Incorporate the enhanced ffs_dirpref() by Grigoriy Orlov, as found in
FreeBSD (three commits; the initial work, man page updates, and a fix
to ffs_reload()), with the following differences:
- Be consistent between newfs(8) and tunefs(8) as to the options which
set and control the tuning parameters for this work (avgfilesize & avgfpdir)
- Use u_int16_t instead of u_int8_t to keep track of the number of
contiguous directories (suggested by Chuck Silvers)
- Work within our FFS_EI framework
- Ensure that fs->fs_maxclusters and fs->fs_contigdirs don't point to
the same area of memory
The new algorithm has a marked performance increase, especially when
performing tasks such as untarring pkgsrc.tar.gz, etc.
The original FreeBSD commit messages are attached:
=====
mckusick 2001/04/10 01:39:00 PDT
Directory layout preference improvements from Grigoriy Orlov <gluk@ptci.ru>.
His description of the problem and solution follow. My own tests show
speedups on typical filesystem intensive workloads of 5% to 12% which
is very impressive considering the small amount of code change involved.
------
One day I noticed that some file operations run much faster on
small file systems then on big ones. I've looked at the ffs
algorithms, thought about them, and redesigned the dirpref algorithm.
First I want to describe the results of my tests. These results are old
and I have improved the algorithm after these tests were done. Nevertheless
they show how big the perfomance speedup may be. I have done two file/directory
intensive tests on a two OpenBSD systems with old and new dirpref algorithm.
The first test is "tar -xzf ports.tar.gz", the second is "rm -rf ports".
The ports.tar.gz file is the ports collection from the OpenBSD 2.8 release.
It contains 6596 directories and 13868 files. The test systems are:
1. Celeron-450, 128Mb, two IDE drives, the system at wd0, file system for
test is at wd1. Size of test file system is 8 Gb, number of cg=991,
size of cg is 8m, block size = 8k, fragment size = 1k OpenBSD-current
from Dec 2000 with BUFCACHEPERCENT=35
2. PIII-600, 128Mb, two IBM DTLA-307045 IDE drives at i815e, the system
at wd0, file system for test is at wd1. Size of test file system is 40 Gb,
number of cg=5324, size of cg is 8m, block size = 8k, fragment size = 1k
OpenBSD-current from Dec 2000 with BUFCACHEPERCENT=50
You can get more info about the test systems and methods at:
http://www.ptci.ru/gluk/dirpref/old/dirpref.html
Test Results
tar -xzf ports.tar.gz rm -rf ports
mode old dirpref new dirpref speedup old dirprefnew dirpref speedup
First system
normal 667 472 1.41 477 331 1.44
async 285 144 1.98 130 14 9.29
sync 768 616 1.25 477 334 1.43
softdep 413 252 1.64 241 38 6.34
Second system
normal 329 81 4.06 263.5 93.5 2.81
async 302 25.7 11.75 112 2.26 49.56
sync 281 57.0 4.93 263 90.5 2.9
softdep 341 40.6 8.4 284 4.76 59.66
"old dirpref" and "new dirpref" columns give a test time in seconds.
speedup - speed increasement in times, ie. old dirpref / new dirpref.
------
Algorithm description
The old dirpref algorithm is described in comments:
/*
* Find a cylinder to place a directory.
*
* The policy implemented by this algorithm is to select from
* among those cylinder groups with above the average number of
* free inodes, the one with the smallest number of directories.
*/
A new directory is allocated in a different cylinder groups than its
parent directory resulting in a directory tree that is spreaded across
all the cylinder groups. This spreading out results in a non-optimal
access to the directories and files. When we have a small filesystem
it is not a problem but when the filesystem is big then perfomance
degradation becomes very apparent.
What I mean by a big file system ?
1. A big filesystem is a filesystem which occupy 20-30 or more percent
of total drive space, i.e. first and last cylinder are physically
located relatively far from each other.
2. It has a relatively large number of cylinder groups, for example
more cylinder groups than 50% of the buffers in the buffer cache.
