread;
|
2003-03-04 21:35:28 +00:00
|
|
|
nblks = howmany(totread, size);
|
|
|
|
if (nblks > racluster)
|
|
|
|
nblks = racluster;
|
1996-12-29 02:45:28 +00:00
|
|
|
|
2003-03-04 21:35:28 +00:00
|
|
|
/*
|
|
|
|
* Now compute the number of contiguous blocks.
|
|
|
|
*/
|
|
|
|
if (nblks > 1) {
|
1996-12-29 02:45:28 +00:00
|
|
|
error = VOP_BMAP(vp, lblkno, NULL,
|
2003-03-11 06:14:03 +00:00
|
|
|
&blkno, &ncontig, NULL);
|
2003-03-04 21:35:28 +00:00
|
|
|
/*
|
|
|
|
* If this failed to map just do the original block.
|
|
|
|
*/
|
|
|
|
if (error || blkno == -1)
|
2003-03-11 06:14:03 +00:00
|
|
|
ncontig = 0;
|
2003-03-04 21:35:28 +00:00
|
|
|
}
|
1996-12-29 02:45:28 +00:00
|
|
|
|
2003-03-04 21:35:28 +00:00
|
|
|
/*
|
|
|
|
* If we have contiguous data available do a cluster
|
|
|
|
* otherwise just read the requested block.
|
|
|
|
*/
|
2003-03-11 06:14:03 +00:00
|
|
|
if (ncontig) {
|
2003-03-04 21:35:28 +00:00
|
|
|
/* Account for our first block. */
|
2003-03-11 06:14:03 +00:00
|
|
|
ncontig = min(ncontig + 1, nblks);
|
|
|
|
if (ncontig < nblks)
|
|
|
|
nblks = ncontig;
|
1996-12-29 02:45:28 +00:00
|
|
|
bp = cluster_rbuild(vp, filesize, lblkno,
|
|
|
|
blkno, size, nblks, bp);
|
In kern_physio.c fix tsleep priority messup.
In vfs_bio.c, remove b_generation count usage,
remove redundant reassignbuf,
remove redundant spl(s),
manage page PG_ZERO flags more correctly,
utilize in invalid value for b_offset until it
is properly initialized. Add asserts
for #ifdef DIAGNOSTIC, when b_offset is
improperly used.
when a process is not performing I/O, and just waiting
on a buffer generally, make the sleep priority
low.
only check page validity in getblk for B_VMIO buffers.
In vfs_cluster, add b_offset asserts, correct pointer calculation
for clustered reads. Improve readability of certain parts of
the code. Remove redundant spl(s).
In vfs_subr, correct usage of vfs_bio_awrite (From Andrew Gallatin
<gallatin@cs.duke.edu>). More vtruncbuf problems fixed.
1998-03-19 22:48:16 +00:00
|
|
|
lblkno += (bp->b_bufsize / size);
|
1995-09-03 19:56:15 +00:00
|
|
|
} else {
|
2000-03-20 10:44:49 +00:00
|
|
|
bp->b_flags |= B_RAM;
|
|
|
|
bp->b_iocmd = BIO_READ;
|
1995-09-03 19:56:15 +00:00
|
|
|
lblkno += 1;
|
1995-05-30 08:16:23 +00:00
|
|
|
}
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
|
|
|
/*
|
2003-03-11 06:14:03 +00:00
|
|
|
* handle the synchronous read so that it is available ASAP.
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
*/
|
|
|
|
if (bp) {
|
2000-12-26 19:41:38 +00:00
|
|
|
if ((bp->b_flags & B_CLUSTER) == 0) {
|
1998-05-21 07:47:58 +00:00
|
|
|
vfs_busy_pages(bp, 0);
|
2000-12-26 19:41:38 +00:00
|
|
|
}
|
2000-04-02 15:24:56 +00:00
|
|
|
bp->b_flags &= ~B_INVAL;
|
|
|
|
bp->b_ioflags &= ~BIO_ERROR;
|
2000-03-20 10:44:49 +00:00
|
|
|
if ((bp->b_flags & B_ASYNC) || bp->b_iodone != NULL)
|
1999-06-29 05:59:47 +00:00
|
|
|
BUF_KERNPROC(bp);
|
2003-10-18 14:10:28 +00:00
|
|
|
bp->b_iooffset = dbtob(bp->b_blkno);
|
2003-07-22 10:36:36 +00:00
|
|
|
error = VOP_STRATEGY(vp, bp);
|
1998-05-21 07:47:58 +00:00
|
|
|
curproc->p_stats->p_ru.ru_inblock++;
|
2003-03-11 06:14:03 +00:00
|
|
|
if (error)
|
|
|
|
return (error);
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
}
|
Some VM improvements, including elimination of alot of Sig-11
problems. Tor Egge and others have helped with various VM bugs
lately, but don't blame him -- blame me!!!
pmap.c:
1) Create an object for kernel page table allocations. This
fixes a bogus allocation method previously used for such, by
grabbing pages from the kernel object, using bogus pindexes.
(This was a code cleanup, and perhaps a minor system stability
issue.)
pmap.c:
2) Pre-set the modify and accessed bits when prudent. This will
decrease bus traffic under certain circumstances.
vfs_bio.c, vfs_cluster.c:
3) Rather than calculating the beginning virtual byte offset
multiple times, stick the offset into the buffer header, so
that the calculated offset can be reused. (Long long multiplies
are often expensive, and this is a probably unmeasurable performance
improvement, and code cleanup.)
vfs_bio.c:
4) Handle write recursion more intelligently (but not perfectly) so
that it is less likely to cause a system panic, and is also
much more robust.
vfs_bio.c:
5) getblk incorrectly wrote out blocks that are incorrectly sized.
The problem is fixed, and writes blocks out ONLY when B_DELWRI
is true.
vfs_bio.c:
6) Check that already constituted buffers have fully valid pages. If
not, then make sure that the B_CACHE bit is not set. (This was
a major source of Sig-11 type problems.)
vfs_bio.c:
7) Fix a potential system deadlock due to an incorrectly specified
sleep priority while waiting for a buffer write operation. The
change that I made opens the system up to serious problems, and
we need to examine the issue of process sleep priorities.
vfs_cluster.c, vfs_bio.c:
8) Make clustered reads work more correctly (and more completely)
when buffers are already constituted, but not fully valid.
(This was another system reliability issue.)
vfs_subr.c, ffs_inode.c:
9) Create a vtruncbuf function, which is used by filesystems that
can truncate files. The vinvalbuf forced a file sync type operation,
while vtruncbuf only invalidates the buffers past the new end of file,
and also invalidates the appropriate pages. (This was a system reliabiliy
and performance issue.)
10) Modify FFS to use vtruncbuf.
vm_object.c:
11) Make the object rundown mechanism for OBJT_VNODE type objects work
more correctly. Included in that fix, create pager entries for
the OBJT_DEAD pager type, so that paging requests that might slip
in during race conditions are properly handled. (This was a system
reliability issue.)
vm_page.c:
12) Make some of the page validation routines be a little less picky
about arguments passed to them. Also, support page invalidation
change the object generation count so that we handle generation
counts a little more robustly.
vm_pageout.c:
13) Further reduce pageout daemon activity when the system doesn't
need help from it. There should be no additional performance
decrease even when the pageout daemon is running. (This was
a significant performance issue.)
vnode_pager.c:
14) Teach the vnode pager to handle race conditions during vnode
deallocations.
1998-03-16 01:56:03 +00:00
|
|
|
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
/*
|
2003-03-11 06:14:03 +00:00
|
|
|
* If we have been doing sequential I/O, then do some read-ahead.
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
*/
|
2003-03-11 06:14:03 +00:00
|
|
|
while (lblkno < (origblkno + maxra)) {
|
|
|
|
error = VOP_BMAP(vp, lblkno, NULL, &blkno, &ncontig, NULL);
|
|
|
|
if (error)
|
|
|
|
break;
|
|
|
|
|
|
|
|
if (blkno == -1)
|
|
|
|
break;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We could throttle ncontig here by maxra but we might as
|
|
|
|
* well read the data if it is contiguous. We're throttled
|
|
|
|
* by racluster anyway.
|
|
|
|
*/
|
|
|
|
if (ncontig) {
|
|
|
|
ncontig = min(ncontig + 1, racluster);
|
|
|
|
rbp = cluster_rbuild(vp, filesize, lblkno, blkno,
|
|
|
|
size, ncontig, NULL);
|
|
|
|
lblkno += (rbp->b_bufsize / size);
|
2003-03-30 02:57:32 +00:00
|
|
|
if (rbp->b_flags & B_DELWRI) {
|
|
|
|
bqrelse(rbp);
|
|
|
|
continue;
|
|
|
|
}
|
2003-03-11 06:14:03 +00:00
|
|
|
} else {
|
|
|
|
rbp = getblk(vp, lblkno, size, 0, 0, 0);
|
2003-03-30 02:57:32 +00:00
|
|
|
lblkno += 1;
|
|
|
|
if (rbp->b_flags & B_DELWRI) {
|
|
|
|
bqrelse(rbp);
|
|
|
|
continue;
|
|
|
|
}
|
2003-03-11 06:14:03 +00:00
|
|
|
rbp->b_flags |= B_ASYNC | B_RAM;
|
|
|
|
rbp->b_iocmd = BIO_READ;
|
|
|
|
rbp->b_blkno = blkno;
|
|
|
|
}
|
|
|
|
if (rbp->b_flags & B_CACHE) {
|
2000-03-20 10:44:49 +00:00
|
|
|
rbp->b_flags &= ~B_ASYNC;
|
1996-01-19 04:00:31 +00:00
|
|
|
bqrelse(rbp);
|
2003-03-11 06:14:03 +00:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if ((rbp->b_flags & B_CLUSTER) == 0) {
|
|
|
|
vfs_busy_pages(rbp, 0);
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
}
|
2003-03-11 06:14:03 +00:00
|
|
|
rbp->b_flags &= ~B_INVAL;
|
|
|
|
rbp->b_ioflags &= ~BIO_ERROR;
|
|
|
|
if ((rbp->b_flags & B_ASYNC) || rbp->b_iodone != NULL)
|
|
|
|
BUF_KERNPROC(rbp);
|
2003-10-18 14:10:28 +00:00
|
|
|
rbp->b_iooffset = dbtob(rbp->b_blkno);
|
2003-07-22 10:36:36 +00:00
|
|
|
(void) VOP_STRATEGY(vp, rbp);
|
2003-03-11 06:14:03 +00:00
|
|
|
curproc->p_stats->p_ru.ru_inblock++;
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
}
|
2003-03-11 06:14:03 +00:00
|
|
|
|
1996-12-29 02:45:28 +00:00
|
|
|
if (reqbp)
|
2000-04-29 16:25:22 +00:00
|
|
|
return (bufwait(reqbp));
|
1996-12-29 02:45:28 +00:00
|
|
|
else
|
|
|
|
return (error);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If blocks are contiguous on disk, use this to provide clustered
|
|
|
|
* read ahead. We will read as many blocks as possible sequentially
|
|
|
|
* and then parcel them up into logical blocks in the buffer hash table.