The first results in long access times, while the second results in
many buffers being used by metadata operations. Such operations use
cylinder group blocks and on-disk inode blocks. The cylinder group
block (fs->fs_cblkno) contains struct cg, inode and block bit maps.
It is 2k in size for the default filesystem parameters. If new and
parent directories are located in different cylinder groups then the
system performs more input/output operations and uses more buffers.
On filesystems with many cylinder groups, lots of cache buffers are
used for metadata operations.
My solution for this problem is very simple. I allocate many directories
in one cylinder group. I also do some things, so that the new allocation
method does not cause excessive fragmentation and all directory inodes
will not be located at a location far from its file's inodes and data.
The algorithm is:
/*
* Find a cylinder group to place a directory.
*
* The policy implemented by this algorithm is to allocate a
* directory inode in the same cylinder group as its parent
* directory, but also to reserve space for its files inodes
* and data. Restrict the number of directories which may be
* allocated one after another in the same cylinder group
* without intervening allocation of files.
*
* If we allocate a first level directory then force allocation
* in another cylinder group.
*/
My early versions of dirpref give me a good results for a wide range of
file operations and different filesystem capacities except one case:
those applications that create their entire directory structure first
and only later fill this structure with files.
My solution for such and similar cases is to limit a number of
directories which may be created one after another in the same cylinder
group without intervening file creations. For this purpose, I allocate
an array of counters at mount time. This array is linked to the superblock
fs->fs_contigdirs[cg]. Each time a directory is created the counter
increases and each time a file is created the counter decreases. A 60Gb
filesystem with 8mb/cg requires 10kb of memory for the counters array.
The maxcontigdirs is a maximum number of directories which may be created
without an intervening file creation. I found in my tests that the best
performance occurs when I restrict the number of directories in one cylinder
group such that all its files may be located in the same cylinder group.
There may be some deterioration in performance if all the file inodes
are in the same cylinder group as its containing directory, but their
data partially resides in a different cylinder group. The maxcontigdirs
value is calculated to try to prevent this condition. Since there is
no way to know how many files and directories will be allocated later
I added two optimization parameters in superblock/tunefs. They are:
int32_t fs_avgfilesize; /* expected average file size */
int32_t fs_avgfpdir; /* expected # of files per directory */
These parameters have reasonable defaults but may be tweeked for special
uses of a filesystem. They are only necessary in rare cases like better
tuning a filesystem being used to store a squid cache.
I have been using this algorithm for about 3 months. I have done
a lot of testing on filesystems with different capacities, average
filesize, average number of files per directory, and so on. I think
this algorithm has no negative impact on filesystem perfomance. It
works better than the default one in all cases. The new dirpref
will greatly improve untarring/removing/coping of big directories,
decrease load on cvs servers and much more. The new dirpref doesn't
speedup a compilation process, but also doesn't slow it down.
Obtained from: Grigoriy Orlov <gluk@ptci.ru>
=====
=====
iedowse 2001/04/23 17:37:17 PDT
Pre-dirpref versions of fsck may zero out the new superblock fields
fs_contigdirs, fs_avgfilesize and fs_avgfpdir. This could cause
panics if these fields were zeroed while a filesystem was mounted
read-only, and then remounted read-write.
Add code to ffs_reload() which copies the fs_contigdirs pointer
from the previous superblock, and reinitialises fs_avgf* if necessary.
Reviewed by: mckusick
=====
=====
nik 2001/04/10 03:36:44 PDT
Add information about the new options to newfs and tunefs which set the
expected average file size and number of files per directory. Could do
with some fleshing out.
=====
2001-09-06 06:16:00 +04:00
|
|
|
/* $NetBSD: newfs.c,v 1.47 2001/09/06 02:16:01 lukem Exp $ */
|
1995-03-18 17:54:19 +03:00
|
|
|
|
1993-03-21 12:45:37 +03:00
|
|
|
/*
|
1994-06-08 23:27:32 +04:00
|
|
|
* Copyright (c) 1983, 1989, 1993, 1994
|
|
|
|
* The Regents of the University of California. All rights reserved.
|
1993-03-21 12:45:37 +03:00
|
|
|
*
|
|
|
|
* 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.
|
|
|
|
* 3. All advertising materials mentioning features or use of this software
|
|
|
|
* must display the following acknowledgement:
|
|
|
|
* This product includes software developed by the University of
|
|
|
|
* California, Berkeley and its contributors.
|
|
|
|
* 4. Neither the name of the University nor the names of its contributors
|
|
|
|
* may be used to endorse or promote products derived from this software
|
|
|
|
* without specific prior written permission.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
|
|
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
|
|
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
|
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
|
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
|
|
* SUCH DAMAGE.
|
|
|
|
*/
|
|
|
|
|
1997-07-01 02:20:30 +04:00
|
|
|
#include <sys/cdefs.h>
|
1993-03-21 12:45:37 +03:00
|
|
|
#ifndef lint
|
1997-07-01 02:20:30 +04:00
|
|
|
__COPYRIGHT("@(#) Copyright (c) 1983, 1989, 1993, 1994\n\
|
|
|
|
The Regents of the University of California. All rights reserved.\n");
|
1993-03-21 12:45:37 +03:00
|
|
|
#endif /* not lint */
|
|
|
|
|
|
|
|
#ifndef lint
|
1995-03-18 17:54:19 +03:00
|
|
|
#if 0
|
1997-09-16 18:05:39 +04:00
|
|
|
static char sccsid[] = "@(#)newfs.c 8.13 (Berkeley) 5/1/95";
|
1995-03-18 17:54:19 +03:00
|
|
|
#else
|
Incorporate the enhanced ffs_dirpref() by Grigoriy Orlov, as found in
FreeBSD (three commits; the initial work, man page updates, and a fix
to ffs_reload()), with the following differences:
- Be consistent between newfs(8) and tunefs(8) as to the options which
set and control the tuning parameters for this work (avgfilesize & avgfpdir)
- Use u_int16_t instead of u_int8_t to keep track of the number of
contiguous directories (suggested by Chuck Silvers)
- Work within our FFS_EI framework
- Ensure that fs->fs_maxclusters and fs->fs_contigdirs don't point to
the same area of memory
The new algorithm has a marked performance increase, especially when
performing tasks such as untarring pkgsrc.tar.gz, etc.
The original FreeBSD commit messages are attached:
=====
mckusick 2001/04/10 01:39:00 PDT
Directory layout preference improvements from Grigoriy Orlov <gluk@ptci.ru>.
His description of the problem and solution follow. My own tests show
speedups on typical filesystem intensive workloads of 5% to 12% which
is very impressive considering the small amount of code change involved.
------
One day I noticed that some file operations run much faster on
small file systems then on big ones. I've looked at the ffs
algorithms, thought about them, and redesigned the dirpref algorithm.
First I want to describe the results of my tests. These results are old
and I have improved the algorithm after these tests were done. Nevertheless
they show how big the perfomance speedup may be. I have done two file/directory
intensive tests on a two OpenBSD systems with old and new dirpref algorithm.
The first test is "tar -xzf ports.tar.gz", the second is "rm -rf ports".
The ports.tar.gz file is the ports collection from the OpenBSD 2.8 release.
It contains 6596 directories and 13868 files. The test systems are:
1. Celeron-450, 128Mb, two IDE drives, the system at wd0, file system for
test is at wd1. Size of test file system is 8 Gb, number of cg=991,
size of cg is 8m, block size = 8k, fragment size = 1k OpenBSD-current
from Dec 2000 with BUFCACHEPERCENT=35
2. PIII-600, 128Mb, two IBM DTLA-307045 IDE drives at i815e, the system
at wd0, file system for test is at wd1. Size of test file system is 40 Gb,
number of cg=5324, size of cg is 8m, block size = 8k, fragment size = 1k
OpenBSD-current from Dec 2000 with BUFCACHEPERCENT=50
You can get more info about the test systems and methods at:
http://www.ptci.ru/gluk/dirpref/old/dirpref.html
Test Results
tar -xzf ports.tar.gz rm -rf ports
mode old dirpref new dirpref speedup old dirprefnew dirpref speedup
First system
normal 667 472 1.41 477 331 1.44
async 285 144 1.98 130 14 9.29
sync 768 616 1.25 477 334 1.43
softdep 413 252 1.64 241 38 6.34
Second system
normal 329 81 4.06 263.5 93.5 2.81
async 302 25.7 11.75 112 2.26 49.56
sync 281 57.0 4.93 263 90.5 2.9
softdep 341 40.6 8.4 284 4.76 59.66
"old dirpref" and "new dirpref" columns give a test time in seconds.