|
|
|
|
*/
|
1995-09-03 19:56:15 +00:00
|
|
|
static struct buf *
|
1996-12-29 02:45:28 +00:00
|
|
|
cluster_rbuild(vp, filesize, lbn, blkno, size, run, fbp)
|
1994-05-24 10:09:53 +00:00
|
|
|
struct vnode *vp;
|
|
|
|
u_quad_t filesize;
|
|
|
|
daddr_t lbn;
|
2002-05-14 11:09:43 +00:00
|
|
|
daddr_t blkno;
|
1994-05-24 10:09:53 +00:00
|
|
|
long size;
|
|
|
|
int run;
|
1996-12-29 02:45:28 +00:00
|
|
|
struct buf *fbp;
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
1995-09-03 19:56:15 +00:00
|
|
|
struct buf *bp, *tbp;
|
1994-05-24 10:09:53 +00:00
|
|
|
daddr_t bn;
|
1998-10-25 17:44:59 +00:00
|
|
|
int i, inc, j;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2001-07-04 16:20:28 +00:00
|
|
|
GIANT_REQUIRED;
|
|
|
|
|
1999-01-08 17:31:30 +00:00
|
|
|
KASSERT(size == vp->v_mount->mnt_stat.f_iosize,
|
2003-11-12 08:01:40 +00:00
|
|
|
("cluster_rbuild: size %ld != filesize %jd\n",
|
|
|
|
size, (intmax_t)vp->v_mount->mnt_stat.f_iosize));
|
1999-01-10 01:58:29 +00:00
|
|
|
|
1995-12-11 04:58:34 +00:00
|
|
|
/*
|
|
|
|
* avoid a division
|
|
|
|
*/
|
|
|
|
while ((u_quad_t) size * (lbn + run) > filesize) {
|
1994-05-24 10:09:53 +00:00
|
|
|
--run;
|
1995-12-11 04:58:34 +00:00
|
|
|
}
|
1995-09-03 19:56:15 +00:00
|
|
|
|
1996-12-29 02:45:28 +00:00
|
|
|
if (fbp) {
|
|
|
|
tbp = fbp;
|
2000-03-20 10:44:49 +00:00
|
|
|
tbp->b_iocmd = BIO_READ;
|
1996-12-29 02:45:28 +00:00
|
|
|
} else {
|
2003-03-04 00:04:44 +00:00
|
|
|
tbp = getblk(vp, lbn, size, 0, 0, 0);
|
1996-12-29 02:45:28 +00:00
|
|
|
if (tbp->b_flags & B_CACHE)
|
|
|
|
return tbp;
|
2000-03-20 10:44:49 +00:00
|
|
|
tbp->b_flags |= B_ASYNC | B_RAM;
|
|
|
|
tbp->b_iocmd = BIO_READ;
|
1996-12-29 02:45:28 +00:00
|
|
|
}
|
1995-09-03 19:56:15 +00:00
|
|
|
|
|
|
|
tbp->b_blkno = blkno;
|
1996-06-03 04:40:35 +00:00
|
|
|
if( (tbp->b_flags & B_MALLOC) ||
|
|
|
|
((tbp->b_flags & B_VMIO) == 0) || (run <= 1) )
|
1995-09-03 19:56:15 +00:00
|
|
|
return tbp;
|
|
|
|
|
1999-01-21 08:29:12 +00:00
|
|
|
bp = trypbuf(&cluster_pbuf_freecnt);
|
1995-09-03 19:56:15 +00:00
|
|
|
if (bp == 0)
|
|
|
|
return tbp;
|
|
|
|
|
2001-10-21 06:12:06 +00:00
|
|
|
/*
|
|
|
|
* We are synthesizing a buffer out of vm_page_t's, but
|
|
|
|
* if the block size is not page aligned then the starting
|
|
|
|
* address may not be either. Inherit the b_data offset
|
|
|
|
* from the original buffer.
|
|
|
|
*/
|
1998-07-07 04:36:23 +00:00
|
|
|
bp->b_data = (char *)((vm_offset_t)bp->b_data |
|
|
|
|
((vm_offset_t)tbp->b_data & PAGE_MASK));
|
2000-03-20 10:44:49 +00:00
|
|
|
bp->b_flags = B_ASYNC | B_CLUSTER | B_VMIO;
|
|
|
|
bp->b_iocmd = BIO_READ;
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
bp->b_iodone = cluster_callback;
|
|
|
|
bp->b_blkno = blkno;
|
|
|
|
bp->b_lblkno = lbn;
|
Some VM improvements, including elimination of alot of Sig-11
problems. Tor Egge and others have helped with various VM bugs
lately, but don't blame him -- blame me!!!
pmap.c:
1) Create an object for kernel page table allocations. This
fixes a bogus allocation method previously used for such, by
grabbing pages from the kernel object, using bogus pindexes.
(This was a code cleanup, and perhaps a minor system stability
issue.)
pmap.c:
2) Pre-set the modify and accessed bits when prudent. This will
decrease bus traffic under certain circumstances.
vfs_bio.c, vfs_cluster.c:
3) Rather than calculating the beginning virtual byte offset
multiple times, stick the offset into the buffer header, so
that the calculated offset can be reused. (Long long multiplies
are often expensive, and this is a probably unmeasurable performance
improvement, and code cleanup.)
vfs_bio.c:
4) Handle write recursion more intelligently (but not perfectly) so
that it is less likely to cause a system panic, and is also
much more robust.
vfs_bio.c:
5) getblk incorrectly wrote out blocks that are incorrectly sized.
The problem is fixed, and writes blocks out ONLY when B_DELWRI
is true.
vfs_bio.c:
6) Check that already constituted buffers have fully valid pages. If
not, then make sure that the B_CACHE bit is not set. (This was
a major source of Sig-11 type problems.)
vfs_bio.c:
7) Fix a potential system deadlock due to an incorrectly specified
sleep priority while waiting for a buffer write operation. The
change that I made opens the system up to serious problems, and
we need to examine the issue of process sleep priorities.
vfs_cluster.c, vfs_bio.c:
8) Make clustered reads work more correctly (and more completely)
when buffers are already constituted, but not fully valid.
(This was another system reliability issue.)
vfs_subr.c, ffs_inode.c:
9) Create a vtruncbuf function, which is used by filesystems that
can truncate files. The vinvalbuf forced a file sync type operation,
while vtruncbuf only invalidates the buffers past the new end of file,
and also invalidates the appropriate pages. (This was a system reliabiliy
and performance issue.)
10) Modify FFS to use vtruncbuf.
vm_object.c:
11) Make the object rundown mechanism for OBJT_VNODE type objects work
more correctly. Included in that fix, create pager entries for
the OBJT_DEAD pager type, so that paging requests that might slip
in during race conditions are properly handled. (This was a system
reliability issue.)
vm_page.c:
12) Make some of the page validation routines be a little less picky
about arguments passed to them. Also, support page invalidation
change the object generation count so that we handle generation
counts a little more robustly.
vm_pageout.c:
13) Further reduce pageout daemon activity when the system doesn't
need help from it. There should be no additional performance
decrease even when the pageout daemon is running. (This was
a significant performance issue.)
vnode_pager.c:
14) Teach the vnode pager to handle race conditions during vnode
deallocations.
1998-03-16 01:56:03 +00:00
|
|
|
bp->b_offset = tbp->b_offset;
|
1999-01-10 01:58:29 +00:00
|
|
|
KASSERT(bp->b_offset != NOOFFSET, ("cluster_rbuild: no buffer offset"));
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
pbgetvp(vp, bp);
|
1994-05-24 10:09:53 +00:00
|
|
|
|
1995-11-19 19:54:31 +00:00
|
|
|
TAILQ_INIT(&bp->b_cluster.cluster_head);
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
|
|
|
bp->b_bcount = 0;
|
|
|
|
bp->b_bufsize = 0;
|
|
|
|
bp->b_npages = 0;
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
inc = btodb(size);
|
1995-09-03 19:56:15 +00:00
|
|
|
for (bn = blkno, i = 0; i < run; ++i, bn += inc) {
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
if (i != 0) {
|
1995-12-11 04:58:34 +00:00
|
|
|
if ((bp->b_npages * PAGE_SIZE) +
|
2001-10-21 06:12:06 +00:00
|
|
|
round_page(size) > vp->v_mount->mnt_iosize_max) {
|
1995-09-03 19:56:15 +00:00
|
|
|
break;
|
2001-10-21 06:12:06 +00:00
|
|
|
}
|
1995-09-23 21:12:45 +00:00
|
|
|
|
2003-03-04 21:35:28 +00:00
|
|
|
tbp = getblk(vp, lbn + i, size, 0, 0, GB_LOCK_NOWAIT);
|
1995-12-11 04:58:34 +00:00
|
|
|
|
2003-03-04 21:35:28 +00:00
|
|
|
/* Don't wait around for locked bufs. */
|
|
|
|
if (tbp == NULL)
|
|
|
|
break;
|
1995-09-03 19:56:15 +00:00
|
|
|
|
2001-01-19 05:31:07 +00:00
|
|
|
/*
|
2001-10-21 06:12:06 +00:00
|
|
|
* Stop scanning if the buffer is fully valid
|
|
|
|
* (marked B_CACHE), or locked (may be doing a
|
|
|
|
* background write), or if the buffer is not
|
|
|
|
* VMIO backed. The clustering code can only deal
|
|
|
|
* with VMIO-backed buffers.