speedup - speed increasement in times, ie. old dirpref / new dirpref.
------
Algorithm description
The old dirpref algorithm is described in comments:
/*
* Find a cylinder to place a directory.
*
* The policy implemented by this algorithm is to select from
* among those cylinder groups with above the average number of
* free inodes, the one with the smallest number of directories.
*/
A new directory is allocated in a different cylinder groups than its
parent directory resulting in a directory tree that is spreaded across
all the cylinder groups. This spreading out results in a non-optimal
access to the directories and files. When we have a small filesystem
it is not a problem but when the filesystem is big then perfomance
degradation becomes very apparent.
What I mean by a big file system ?
1. A big filesystem is a filesystem which occupy 20-30 or more percent
of total drive space, i.e. first and last cylinder are physically
located relatively far from each other.
2. It has a relatively large number of cylinder groups, for example
more cylinder groups than 50% of the buffers in the buffer cache.
The first results in long access times, while the second results in
many buffers being used by metadata operations. Such operations use
cylinder group blocks and on-disk inode blocks. The cylinder group
block (fs->fs_cblkno) contains struct cg, inode and block bit maps.
It is 2k in size for the default filesystem parameters. If new and
parent directories are located in different cylinder groups then the
system performs more input/output operations and uses more buffers.
On filesystems with many cylinder groups, lots of cache buffers are
used for metadata operations.
My solution for this problem is very simple. I allocate many directories
in one cylinder group. I also do some things, so that the new allocation
method does not cause excessive fragmentation and all directory inodes
will not be located at a location far from its file's inodes and data.
The algorithm is:
/*
* Find a cylinder group to place a directory.
*
* The policy implemented by this algorithm is to allocate a
* directory inode in the same cylinder group as its parent
* directory, but also to reserve space for its files inodes
* and data. Restrict the number of directories which may be
* allocated one after another in the same cylinder group
* without intervening allocation of files.
*
* If we allocate a first level directory then force allocation
* in another cylinder group.
*/
My early versions of dirpref give me a good results for a wide range of
file operations and different filesystem capacities except one case:
those applications that create their entire directory structure first
and only later fill this structure with files.
My solution for such and similar cases is to limit a number of
directories which may be created one after another in the same cylinder
group without intervening file creations. For this purpose, I allocate
an array of counters at mount time. This array is linked to the superblock
fs->fs_contigdirs[cg]. Each time a directory is created the counter
increases and each time a file is created the counter decreases. A 60Gb
filesystem with 8mb/cg requires 10kb of memory for the counters array.
The maxcontigdirs is a maximum number of directories which may be created
without an intervening file creation. I found in my tests that the best
performance occurs when I restrict the number of directories in one cylinder
group such that all its files may be located in the same cylinder group.
There may be some deterioration in performance if all the file inodes
are in the same cylinder group as its containing directory, but their
data partially resides in a different cylinder group. The maxcontigdirs
value is calculated to try to prevent this condition. Since there is
no way to know how many files and directories will be allocated later
I added two optimization parameters in superblock/tunefs. They are:
int32_t fs_avgfilesize; /* expected average file size */
int32_t fs_avgfpdir; /* expected # of files per directory */
These parameters have reasonable defaults but may be tweeked for special
uses of a filesystem. They are only necessary in rare cases like better
tuning a filesystem being used to store a squid cache.