|
2001-01-19 05:31:07 +00:00
|
|
|
*/
|
2003-08-28 06:55:18 +00:00
|
|
|
VI_LOCK(bp->b_vp);
|
|
|
|
if ((tbp->b_vflags & BV_BKGRDINPROG) ||
|
|
|
|
(tbp->b_flags & B_CACHE) ||
|
|
|
|
(tbp->b_flags & B_VMIO) == 0) {
|
|
|
|
VI_UNLOCK(bp->b_vp);
|
1996-01-19 04:00:31 +00:00
|
|
|
bqrelse(tbp);
|
1995-09-03 19:56:15 +00:00
|
|
|
break;
|
|
|
|
}
|
2003-08-28 06:55:18 +00:00
|
|
|
VI_UNLOCK(bp->b_vp);
|
1995-09-03 19:56:15 +00:00
|
|
|
|
2001-10-21 06:12:06 +00:00
|
|
|
/*
|
|
|
|
* The buffer must be completely invalid in order to
|
|
|
|
* take part in the cluster. If it is partially valid
|
|
|
|
* then we stop.
|
|
|
|
*/
|
2003-10-20 05:57:55 +00:00
|
|
|
VM_OBJECT_LOCK(tbp->b_object);
|
2001-01-19 05:31:07 +00:00
|
|
|
for (j = 0;j < tbp->b_npages; j++) {
|
2003-10-20 05:57:55 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(tbp->b_pages[j]->object,
|
|
|
|
MA_OWNED);
|
Some VM improvements, including elimination of alot of Sig-11
problems. Tor Egge and others have helped with various VM bugs
lately, but don't blame him -- blame me!!!
pmap.c:
1) Create an object for kernel page table allocations. This
fixes a bogus allocation method previously used for such, by
grabbing pages from the kernel object, using bogus pindexes.
(This was a code cleanup, and perhaps a minor system stability
issue.)
pmap.c:
2) Pre-set the modify and accessed bits when prudent. This will
decrease bus traffic under certain circumstances.
vfs_bio.c, vfs_cluster.c:
3) Rather than calculating the beginning virtual byte offset
multiple times, stick the offset into the buffer header, so
that the calculated offset can be reused. (Long long multiplies
are often expensive, and this is a probably unmeasurable performance
improvement, and code cleanup.)
vfs_bio.c:
4) Handle write recursion more intelligently (but not perfectly) so
that it is less likely to cause a system panic, and is also
much more robust.
vfs_bio.c:
5) getblk incorrectly wrote out blocks that are incorrectly sized.
The problem is fixed, and writes blocks out ONLY when B_DELWRI
is true.
vfs_bio.c:
6) Check that already constituted buffers have fully valid pages. If
not, then make sure that the B_CACHE bit is not set. (This was
a major source of Sig-11 type problems.)
vfs_bio.c:
7) Fix a potential system deadlock due to an incorrectly specified
sleep priority while waiting for a buffer write operation. The
change that I made opens the system up to serious problems, and
we need to examine the issue of process sleep priorities.
vfs_cluster.c, vfs_bio.c:
8) Make clustered reads work more correctly (and more completely)
when buffers are already constituted, but not fully valid.
(This was another system reliability issue.)
vfs_subr.c, ffs_inode.c:
9) Create a vtruncbuf function, which is used by filesystems that
can truncate files. The vinvalbuf forced a file sync type operation,
while vtruncbuf only invalidates the buffers past the new end of file,
and also invalidates the appropriate pages. (This was a system reliabiliy
and performance issue.)
10) Modify FFS to use vtruncbuf.
vm_object.c:
11) Make the object rundown mechanism for OBJT_VNODE type objects work
more correctly. Included in that fix, create pager entries for
the OBJT_DEAD pager type, so that paging requests that might slip
in during race conditions are properly handled. (This was a system
reliability issue.)
vm_page.c:
12) Make some of the page validation routines be a little less picky
about arguments passed to them. Also, support page invalidation
change the object generation count so that we handle generation
counts a little more robustly.
vm_pageout.c:
13) Further reduce pageout daemon activity when the system doesn't
need help from it. There should be no additional performance
decrease even when the pageout daemon is running. (This was
a significant performance issue.)
vnode_pager.c:
14) Teach the vnode pager to handle race conditions during vnode
deallocations.
1998-03-16 01:56:03 +00:00
|
|
|
if (tbp->b_pages[j]->valid)
|
1995-09-03 19:56:15 +00:00
|
|
|
break;
|
2001-01-19 05:31:07 +00:00
|
|
|
}
|
2003-10-20 05:57:55 +00:00
|
|
|
VM_OBJECT_UNLOCK(tbp->b_object);
|
1995-09-03 19:56:15 +00:00
|
|
|
if (j != tbp->b_npages) {
|
Some VM improvements, including elimination of alot of Sig-11
problems. Tor Egge and others have helped with various VM bugs
lately, but don't blame him -- blame me!!!
pmap.c:
1) Create an object for kernel page table allocations. This
fixes a bogus allocation method previously used for such, by
grabbing pages from the kernel object, using bogus pindexes.
(This was a code cleanup, and perhaps a minor system stability
issue.)
pmap.c:
2) Pre-set the modify and accessed bits when prudent. This will
decrease bus traffic under certain circumstances.
vfs_bio.c, vfs_cluster.c:
3) Rather than calculating the beginning virtual byte offset
multiple times, stick the offset into the buffer header, so
that the calculated offset can be reused. (Long long multiplies
are often expensive, and this is a probably unmeasurable performance
improvement, and code cleanup.)
vfs_bio.c:
4) Handle write recursion more intelligently (but not perfectly) so
that it is less likely to cause a system panic, and is also
much more robust.
vfs_bio.c:
5) getblk incorrectly wrote out blocks that are incorrectly sized.
The problem is fixed, and writes blocks out ONLY when B_DELWRI
is true.
vfs_bio.c:
6) Check that already constituted buffers have fully valid pages. If
not, then make sure that the B_CACHE bit is not set. (This was
a major source of Sig-11 type problems.)
vfs_bio.c:
7) Fix a potential system deadlock due to an incorrectly specified
sleep priority while waiting for a buffer write operation. The
change that I made opens the system up to serious problems, and
we need to examine the issue of process sleep priorities.
vfs_cluster.c, vfs_bio.c:
8) Make clustered reads work more correctly (and more completely)
when buffers are already constituted, but not fully valid.
(This was another system reliability issue.)
vfs_subr.c, ffs_inode.c:
9) Create a vtruncbuf function, which is used by filesystems that
can truncate files. The vinvalbuf forced a file sync type operation,
while vtruncbuf only invalidates the buffers past the new end of file,
and also invalidates the appropriate pages. (This was a system reliabiliy
and performance issue.)
10) Modify FFS to use vtruncbuf.
vm_object.c:
11) Make the object rundown mechanism for OBJT_VNODE type objects work
more correctly. Included in that fix, create pager entries for
the OBJT_DEAD pager type, so that paging requests that might slip
in during race conditions are properly handled. (This was a system
reliability issue.)
vm_page.c:
12) Make some of the page validation routines be a little less picky
about arguments passed to them. Also, support page invalidation
change the object generation count so that we handle generation
counts a little more robustly.
vm_pageout.c:
13) Further reduce pageout daemon activity when the system doesn't
need help from it. There should be no additional performance
decrease even when the pageout daemon is running. (This was
a significant performance issue.)
vnode_pager.c:
14) Teach the vnode pager to handle race conditions during vnode
deallocations.
1998-03-16 01:56:03 +00:00
|
|
|
bqrelse(tbp);
|
1995-09-03 19:56:15 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
2001-10-21 06:12:06 +00:00
|
|
|
/*
|
|
|
|
* Set a read-ahead mark as appropriate
|
|
|
|
*/
|
1996-12-29 02:45:28 +00:00
|
|
|
if ((fbp && (i == 1)) || (i == (run - 1)))
|
|
|
|
tbp->b_flags |= B_RAM;
|
2001-10-21 06:12:06 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Set the buffer up for an async read (XXX should
|
|
|
|
* we do this only if we do not wind up brelse()ing?).
|
|
|
|
* Set the block number if it isn't set, otherwise
|
|
|
|
* if it is make sure it matches the block number we
|
|
|
|
* expect.
|
|
|
|
*/
|
2000-03-20 10:44:49 +00:00
|
|
|
tbp->b_flags |= B_ASYNC;
|
|
|
|
tbp->b_iocmd = BIO_READ;
|
1995-12-11 04:58:34 +00:00
|
|
|
if (tbp->b_blkno == tbp->b_lblkno) {
|
1995-09-03 19:56:15 +00:00
|
|
|
tbp->b_blkno = bn;
|
|
|
|
} else if (tbp->b_blkno != bn) {
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
brelse(tbp);
|
|
|
|
break;
|
|
|
|
}
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
1999-06-29 05:59:47 +00:00
|
|
|
/*
|
|
|
|
* XXX fbp from caller may not be B_ASYNC, but we are going
|
|
|
|
* to biodone() it in cluster_callback() anyway
|
|
|
|
*/
|
|
|
|
BUF_KERNPROC(tbp);
|
1995-11-19 19:54:31 +00:00
|
|
|
TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head,
|
|
|
|
tbp, b_cluster.cluster_entry);
|
2003-10-20 05:57:55 +00:00
|
|
|
VM_OBJECT_LOCK(tbp->b_object);
|
2002-07-31 07:27:08 +00:00
|
|
|
vm_page_lock_queues();
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
for (j = 0; j < tbp->b_npages; j += 1) {
|
1995-09-03 19:56:15 +00:00
|
|
|
vm_page_t m;
|
|
|
|
m = tbp->b_pages[j];
|
1998-09-04 08:06:57 +00:00
|
|
|
vm_page_io_start(m);
|
1998-08-24 08:39:39 +00:00
|
|
|
vm_object_pip_add(m->object, 1);
|
1995-09-03 19:56:15 +00:00
|
|
|
if ((bp->b_npages == 0) ||
|
1995-11-20 04:53:45 +00:00
|
|
|
(bp->b_pages[bp->b_npages-1] != m)) {
|
1995-09-03 19:56:15 +00:00
|
|
|
bp->b_pages[bp->b_npages] = m;
|
|
|
|
bp->b_npages++;
|
|
|
|
}
|
1996-10-06 07:50:05 +00:00
|
|
|
if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL)
|
|
|
|
tbp->b_pages[j] = bogus_page;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
2002-07-31 07:27:08 +00:00
|
|
|
vm_page_unlock_queues();
|
2003-10-20 05:57:55 +00:00
|
|
|
VM_OBJECT_UNLOCK(tbp->b_object);
|
2001-10-25 22:49:48 +00:00
|
|
|
/*
|
|
|
|
* XXX shouldn't this be += size for both, like in
|
|
|
|
* cluster_wbuild()?
|
|
|
|
*
|
|
|
|
* Don't inherit tbp->b_bufsize as it may be larger due to
|
|
|
|
* a non-page-aligned size. Instead just aggregate using
|
|
|
|
* 'size'.
|
|
|
|
*/
|
|
|
|
if (tbp->b_bcount != size)
|
|
|
|
printf("warning: tbp->b_bcount wrong %ld vs %ld\n", tbp->b_bcount, size);
|
|
|
|
if (tbp->b_bufsize != size)
|
|
|
|
printf("warning: tbp->b_bufsize wrong %ld vs %ld\n", tbp->b_bufsize, size);
|
|
|
|
bp->b_bcount += size;
|
|
|
|
bp->b_bufsize += size;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
1996-10-06 07:50:05 +00:00
|
|
|
|
2001-10-21 06:12:06 +00:00
|
|
|
/*
|
|
|
|
* Fully valid pages in the cluster are already good and do not need
|
|
|
|
* to be re-read from disk. Replace the page with bogus_page
|
|
|
|
*/
|
2003-10-20 05:57:55 +00:00
|
|
|
VM_OBJECT_LOCK(bp->b_object);
|
2001-10-21 06:12:06 +00:00
|
|
|
for (j = 0; j < bp->b_npages; j++) {
|
2003-10-20 05:57:55 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(bp->b_pages[j]->object, MA_OWNED);
|
1996-10-06 07:50:05 +00:00
|
|
|
if ((bp->b_pages[j]->valid & VM_PAGE_BITS_ALL) ==
|
2001-10-21 06:12:06 +00:00
|
|
|
VM_PAGE_BITS_ALL) {
|
1996-10-06 07:50:05 +00:00
|
|
|
bp->b_pages[j] = bogus_page;
|
2001-10-21 06:12:06 +00:00
|
|
|
}
|
1996-10-06 07:50:05 +00:00
|
|
|
}
|
2003-10-20 05:57:55 +00:00
|
|
|
VM_OBJECT_UNLOCK(bp->b_object);
|
1996-11-30 22:41:49 +00:00
|
|
|
if (bp->b_bufsize > bp->b_kvasize)
|
1998-07-11 10:45:45 +00:00
|
|
|
panic("cluster_rbuild: b_bufsize(%ld) > b_kvasize(%d)\n",
|
|
|
|
bp->b_bufsize, bp->b_kvasize);
|
1996-11-30 22:41:49 +00:00
|
|
|
bp->b_kvasize = bp->b_bufsize;
|
1996-10-06 07:50:05 +00:00
|
|
|
|
1995-09-03 19:56:15 +00:00
|
|
|
pmap_qenter(trunc_page((vm_offset_t) bp->b_data),
|
|
|
|
(vm_page_t *)bp->b_pages, bp->b_npages);
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
return (bp);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Cleanup after a clustered read or write.
|
|
|
|
* This is complicated by the fact that any of the buffers might have
|
|
|
|
* extra memory (if there were no empty buffer headers at allocbuf time)
|
|
|
|
* that we will need to shift around.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
cluster_callback(bp)
|
|
|
|
struct buf *bp;
|
|
|
|
{
|
1995-11-19 19:54:31 +00:00
|
|
|
struct buf *nbp, *tbp;
|
1994-05-24 10:09:53 +00:00
|
|
|
int error = 0;
|
|
|
|
|
2001-07-04 16:20:28 +00:00
|
|
|
GIANT_REQUIRED;
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
|
|
|
* Must propogate errors to all the components.
|
|
|
|
*/
|
2000-04-02 15:24:56 +00:00
|
|
|
if (bp->b_ioflags & BIO_ERROR)
|
1994-05-24 10:09:53 +00:00
|
|
|
error = bp->b_error;
|
|
|
|
|
1995-09-03 19:56:15 +00:00
|
|
|
pmap_qremove(trunc_page((vm_offset_t) bp->b_data), bp->b_npages);
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
|
|
|
* Move memory from the large cluster buffer into the component
|
|
|
|
* buffers and mark IO as done on these.
|
|
|
|
*/
|
1996-12-29 02:45:28 +00:00
|
|
|
for (tbp = TAILQ_FIRST(&bp->b_cluster.cluster_head);
|
1995-11-19 19:54:31 +00:00
|
|
|
tbp; tbp = nbp) {
|
1996-12-29 02:45:28 +00:00
|
|
|
nbp = TAILQ_NEXT(&tbp->b_cluster, cluster_entry);
|
1994-05-24 10:09:53 +00:00
|
|
|
if (error) {
|
2000-04-02 15:24:56 +00:00
|
|
|
tbp->b_ioflags |= BIO_ERROR;
|
1994-05-24 10:09:53 +00:00
|
|
|
tbp->b_error = error;
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
} else {
|
|
|
|
tbp->b_dirtyoff = tbp->b_dirtyend = 0;
|
2000-04-02 15:24:56 +00:00
|
|
|
tbp->b_flags &= ~B_INVAL;
|
|
|
|
tbp->b_ioflags &= ~BIO_ERROR;
|
2001-05-24 07:22:27 +00:00
|
|
|
/*
|
|
|
|
* XXX the bdwrite()/bqrelse() issued during
|
|
|
|
* cluster building clears B_RELBUF (see bqrelse()
|
|
|
|
* comment). If direct I/O was specified, we have
|
|
|
|
* to restore it here to allow the buffer and VM
|
|
|
|
* to be freed.
|
|
|
|
*/
|
|
|
|
if (tbp->b_flags & B_DIRECT)
|
|
|
|
tbp->b_flags |= B_RELBUF;
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
}
|
2000-04-15 05:54:02 +00:00
|
|
|
bufdone(tbp);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
1999-01-21 08:29:12 +00:00
|
|
|
relpbuf(bp, &cluster_pbuf_freecnt);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
1999-07-04 00:31:17 +00:00
|
|
|
/*
|
|
|
|
* cluster_wbuild_wb:
|
|
|
|
*
|
|
|
|
* Implement modified write build for cluster.
|
|
|
|
*
|
|
|
|
* write_behind = 0 write behind disabled
|
|
|
|
* write_behind = 1 write behind normal (default)
|
|
|
|
* write_behind = 2 write behind backed-off
|
|
|
|
*/
|
|
|
|
|
|
|
|
static __inline int
|
|
|
|
cluster_wbuild_wb(struct vnode *vp, long size, daddr_t start_lbn, int len)
|
|
|
|
{
|
|
|
|
int r = 0;
|
|
|
|
|
|
|
|
switch(write_behind) {
|
|
|
|
case 2:
|
|
|
|
if (start_lbn < len)
|
|
|
|
break;
|
|
|
|
start_lbn -= len;
|
2002-08-25 13:23:09 +00:00
|
|
|
/* FALLTHROUGH */
|
1999-07-04 00:31:17 +00:00
|
|
|
case 1:
|
|
|
|
r = cluster_wbuild(vp, size, start_lbn, len);
|
2002-08-25 13:23:09 +00:00
|
|
|
/* FALLTHROUGH */
|
1999-07-04 00:31:17 +00:00
|
|
|
default:
|
2002-08-25 13:23:09 +00:00
|
|
|
/* FALLTHROUGH */
|
1999-07-04 00:31:17 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
return(r);
|
|
|
|
}
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
|
|
|
* Do clustered write for FFS.
|
|
|
|
*
|
|
|
|
* Three cases:
|
|
|
|
* 1. Write is not sequential (write asynchronously)
|
|
|
|
* Write is sequential:
|
|
|
|
* 2. beginning of cluster - begin cluster
|
|
|
|
* 3. middle of a cluster - add to cluster
|
|
|
|
* 4. end of a cluster - asynchronously write cluster
|
|
|
|
*/
|
|
|
|
void
|
2004-09-27 19:14:10 +00:00
|
|
|
cluster_write(struct vnode *vp, struct buf *bp, u_quad_t filesize, int seqcount)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
daddr_t lbn;
|
|
|
|
int maxclen, cursize;
|
|
|
|
int lblocksize;
|
1995-11-19 19:54:31 +00:00
|
|
|
int async;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
Make our v_usecount vnode reference count work identically to the
original BSD code. The association between the vnode and the vm_object
no longer includes reference counts. The major difference is that
vm_object's are no longer freed gratuitiously from the vnode, and so
once an object is created for the vnode, it will last as long as the
vnode does.
When a vnode object reference count is incremented, then the underlying
vnode reference count is incremented also. The two "objects" are now
more intimately related, and so the interactions are now much less
complex.
When vnodes are now normally placed onto the free queue with an object still
attached. The rundown of the object happens at vnode rundown time, and
happens with exactly the same filesystem semantics of the original VFS
code. There is absolutely no need for vnode_pager_uncache and other
travesties like that anymore.
A side-effect of these changes is that SMP locking should be much simpler,
the I/O copyin/copyout optimizations work, NFS should be more ponderable,
and further work on layered filesystems should be less frustrating, because
of the totally coherent management of the vnode objects and vnodes.
Please be careful with your system while running this code, but I would
greatly appreciate feedback as soon a reasonably possible.