I have been using this algorithm for about 3 months. I have done
a lot of testing on filesystems with different capacities, average
filesize, average number of files per directory, and so on. I think
this algorithm has no negative impact on filesystem perfomance. It
works better than the default one in all cases. The new dirpref
will greatly improve untarring/removing/coping of big directories,
decrease load on cvs servers and much more. The new dirpref doesn't
speedup a compilation process, but also doesn't slow it down.
Obtained from: Grigoriy Orlov <gluk@ptci.ru>
=====
=====
iedowse 2001/04/23 17:37:17 PDT
Pre-dirpref versions of fsck may zero out the new superblock fields
fs_contigdirs, fs_avgfilesize and fs_avgfpdir. This could cause
panics if these fields were zeroed while a filesystem was mounted
read-only, and then remounted read-write.
Add code to ffs_reload() which copies the fs_contigdirs pointer
from the previous superblock, and reinitialises fs_avgf* if necessary.
Reviewed by: mckusick
=====
=====
nik 2001/04/10 03:36:44 PDT
Add information about the new options to newfs and tunefs which set the
expected average file size and number of files per directory. Could do
with some fleshing out.
=====
2001-09-06 06:16:00 +04:00
|
|
|
__RCSID("$NetBSD: newfs.c,v 1.47 2001/09/06 02:16:01 lukem Exp $");
|
1995-03-18 17:54:19 +03:00
|
|
|
#endif
|
1993-03-21 12:45:37 +03:00
|
|
|
#endif /* not lint */
|
|
|
|
|
|
|
|
/*
|
|
|
|
* newfs: friendly front end to mkfs
|
|
|
|
*/
|
|
|
|
#include <sys/param.h>
|
|
|
|
#include <sys/stat.h>
|
|
|
|
#include <sys/ioctl.h>
|
|
|
|
#include <sys/disklabel.h>
|
|
|
|
#include <sys/file.h>
|
|
|
|
#include <sys/mount.h>
|
1995-06-28 06:21:02 +04:00
|
|
|
#include <sys/sysctl.h>
|
1997-11-01 21:25:46 +03:00
|
|
|
#include <sys/wait.h>
|
1993-03-21 12:45:37 +03:00
|
|
|
|
1994-06-08 23:27:32 +04:00
|
|
|
#include <ufs/ufs/dir.h>
|
1997-09-16 18:05:39 +04:00
|
|
|
#include <ufs/ufs/dinode.h>
|
1998-03-01 05:20:01 +03:00
|
|
|
#include <ufs/ufs/ufsmount.h>
|
1994-06-08 23:27:32 +04:00
|
|
|
#include <ufs/ffs/fs.h>
|
|
|
|
|
|
|
|
#include <ctype.h>
|
1999-01-19 22:34:50 +03:00
|
|
|
#include <disktab.h>
|
1993-03-21 12:45:37 +03:00
|
|
|
#include <errno.h>
|
1994-06-08 23:27:32 +04:00
|
|
|
#include <paths.h>
|
1993-03-21 12:45:37 +03:00
|
|
|
#include <stdio.h>
|
|
|
|
#include <stdlib.h>
|
1994-06-08 23:27:32 +04:00
|
|
|
#include <string.h>
|
|
|
|
#include <syslog.h>
|
|
|
|
#include <unistd.h>
|
1997-11-01 21:25:46 +03:00
|
|
|
#include <signal.h>
|
1997-07-01 02:20:30 +04:00
|
|
|
#include <err.h>
|
1996-05-16 11:13:01 +04:00
|
|
|
#include <util.h>
|
1994-06-08 23:27:32 +04:00
|
|
|
|
|
|
|
#include "mntopts.h"
|
1997-07-01 02:20:30 +04:00
|
|
|
#include "dkcksum.h"
|
|
|
|
#include "extern.h"
|
1994-06-08 23:27:32 +04:00
|
|
|
|
|
|
|
struct mntopt mopts[] = {
|
|
|
|
MOPT_STDOPTS,
|
|
|
|
MOPT_ASYNC,
|
1996-10-24 02:46:16 +04:00
|
|
|
MOPT_UPDATE,
|
1997-01-30 12:56:10 +03:00
|
|
|
MOPT_NOATIME,
|
1994-06-08 23:27:32 +04:00
|
|
|
{ NULL },
|
|
|
|
};
|
|
|
|
|
2000-12-01 14:52:54 +03:00
|
|
|
static struct disklabel *getdisklabel(char *, int);
|
|
|
|
static void rewritelabel(char *, int, struct disklabel *);
|
2001-07-29 13:55:22 +04:00
|
|
|
static int strsuftoi(const char *, const char *, int, int);
|
2000-12-01 14:52:54 +03:00
|
|
|
static void usage(void);
|
|
|
|
int main(int, char *[]);
|
1993-03-21 12:45:37 +03:00
|
|
|
|
|
|
|
#define COMPAT /* allow non-labeled disks */
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The following two constants set the default block and fragment sizes.