1998-01-06 05:26:17 +00:00
|
|
|
if (vp->v_type == VREG) {
|
|
|
|
async = vp->v_mount->mnt_flag & MNT_ASYNC;
|
|
|
|
lblocksize = vp->v_mount->mnt_stat.f_iosize;
|
|
|
|
} else {
|
|
|
|
async = 0;
|
|
|
|
lblocksize = bp->b_bufsize;
|
|
|
|
}
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
lbn = bp->b_lblkno;
|
1999-01-08 17:31:30 +00:00
|
|
|
KASSERT(bp->b_offset != NOOFFSET, ("cluster_write: no buffer offset"));
|
In kern_physio.c fix tsleep priority messup.
In vfs_bio.c, remove b_generation count usage,
remove redundant reassignbuf,
remove redundant spl(s),
manage page PG_ZERO flags more correctly,
utilize in invalid value for b_offset until it
is properly initialized. Add asserts
for #ifdef DIAGNOSTIC, when b_offset is
improperly used.
when a process is not performing I/O, and just waiting
on a buffer generally, make the sleep priority
low.
only check page validity in getblk for B_VMIO buffers.
In vfs_cluster, add b_offset asserts, correct pointer calculation
for clustered reads. Improve readability of certain parts of
the code. Remove redundant spl(s).
In vfs_subr, correct usage of vfs_bio_awrite (From Andrew Gallatin
<gallatin@cs.duke.edu>). More vtruncbuf problems fixed.
1998-03-19 22:48:16 +00:00
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/* Initialize vnode to beginning of file. */
|
|
|
|
if (lbn == 0)
|
|
|
|
vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
|
|
|
|
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
if (vp->v_clen == 0 || lbn != vp->v_lastw + 1 ||
|
|
|
|
(bp->b_blkno != vp->v_lasta + btodb(lblocksize))) {
|
1999-09-29 20:05:33 +00:00
|
|
|
maxclen = vp->v_mount->mnt_iosize_max / lblocksize - 1;
|
1994-05-24 10:09:53 +00:00
|
|
|
if (vp->v_clen != 0) {
|
|
|
|
/*
|
|
|
|
* Next block is not sequential.
|
1995-05-30 08:16:23 +00:00
|
|
|
*
|
1994-05-24 10:09:53 +00:00
|
|
|
* If we are not writing at end of file, the process
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
* seeked to another point in the file since its last
|
|
|
|
* write, or we have reached our maximum cluster size,
|
|
|
|
* then push the previous cluster. Otherwise try
|
|
|
|
* reallocating to make it sequential.
|
2000-04-02 00:55:28 +00:00
|
|
|
*
|
|
|
|
* Change to algorithm: only push previous cluster if
|
|
|
|
* it was sequential from the point of view of the
|
|
|
|
* seqcount heuristic, otherwise leave the buffer
|
|
|
|
* intact so we can potentially optimize the I/O
|
|
|
|
* later on in the buf_daemon or update daemon
|
|
|
|
* flush.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
|
|
|
cursize = vp->v_lastw - vp->v_cstart + 1;
|
Some VM improvements, including elimination of alot of Sig-11
problems. Tor Egge and others have helped with various VM bugs
lately, but don't blame him -- blame me!!!
pmap.c:
1) Create an object for kernel page table allocations. This
fixes a bogus allocation method previously used for such, by
grabbing pages from the kernel object, using bogus pindexes.
(This was a code cleanup, and perhaps a minor system stability
issue.)
pmap.c:
2) Pre-set the modify and accessed bits when prudent. This will
decrease bus traffic under certain circumstances.
vfs_bio.c, vfs_cluster.c:
3) Rather than calculating the beginning virtual byte offset
multiple times, stick the offset into the buffer header, so
that the calculated offset can be reused. (Long long multiplies
are often expensive, and this is a probably unmeasurable performance
improvement, and code cleanup.)
vfs_bio.c:
4) Handle write recursion more intelligently (but not perfectly) so
that it is less likely to cause a system panic, and is also
much more robust.
vfs_bio.c:
5) getblk incorrectly wrote out blocks that are incorrectly sized.
The problem is fixed, and writes blocks out ONLY when B_DELWRI
is true.
vfs_bio.c:
6) Check that already constituted buffers have fully valid pages. If
not, then make sure that the B_CACHE bit is not set. (This was
a major source of Sig-11 type problems.)
vfs_bio.c:
7) Fix a potential system deadlock due to an incorrectly specified
sleep priority while waiting for a buffer write operation. The
change that I made opens the system up to serious problems, and
we need to examine the issue of process sleep priorities.
vfs_cluster.c, vfs_bio.c:
8) Make clustered reads work more correctly (and more completely)
when buffers are already constituted, but not fully valid.
(This was another system reliability issue.)
vfs_subr.c, ffs_inode.c:
9) Create a vtruncbuf function, which is used by filesystems that
can truncate files. The vinvalbuf forced a file sync type operation,
while vtruncbuf only invalidates the buffers past the new end of file,
and also invalidates the appropriate pages. (This was a system reliabiliy
and performance issue.)
10) Modify FFS to use vtruncbuf.
vm_object.c:
11) Make the object rundown mechanism for OBJT_VNODE type objects work
more correctly. Included in that fix, create pager entries for
the OBJT_DEAD pager type, so that paging requests that might slip
in during race conditions are properly handled. (This was a system
reliability issue.)
vm_page.c:
12) Make some of the page validation routines be a little less picky
about arguments passed to them. Also, support page invalidation
change the object generation count so that we handle generation
counts a little more robustly.
vm_pageout.c:
13) Further reduce pageout daemon activity when the system doesn't
need help from it. There should be no additional performance
decrease even when the pageout daemon is running. (This was
a significant performance issue.)
vnode_pager.c:
14) Teach the vnode pager to handle race conditions during vnode
deallocations.
1998-03-16 01:56:03 +00:00
|
|
|
if (((u_quad_t) bp->b_offset + lblocksize) != filesize ||
|
1995-09-03 19:56:15 +00:00
|
|
|
lbn != vp->v_lastw + 1 || vp->v_clen <= cursize) {
|
2000-04-02 00:55:28 +00:00
|
|
|
if (!async && seqcount > 0) {
|
1999-07-08 06:06:00 +00:00
|
|
|
cluster_wbuild_wb(vp, lblocksize,
|
1995-11-19 19:54:31 +00:00
|
|
|
vp->v_cstart, cursize);
|
2000-04-02 00:55:28 +00:00
|
|
|
}
|
1995-09-03 19:56:15 +00:00
|
|
|
} else {
|
|
|
|
struct buf **bpp, **endbp;
|
|
|
|
struct cluster_save *buflist;
|
|
|
|
|
|
|
|
buflist = cluster_collectbufs(vp, bp);
|
|
|
|
endbp = &buflist->bs_children
|
|
|
|
[buflist->bs_nchildren - 1];
|
|
|
|
if (VOP_REALLOCBLKS(vp, buflist)) {
|
|
|
|
/*
|
2000-04-02 00:55:28 +00:00
|
|
|
* Failed, push the previous cluster
|
|
|
|
* if *really* writing sequentially
|
|
|
|
* in the logical file (seqcount > 1),
|
|
|
|
* otherwise delay it in the hopes that
|
|
|
|
* the low level disk driver can
|
|
|
|
* optimize the write ordering.
|
1995-09-03 19:56:15 +00:00
|
|
|
*/
|
|
|
|
for (bpp = buflist->bs_children;
|
|
|
|
bpp < endbp; bpp++)
|
|
|
|
brelse(*bpp);
|
|
|
|
free(buflist, M_SEGMENT);
|
2000-04-02 00:55:28 +00:00
|
|
|
if (seqcount > 1) {
|
|
|
|
cluster_wbuild_wb(vp,
|
|
|
|
lblocksize, vp->v_cstart,
|
|
|
|
cursize);
|
|
|
|
}
|
1995-09-03 19:56:15 +00:00
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* Succeeded, keep building cluster.
|
|
|
|
*/
|
|
|
|
for (bpp = buflist->bs_children;
|
|
|
|
bpp <= endbp; bpp++)
|
|
|
|
bdwrite(*bpp);
|
|
|
|
free(buflist, M_SEGMENT);
|
|
|
|
vp->v_lastw = lbn;
|
|
|
|
vp->v_lasta = bp->b_blkno;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
/*
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
* Consider beginning a cluster. If at end of file, make
|
|
|
|
* cluster as large as possible, otherwise find size of
|
|
|
|
* existing cluster.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
Make our v_usecount vnode reference count work identically to the
original BSD code. The association between the vnode and the vm_object
no longer includes reference counts. The major difference is that
vm_object's are no longer freed gratuitiously from the vnode, and so
once an object is created for the vnode, it will last as long as the
vnode does.
When a vnode object reference count is incremented, then the underlying
vnode reference count is incremented also. The two "objects" are now
more intimately related, and so the interactions are now much less
complex.
When vnodes are now normally placed onto the free queue with an object still
attached. The rundown of the object happens at vnode rundown time, and
happens with exactly the same filesystem semantics of the original VFS
code. There is absolutely no need for vnode_pager_uncache and other
travesties like that anymore.
A side-effect of these changes is that SMP locking should be much simpler,
the I/O copyin/copyout optimizations work, NFS should be more ponderable,
and further work on layered filesystems should be less frustrating, because
of the totally coherent management of the vnode objects and vnodes.
Please be careful with your system while running this code, but I would
greatly appreciate feedback as soon a reasonably possible.
1998-01-06 05:26:17 +00:00
|
|
|
if ((vp->v_type == VREG) &&
|
Some VM improvements, including elimination of alot of Sig-11
problems. Tor Egge and others have helped with various VM bugs
lately, but don't blame him -- blame me!!!
pmap.c:
1) Create an object for kernel page table allocations. This
fixes a bogus allocation method previously used for such, by
grabbing pages from the kernel object, using bogus pindexes.
(This was a code cleanup, and perhaps a minor system stability
issue.)
pmap.c:
2) Pre-set the modify and accessed bits when prudent. This will
decrease bus traffic under certain circumstances.
vfs_bio.c, vfs_cluster.c:
3) Rather than calculating the beginning virtual byte offset
multiple times, stick the offset into the buffer header, so
that the calculated offset can be reused. (Long long multiplies
are often expensive, and this is a probably unmeasurable performance
improvement, and code cleanup.)
vfs_bio.c:
4) Handle write recursion more intelligently (but not perfectly) so
that it is less likely to cause a system panic, and is also
much more robust.
vfs_bio.c:
5) getblk incorrectly wrote out blocks that are incorrectly sized.
The problem is fixed, and writes blocks out ONLY when B_DELWRI
is true.
vfs_bio.c:
6) Check that already constituted buffers have fully valid pages. If
not, then make sure that the B_CACHE bit is not set. (This was
a major source of Sig-11 type problems.)
vfs_bio.c:
7) Fix a potential system deadlock due to an incorrectly specified
sleep priority while waiting for a buffer write operation. The
change that I made opens the system up to serious problems, and
we need to examine the issue of process sleep priorities.
vfs_cluster.c, vfs_bio.c:
8) Make clustered reads work more correctly (and more completely)
when buffers are already constituted, but not fully valid.
(This was another system reliability issue.)
vfs_subr.c, ffs_inode.c:
9) Create a vtruncbuf function, which is used by filesystems that
can truncate files. The vinvalbuf forced a file sync type operation,
while vtruncbuf only invalidates the buffers past the new end of file,
and also invalidates the appropriate pages. (This was a system reliabiliy
and performance issue.)
10) Modify FFS to use vtruncbuf.
vm_object.c:
11) Make the object rundown mechanism for OBJT_VNODE type objects work
more correctly. Included in that fix, create pager entries for
the OBJT_DEAD pager type, so that paging requests that might slip
in during race conditions are properly handled. (This was a system
reliability issue.)
vm_page.c:
12) Make some of the page validation routines be a little less picky
about arguments passed to them. Also, support page invalidation
change the object generation count so that we handle generation
counts a little more robustly.
vm_pageout.c:
13) Further reduce pageout daemon activity when the system doesn't
need help from it. There should be no additional performance
decrease even when the pageout daemon is running. (This was
a significant performance issue.)
vnode_pager.c:
14) Teach the vnode pager to handle race conditions during vnode
deallocations.
1998-03-16 01:56:03 +00:00
|
|
|
((u_quad_t) bp->b_offset + lblocksize) != filesize &&
|
1995-04-04 02:10:17 +00:00
|
|
|
(bp->b_blkno == bp->b_lblkno) &&
|
1995-09-04 00:21:16 +00:00
|
|
|
(VOP_BMAP(vp, lbn, NULL, &bp->b_blkno, &maxclen, NULL) ||
|
1995-09-03 19:56:15 +00:00
|
|
|
bp->b_blkno == -1)) {
|
1994-05-24 10:09:53 +00:00
|
|
|
bawrite(bp);
|
|
|
|
vp->v_clen = 0;
|
|
|
|
vp->v_lasta = bp->b_blkno;
|
|
|
|
vp->v_cstart = lbn + 1;
|
|
|
|
vp->v_lastw = lbn;
|
|
|
|
return;
|
|
|
|
}
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
vp->v_clen = maxclen;
|
1995-11-19 19:54:31 +00:00
|
|
|
if (!async && maxclen == 0) { /* I/O not contiguous */
|
1994-05-24 10:09:53 +00:00
|
|
|
vp->v_cstart = lbn + 1;
|
1996-01-19 04:00:31 +00:00
|
|
|
bawrite(bp);
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
} else { /* Wait for rest of cluster */
|
1994-05-24 10:09:53 +00:00
|
|
|
vp->v_cstart = lbn;
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
bdwrite(bp);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
} else if (lbn == vp->v_cstart + vp->v_clen) {
|
|
|
|
/*
|
2000-04-02 00:55:28 +00:00
|
|
|
* At end of cluster, write it out if seqcount tells us we
|
|
|
|
* are operating sequentially, otherwise let the buf or
|
|
|
|
* update daemon handle it.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1995-11-19 19:54:31 +00:00
|
|
|
bdwrite(bp);
|
2000-04-02 00:55:28 +00:00
|
|
|
if (seqcount > 1)
|
|
|
|
cluster_wbuild_wb(vp, lblocksize, vp->v_cstart, vp->v_clen + 1);
|
1994-05-24 10:09:53 +00:00
|
|
|
vp->v_clen = 0;
|
|
|
|
vp->v_cstart = lbn + 1;
|
Implement a low-memory deadlock solution.
Removed most of the hacks that were trying to deal with low-memory
situations prior to now.
The new code is based on the concept that I/O must be able to function in
a low memory situation. All major modules related to I/O (except
networking) have been adjusted to allow allocation out of the system
reserve memory pool. These modules now detect a low memory situation but
rather then block they instead continue to operate, then return resources
to the memory pool instead of cache them or leave them wired.
Code has been added to stall in a low-memory situation prior to a vnode
being locked.
Thus situations where a process blocks in a low-memory condition while
holding a locked vnode have been reduced to near nothing. Not only will
I/O continue to operate, but many prior deadlock conditions simply no
longer exist.
Implement a number of VFS/BIO fixes
(found by Ian): in biodone(), bogus-page replacement code, the loop
was not properly incrementing loop variables prior to a continue
statement. We do not believe this code can be hit anyway but we
aren't taking any chances. We'll turn the whole section into a
panic (as it already is in brelse()) after the release is rolled.
In biodone(), the foff calculation was incorrectly
clamped to the iosize, causing the wrong foff to be calculated
for pages in the case of an I/O error or biodone() called without
initiating I/O. The problem always caused a panic before. Now it
doesn't. The problem is mainly an issue with NFS.
Fixed casts for ~PAGE_MASK. This code worked properly before only
because the calculations use signed arithmatic. Better to properly
extend PAGE_MASK first before inverting it for the 64 bit masking
op.
In brelse(), the bogus_page fixup code was improperly throwing
away the original contents of 'm' when it did the j-loop to
fix the bogus pages. The result was that it would potentially
invalidate parts of the *WRONG* page(!), leading to corruption.
There may still be cases where a background bitmap write is
being duplicated, causing potential corruption. We have identified
a potentially serious bug related to this but the fix is still TBD.
So instead this patch contains a KASSERT to detect the problem
and panic the machine rather then continue to corrupt the filesystem.
The problem does not occur very often.. it is very hard to
reproduce, and it may or may not be the cause of the corruption
people have reported.
Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>)
Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
|
|
|
} else if (vm_page_count_severe()) {
|
|
|
|
/*
|
|
|
|
* We are low on memory, get it going NOW
|
|
|
|
*/
|
|
|
|
bawrite(bp);
|
2000-04-02 00:55:28 +00:00
|
|
|
} else {
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
* In the middle of a cluster, so just delay the I/O for now.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
|
|
|
bdwrite(bp);
|
2000-04-02 00:55:28 +00:00
|
|
|
}
|
1994-05-24 10:09:53 +00:00
|
|
|
vp->v_lastw = lbn;
|
|
|
|
vp->v_lasta = bp->b_blkno;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is an awful lot like cluster_rbuild...wish they could be combined.
|
|
|
|
* The last lbn argument is the current block on which I/O is being
|
|
|
|
* performed. Check to see that it doesn't fall in the middle of
|
|
|
|
* the current block (if last_bp == NULL).
|
|
|
|
*/
|
1995-12-11 04:58:34 +00:00
|
|
|
int
|
1995-11-19 19:54:31 +00:00
|
|
|
cluster_wbuild(vp, size, start_lbn, len)
|
1994-05-24 10:09:53 +00:00
|
|
|
struct vnode *vp;
|
|
|
|
long size;
|
|
|
|
daddr_t start_lbn;
|
|
|
|
int len;
|
|
|
|
{
|
1995-11-19 19:54:31 +00:00
|
|
|
struct buf *bp, *tbp;
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
int i, j, s;
|
1995-12-11 04:58:34 +00:00
|
|
|
int totalwritten = 0;
|
1995-11-19 19:54:31 +00:00
|
|
|
int dbsize = btodb(size);
|
1998-05-01 16:29:27 +00:00
|
|
|
|
2001-07-04 16:20:28 +00:00
|
|
|
GIANT_REQUIRED;
|
|
|
|
|
1995-12-11 04:58:34 +00:00
|
|
|
while (len > 0) {
|
|
|
|
s = splbio();
|
2001-01-19 05:31:07 +00:00
|
|
|
/*
|
|
|
|
* If the buffer is not delayed-write (i.e. dirty), or it
|
|
|
|
* is delayed-write but either locked or inval, it cannot
|
2001-02-06 09:18:39 +00:00
|
|
|
* partake in the clustered write.
|
2001-01-19 05:31:07 +00:00
|
|
|
*/
|
2003-03-04 21:35:28 +00:00
|
|
|
VI_LOCK(vp);
|
2004-10-22 08:47:20 +00:00
|
|
|
if ((tbp = gbincore(&vp->v_bufobj, start_lbn)) == NULL ||
|
2003-08-28 06:55:18 +00:00
|
|
|
(tbp->b_vflags & BV_BKGRDINPROG)) {
|
2003-03-04 21:35:28 +00:00
|
|
|
VI_UNLOCK(vp);
|
|
|
|
++start_lbn;
|
|
|
|
--len;
|
|
|
|
splx(s);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if (BUF_LOCK(tbp,
|
|
|
|
LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, VI_MTX(vp))) {
|
|
|
|
++start_lbn;
|
|
|
|
--len;
|
|
|
|
splx(s);
|
|
|
|
continue;
|
|
|
|
}
|
2003-08-28 06:55:18 +00:00
|
|
|
if ((tbp->b_flags & (B_INVAL | B_DELWRI)) != B_DELWRI) {
|
2003-03-04 21:35:28 +00:00
|
|
|
BUF_UNLOCK(tbp);
|
1995-12-11 04:58:34 +00:00
|
|
|
++start_lbn;
|
|
|
|
--len;
|
|
|
|
splx(s);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
bremfree(tbp);
|
|
|
|
tbp->b_flags &= ~B_DONE;
|
|
|
|
splx(s);
|
1995-03-02 19:36:50 +00:00
|
|
|
|
1999-06-17 01:25:25 +00:00
|
|
|
/*
|
|
|
|
* Extra memory in the buffer, punt on this buffer.