|
|
|
|
* Both constants must be a power of 2 and meet the following constraints:
|
|
|
|
* MINBSIZE <= DESBLKSIZE <= MAXBSIZE
|
|
|
|
* sectorsize <= DESFRAGSIZE <= DESBLKSIZE
|
|
|
|
* DESBLKSIZE / DESFRAGSIZE <= 8
|
|
|
|
*/
|
|
|
|
#define DFL_FRAGSIZE 1024
|
|
|
|
#define DFL_BLKSIZE 8192
|
|
|
|
|
2001-07-29 13:55:22 +04:00
|
|
|
/*
|
|
|
|
* Default sector size.
|
|
|
|
*/
|
|
|
|
#define DFL_SECSIZE 512
|
|
|
|
|
1993-03-21 12:45:37 +03:00
|
|
|
/*
|
|
|
|
* Cylinder groups may have up to many cylinders. The actual
|
|
|
|
* number used depends upon how much information can be stored
|
|
|
|
* on a single cylinder. The default is to use 16 cylinders
|
|
|
|
* per group.
|
|
|
|
*/
|
|
|
|
#define DESCPG 16 /* desired fs_cpg */
|
|
|
|
|
|
|
|
/*
|
|
|
|
* ROTDELAY gives the minimum number of milliseconds to initiate
|
|
|
|
* another disk transfer on the same cylinder. It is used in
|
|
|
|
* determining the rotationally optimal layout for disk blocks
|
1994-12-01 21:46:37 +03:00
|
|
|
* within a file; the default of fs_rotdelay is 0ms.
|
1993-03-21 12:45:37 +03:00
|
|
|
*/
|
1994-12-01 21:46:37 +03:00
|
|
|
#define ROTDELAY 0
|
1993-03-21 12:45:37 +03:00
|
|
|
|
|
|
|
/*
|
|
|
|
* MAXBLKPG determines the maximum number of data blocks which are
|
|
|
|
* placed in a single cylinder group. The default is one indirect
|
|
|
|
* block worth of data blocks.
|
|
|
|
*/
|
|
|
|
#define MAXBLKPG(bsize) ((bsize) / sizeof(daddr_t))
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Each file system has a number of inodes statically allocated.
|
|
|
|
* We allocate one inode slot per NFPI fragments, expecting this
|
|
|
|
* to be far more than we will ever need.
|
|
|
|
*/
|
|
|
|
#define NFPI 4
|
|
|
|
|
|
|
|
/*
|
|
|
|
* For each cylinder we keep track of the availability of blocks at different
|
|
|
|
* rotational positions, so that we can lay out the data to be picked
|
|
|
|
* up with minimum rotational latency. NRPOS is the default number of
|
|
|
|
* rotational positions that we distinguish. With NRPOS of 8 the resolution
|
1994-12-01 21:46:37 +03:00
|
|
|
* of our summary information is 2ms for a typical 3600 rpm drive. Caching
|
|
|
|
* and zoning pretty much defeats rotational optimization, so we now use a
|
|
|
|
* default of 1.