|
|
|
|
* XXX we could handle this in most cases, but we would
|
|
|
|
* have to push the extra memory down to after our max
|
|
|
|
* possible cluster size and then potentially pull it back
|
|
|
|
* up if the cluster was terminated prematurely--too much
|
|
|
|
* hassle.
|
|
|
|
*/
|
2000-11-17 23:40:08 +00:00
|
|
|
if (((tbp->b_flags & (B_CLUSTEROK | B_MALLOC | B_VMIO)) !=
|
|
|
|
(B_CLUSTEROK | B_VMIO)) ||
|
1998-03-16 18:39:41 +00:00
|
|
|
(tbp->b_bcount != tbp->b_bufsize) ||
|
|
|
|
(tbp->b_bcount != size) ||
|
|
|
|
(len == 1) ||
|
1999-06-16 15:54:30 +00:00
|
|
|
((bp = getpbuf(&cluster_pbuf_freecnt)) == NULL)) {
|
1995-12-11 04:58:34 +00:00
|
|
|
totalwritten += tbp->b_bufsize;
|
|
|
|
bawrite(tbp);
|
|
|
|
++start_lbn;
|
|
|
|
--len;
|
|
|
|
continue;
|
|
|
|
}
|
1995-11-19 19:54:31 +00:00
|
|
|
|
1998-03-16 18:39:41 +00:00
|
|
|
/*
|
|
|
|
* We got a pbuf to make the cluster in.
|
|
|
|
* so initialise it.
|
|
|
|
*/
|
1995-12-11 04:58:34 +00:00
|
|
|
TAILQ_INIT(&bp->b_cluster.cluster_head);
|
|
|
|
bp->b_bcount = 0;
|
2001-04-17 08:56:39 +00:00
|
|
|
bp->b_magic = tbp->b_magic;
|
|
|
|
bp->b_op = tbp->b_op;
|
2004-10-22 08:47:20 +00:00
|
|
|
bp->b_bufobj = tbp->b_bufobj;
|
1995-12-11 04:58:34 +00:00
|
|
|
bp->b_bufsize = 0;
|
|
|
|
bp->b_npages = 0;
|
2001-10-11 23:38:17 +00:00
|
|
|
if (tbp->b_wcred != NOCRED)
|
|
|
|
bp->b_wcred = crhold(tbp->b_wcred);
|
1995-12-11 04:58:34 +00:00
|
|
|
|
|
|
|
bp->b_blkno = tbp->b_blkno;
|
|
|
|
bp->b_lblkno = tbp->b_lblkno;
|
Some VM improvements, including elimination of alot of Sig-11
problems. Tor Egge and others have helped with various VM bugs
lately, but don't blame him -- blame me!!!
pmap.c:
1) Create an object for kernel page table allocations. This
fixes a bogus allocation method previously used for such, by
grabbing pages from the kernel object, using bogus pindexes.
(This was a code cleanup, and perhaps a minor system stability
issue.)
pmap.c:
2) Pre-set the modify and accessed bits when prudent. This will
decrease bus traffic under certain circumstances.
vfs_bio.c, vfs_cluster.c:
3) Rather than calculating the beginning virtual byte offset
multiple times, stick the offset into the buffer header, so
that the calculated offset can be reused. (Long long multiplies
are often expensive, and this is a probably unmeasurable performance
improvement, and code cleanup.)
vfs_bio.c:
4) Handle write recursion more intelligently (but not perfectly) so
that it is less likely to cause a system panic, and is also
much more robust.
vfs_bio.c:
5) getblk incorrectly wrote out blocks that are incorrectly sized.
The problem is fixed, and writes blocks out ONLY when B_DELWRI
is true.
vfs_bio.c:
6) Check that already constituted buffers have fully valid pages. If
not, then make sure that the B_CACHE bit is not set. (This was
a major source of Sig-11 type problems.)
vfs_bio.c:
7) Fix a potential system deadlock due to an incorrectly specified
sleep priority while waiting for a buffer write operation. The
change that I made opens the system up to serious problems, and
we need to examine the issue of process sleep priorities.
vfs_cluster.c, vfs_bio.c:
8) Make clustered reads work more correctly (and more completely)
when buffers are already constituted, but not fully valid.
(This was another system reliability issue.)
vfs_subr.c, ffs_inode.c:
9) Create a vtruncbuf function, which is used by filesystems that
can truncate files. The vinvalbuf forced a file sync type operation,
while vtruncbuf only invalidates the buffers past the new end of file,
and also invalidates the appropriate pages. (This was a system reliabiliy
and performance issue.)
10) Modify FFS to use vtruncbuf.
vm_object.c:
11) Make the object rundown mechanism for OBJT_VNODE type objects work
more correctly. Included in that fix, create pager entries for
the OBJT_DEAD pager type, so that paging requests that might slip
in during race conditions are properly handled. (This was a system
reliability issue.)
vm_page.c:
12) Make some of the page validation routines be a little less picky
about arguments passed to them. Also, support page invalidation
change the object generation count so that we handle generation
counts a little more robustly.
vm_pageout.c:
13) Further reduce pageout daemon activity when the system doesn't
need help from it. There should be no additional performance
decrease even when the pageout daemon is running. (This was
a significant performance issue.)
vnode_pager.c:
14) Teach the vnode pager to handle race conditions during vnode
deallocations.
1998-03-16 01:56:03 +00:00
|
|
|
bp->b_offset = tbp->b_offset;
|
2001-10-21 06:12:06 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* We are synthesizing a buffer out of vm_page_t's, but
|
|
|
|
* if the block size is not page aligned then the starting
|
|
|
|
* address may not be either. Inherit the b_data offset
|
|
|
|
* from the original buffer.
|
|
|
|
*/
|
1998-07-07 04:36:23 +00:00
|
|
|
bp->b_data = (char *)((vm_offset_t)bp->b_data |
|
|
|
|
((vm_offset_t)tbp->b_data & PAGE_MASK));
|
2000-03-20 10:44:49 +00:00
|
|
|
bp->b_flags |= B_CLUSTER |
|
2003-05-31 16:42:45 +00:00
|
|
|
(tbp->b_flags & (B_VMIO | B_NEEDCOMMIT));
|
1995-12-11 04:58:34 +00:00
|
|
|
bp->b_iodone = cluster_callback;
|
|
|
|
pbgetvp(vp, bp);
|
1998-03-16 18:39:41 +00:00
|
|
|
/*
|
|
|
|
* From this location in the file, scan forward to see
|
|
|
|
* if there are buffers with adjacent data that need to
|
|
|
|
* be written as well.
|
|
|
|
*/
|
1995-12-11 04:58:34 +00:00
|
|
|
for (i = 0; i < len; ++i, ++start_lbn) {
|
1998-03-16 18:39:41 +00:00
|
|
|
if (i != 0) { /* If not the first buffer */
|
1995-12-11 04:58:34 +00:00
|
|
|
s = splbio();
|
1998-03-16 18:39:41 +00:00
|
|
|
/*
|
|
|
|
* If the adjacent data is not even in core it
|
|
|
|
* can't need to be written.
|
|
|
|
*/
|
2003-03-04 21:35:28 +00:00
|
|
|
VI_LOCK(vp);
|
2004-10-22 08:47:20 +00:00
|
|
|
if ((tbp = gbincore(&vp->v_bufobj, start_lbn)) == NULL ||
|
2003-08-28 06:55:18 +00:00
|
|
|
(tbp->b_vflags & BV_BKGRDINPROG)) {
|
2003-03-04 21:35:28 +00:00
|
|
|
VI_UNLOCK(vp);
|
1995-12-11 04:58:34 +00:00
|
|
|
splx(s);
|
|
|
|
break;
|
|
|
|
}
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
1998-03-16 18:39:41 +00:00
|
|
|
/*
|
|
|
|
* If it IS in core, but has different
|
2001-01-19 05:31:07 +00:00
|
|
|
* characteristics, or is locked (which
|
|
|
|
* means it could be undergoing a background
|
|
|
|
* I/O or be in a weird state), then don't
|
|
|
|
* cluster with it.
|
1998-03-16 18:39:41 +00:00
|
|
|
*/
|
2003-03-04 21:35:28 +00:00
|
|
|
if (BUF_LOCK(tbp,
|
|
|
|
LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK,
|
|
|
|
VI_MTX(vp))) {
|
|
|
|
splx(s);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
1999-06-26 02:47:16 +00:00
|
|
|
if ((tbp->b_flags & (B_VMIO | B_CLUSTEROK |
|
|
|
|
B_INVAL | B_DELWRI | B_NEEDCOMMIT))
|
2003-03-04 21:35:28 +00:00
|
|
|
!= (B_DELWRI | B_CLUSTEROK |
|
1999-06-26 02:47:16 +00:00
|
|
|
(bp->b_flags & (B_VMIO | B_NEEDCOMMIT))) ||
|
2003-03-04 21:35:28 +00:00
|
|
|
tbp->b_wcred != bp->b_wcred) {
|
2003-03-17 18:38:49 +00:00
|
|
|
BUF_UNLOCK(tbp);
|
1995-12-11 04:58:34 +00:00
|
|
|
splx(s);
|
|
|
|
break;
|
|
|
|
}
|
1994-05-24 10:09:53 +00:00
|
|
|
|
1998-03-16 18:39:41 +00:00
|
|
|
/*
|
|
|
|
* Check that the combined cluster
|
|
|
|
* would make sense with regard to pages
|
|
|
|
* and would not be too large
|
|
|
|
*/
|
1995-12-11 04:58:34 +00:00
|
|
|
if ((tbp->b_bcount != size) ||
|
1998-03-16 18:39:41 +00:00
|
|
|
((bp->b_blkno + (dbsize * i)) !=
|
In kern_physio.c fix tsleep priority messup.
In vfs_bio.c, remove b_generation count usage,
remove redundant reassignbuf,
remove redundant spl(s),
manage page PG_ZERO flags more correctly,
utilize in invalid value for b_offset until it
is properly initialized. Add asserts
for #ifdef DIAGNOSTIC, when b_offset is
improperly used.
when a process is not performing I/O, and just waiting
on a buffer generally, make the sleep priority
low.
only check page validity in getblk for B_VMIO buffers.
In vfs_cluster, add b_offset asserts, correct pointer calculation
for clustered reads. Improve readability of certain parts of
the code. Remove redundant spl(s).
In vfs_subr, correct usage of vfs_bio_awrite (From Andrew Gallatin
<gallatin@cs.duke.edu>). More vtruncbuf problems fixed.
1998-03-19 22:48:16 +00:00
|
|
|
tbp->b_blkno) ||
|
1998-03-16 18:39:41 +00:00
|
|
|
((tbp->b_npages + bp->b_npages) >
|
1999-09-29 20:05:33 +00:00
|
|
|
(vp->v_mount->mnt_iosize_max / PAGE_SIZE))) {
|
1999-06-26 02:47:16 +00:00
|
|
|
BUF_UNLOCK(tbp);
|
1995-12-11 04:58:34 +00:00
|
|
|
splx(s);
|
|
|
|
break;
|
|
|
|
}
|
1998-03-16 18:39:41 +00:00
|
|
|
/*
|
|
|
|
* Ok, it's passed all the tests,
|
|
|
|
* so remove it from the free list
|
|
|
|
* and mark it busy. We will use it.
|
|
|
|
*/
|
1995-12-11 04:58:34 +00:00
|
|
|
bremfree(tbp);
|
|
|
|
tbp->b_flags &= ~B_DONE;
|
1995-11-19 19:54:31 +00:00
|
|
|
splx(s);
|
1998-03-16 18:39:41 +00:00
|
|
|
} /* end of code for non-first buffers only */
|
1998-03-08 09:59:44 +00:00
|
|
|
/* check for latent dependencies to be handled */
|
2003-05-28 13:22:10 +00:00
|
|
|
if ((LIST_FIRST(&tbp->b_dep)) != NULL) {
|
|
|
|
tbp->b_iocmd = BIO_WRITE;
|
2000-06-16 08:48:51 +00:00
|
|
|
buf_start(tbp);
|
2003-05-28 13:22:10 +00:00
|
|
|
}
|
1998-03-16 18:39:41 +00:00
|
|
|
/*
|
|
|
|
* If the IO is via the VM then we do some
|
2001-10-21 06:12:06 +00:00
|
|
|
* special VM hackery (yuck). Since the buffer's
|
|
|
|
* block size may not be page-aligned it is possible
|
|
|
|
* for a page to be shared between two buffers. We
|
|
|
|
* have to get rid of the duplication when building
|
|
|
|
* the cluster.
|
1998-03-16 18:39:41 +00:00
|
|
|
*/
|
1996-01-19 04:00:31 +00:00
|
|
|
if (tbp->b_flags & B_VMIO) {
|
1998-01-31 11:56:53 +00:00
|
|
|
vm_page_t m;
|
|
|
|
|
1998-03-16 18:39:41 +00:00
|
|
|
if (i != 0) { /* if not first buffer */
|
1998-01-31 11:56:53 +00:00
|
|
|
for (j = 0; j < tbp->b_npages; j += 1) {
|
|
|
|
m = tbp->b_pages[j];
|
1999-08-31 14:18:32 +00:00
|
|
|
if (m->flags & PG_BUSY) {
|
|
|
|
bqrelse(tbp);
|
1998-01-31 11:56:53 +00:00
|
|
|
goto finishcluster;
|
1999-08-31 14:18:32 +00:00
|
|
|
}
|
1998-01-31 11:56:53 +00:00
|
|
|
}
|
|
|
|
}
|
2003-04-20 07:29:50 +00:00
|
|
|
if (tbp->b_object != NULL)
|
|
|
|
VM_OBJECT_LOCK(tbp->b_object);
|
2002-07-31 07:27:08 +00:00
|
|
|
vm_page_lock_queues();
|
1996-01-19 04:00:31 +00:00
|
|
|
for (j = 0; j < tbp->b_npages; j += 1) {
|
|
|
|
m = tbp->b_pages[j];
|
1998-09-04 08:06:57 +00:00
|
|
|
vm_page_io_start(m);
|
1998-08-24 08:39:39 +00:00
|
|
|
vm_object_pip_add(m->object, 1);
|
1996-01-19 04:00:31 +00:00
|
|
|
if ((bp->b_npages == 0) ||
|
1998-03-16 18:39:41 +00:00
|
|
|
(bp->b_pages[bp->b_npages - 1] != m)) {
|
1996-01-19 04:00:31 +00:00
|
|
|
bp->b_pages[bp->b_npages] = m;
|
|
|
|
bp->b_npages++;
|
|
|
|
}
|
1995-12-11 04:58:34 +00:00
|
|
|
}
|
2002-07-31 07:27:08 +00:00
|
|
|
vm_page_unlock_queues();
|
2003-04-20 07:29:50 +00:00
|
|
|
if (tbp->b_object != NULL)
|
|
|
|
VM_OBJECT_UNLOCK(tbp->b_object);
|
1995-11-19 19:54:31 +00:00
|
|
|
}
|
1995-12-11 04:58:34 +00:00
|
|
|
bp->b_bcount += size;
|
|
|
|
bp->b_bufsize += size;
|
1995-03-02 19:36:50 +00:00
|
|
|
|
1998-08-13 08:09:08 +00:00
|
|
|
s = splbio();
|
1999-03-12 02:24:58 +00:00
|
|
|
bundirty(tbp);
|
2000-04-02 15:24:56 +00:00
|
|
|
tbp->b_flags &= ~B_DONE;
|
|
|
|
tbp->b_ioflags &= ~BIO_ERROR;
|
1995-12-11 04:58:34 +00:00
|
|
|
tbp->b_flags |= B_ASYNC;
|
2000-03-20 10:44:49 +00:00
|
|
|
tbp->b_iocmd = BIO_WRITE;
|
2004-07-25 21:24:23 +00:00
|
|
|
reassignbuf(tbp); /* put on clean list */
|
2004-10-22 08:47:20 +00:00
|
|
|
bufobj_wref(tbp->b_bufobj);
|
1998-08-13 08:09:08 +00:00
|
|
|
splx(s);
|
1999-06-29 05:59:47 +00:00
|
|
|
BUF_KERNPROC(tbp);
|
1995-12-11 04:58:34 +00:00
|
|
|
TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head,
|
|
|
|
tbp, b_cluster.cluster_entry);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
1998-01-31 11:56:53 +00:00
|
|
|
finishcluster:
|
1995-12-11 04:58:34 +00:00
|
|
|
pmap_qenter(trunc_page((vm_offset_t) bp->b_data),
|
|
|
|
(vm_page_t *) bp->b_pages, bp->b_npages);
|
1996-11-30 22:41:49 +00:00
|
|
|
if (bp->b_bufsize > bp->b_kvasize)
|
1998-07-11 10:45:45 +00:00
|
|
|
panic(
|
|
|
|
"cluster_wbuild: b_bufsize(%ld) > b_kvasize(%d)\n",
|
|
|
|
bp->b_bufsize, bp->b_kvasize);
|
1996-11-30 22:41:49 +00:00
|
|
|
bp->b_kvasize = bp->b_bufsize;
|
1995-12-11 04:58:34 +00:00
|
|
|
totalwritten += bp->b_bufsize;
|
1996-07-27 18:49:18 +00:00
|
|
|
bp->b_dirtyoff = 0;
|
|
|
|
bp->b_dirtyend = bp->b_bufsize;
|
1995-12-11 04:58:34 +00:00
|
|
|
bawrite(bp);
|
1994-05-24 10:09:53 +00:00
|
|
|
|
1995-12-11 04:58:34 +00:00
|
|
|
len -= i;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
1995-12-11 04:58:34 +00:00
|
|
|
return totalwritten;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Collect together all the buffers in a cluster.
|
|
|
|
* Plus add one additional buffer.
|
|
|
|
*/
|
1995-12-22 16:06:46 +00:00
|
|
|
static struct cluster_save *
|
1994-05-24 10:09:53 +00:00
|
|
|
cluster_collectbufs(vp, last_bp)
|
|
|
|
struct vnode *vp;
|
|
|
|
struct buf *last_bp;
|
|
|
|
{
|
|
|
|
struct cluster_save *buflist;
|
1998-11-17 00:31:12 +00:00
|
|
|
struct buf *bp;
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
daddr_t lbn;
|
1994-05-24 10:09:53 +00:00
|
|
|
int i, len;
|
|
|
|
|
|
|
|
len = vp->v_lastw - vp->v_cstart + 1;
|
|
|
|
buflist = malloc(sizeof(struct buf *) * (len + 1) + sizeof(*buflist),
|
2003-02-19 05:47:46 +00:00
|
|
|
M_SEGMENT, M_WAITOK);
|
1994-05-24 10:09:53 +00:00
|
|
|
buflist->bs_nchildren = 0;
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
buflist->bs_children = (struct buf **) (buflist + 1);
|
1998-11-17 00:31:12 +00:00
|
|
|
for (lbn = vp->v_cstart, i = 0; i < len; lbn++, i++) {
|
|
|
|
(void) bread(vp, lbn, last_bp->b_bcount, NOCRED, &bp);
|
|
|
|
buflist->bs_children[i] = bp;
|
|
|
|
if (bp->b_blkno == bp->b_lblkno)
|
|
|
|
VOP_BMAP(bp->b_vp, bp->b_lblkno, NULL, &bp->b_blkno,
|
|
|
|
NULL, NULL);
|
|
|
|
}
|
1998-12-05 06:12:14 +00:00
|
|
|
buflist->bs_children[i] = bp = last_bp;
|
|
|
|
if (bp->b_blkno == bp->b_lblkno)
|
|
|
|
VOP_BMAP(bp->b_vp, bp->b_lblkno, NULL, &bp->b_blkno,
|
|
|
|
NULL, NULL);
|
1994-05-24 10:09:53 +00:00
|
|
|
buflist->bs_nchildren = i + 1;
|
|
|
|
return (buflist);
|
|
|
|
}
|