|
1993-03-21 12:45:37 +03:00
|
|
|
*/
|
1994-12-01 21:46:37 +03:00
|
|
|
#define NRPOS 1 /* number distinct rotational positions */
|
1993-03-21 12:45:37 +03:00
|
|
|
|
|
|
|
|
|
|
|
int mfs; /* run as the memory based filesystem */
|
|
|
|
int Nflag; /* run without writing file system */
|
1994-06-08 23:27:32 +04:00
|
|
|
int Oflag; /* format as an 4.3BSD file system */
|
1993-03-21 12:45:37 +03:00
|
|
|
int fssize; /* file system size */
|
|
|
|
int ntracks; /* # tracks/cylinder */
|
|
|
|
int nsectors; /* # sectors/track */
|
|
|
|
int nphyssectors; /* # sectors/track including spares */
|
|
|
|
int secpercyl; /* sectors per cylinder */
|
|
|
|
int trackspares = -1; /* spare sectors per track */
|
|
|
|
int cylspares = -1; /* spare sectors per cylinder */
|
|
|
|
int sectorsize; /* bytes/sector */
|
|
|
|
int rpm; /* revolutions/minute of drive */
|
|
|
|
int interleave; /* hardware sector interleave */
|
|
|
|
int trackskew = -1; /* sector 0 skew, per track */
|
|
|
|
int fsize = 0; /* fragment size */
|
|
|
|
int bsize = 0; /* block size */
|
|
|
|
int cpg = DESCPG; /* cylinders/cylinder group */
|
|
|
|
int cpgflg; /* cylinders/cylinder group flag was given */
|
|
|
|
int minfree = MINFREE; /* free space threshold */
|
|
|
|
int opt = DEFAULTOPT; /* optimization preference (space or time) */
|
|
|
|
int density; /* number of bytes per inode */
|
1998-10-22 12:55:42 +04:00
|
|
|
int maxcontig = 0; /* max contiguous blocks to allocate */
|
1993-03-21 12:45:37 +03:00
|
|
|
int rotdelay = ROTDELAY; /* rotational delay between blocks */
|
|
|
|
int maxbpg; /* maximum blocks per file in a cyl group */
|
|
|
|
int nrpos = NRPOS; /* # of distinguished rotational positions */
|
Incorporate the enhanced ffs_dirpref() by Grigoriy Orlov, as found in
FreeBSD (three commits; the initial work, man page updates, and a fix
to ffs_reload()), with the following differences:
- Be consistent between newfs(8) and tunefs(8) as to the options which
set and control the tuning parameters for this work (avgfilesize & avgfpdir)
- Use u_int16_t instead of u_int8_t to keep track of the number of
contiguous directories (suggested by Chuck Silvers)
- Work within our FFS_EI framework
- Ensure that fs->fs_maxclusters and fs->fs_contigdirs don't point to
the same area of memory
The new algorithm has a marked performance increase, especially when
performing tasks such as untarring pkgsrc.tar.gz, etc.
The original FreeBSD commit messages are attached:
=====
mckusick 2001/04/10 01:39:00 PDT
Directory layout preference improvements from Grigoriy Orlov <gluk@ptci.ru>.
His description of the problem and solution follow. My own tests show
speedups on typical filesystem intensive workloads of 5% to 12% which
is very impressive considering the small amount of code change involved.
------
One day I noticed that some file operations run much faster on
small file systems then on big ones. I've looked at the ffs
algorithms, thought about them, and redesigned the dirpref algorithm.
First I want to describe the results of my tests. These results are old
and I have improved the algorithm after these tests were done. Nevertheless
they show how big the perfomance speedup may be. I have done two file/directory
intensive tests on a two OpenBSD systems with old and new dirpref algorithm.
The first test is "tar -xzf ports.tar.gz", the second is "rm -rf ports".
The ports.tar.gz file is the ports collection from the OpenBSD 2.8 release.
It contains 6596 directories and 13868 files. The test systems are:
1. Celeron-450, 128Mb, two IDE drives, the system at wd0, file system for
test is at wd1. Size of test file system is 8 Gb, number of cg=991,
size of cg is 8m, block size = 8k, fragment size = 1k OpenBSD-current
from Dec 2000 with BUFCACHEPERCENT=35
2. PIII-600, 128Mb, two IBM DTLA-307045 IDE drives at i815e, the system
at wd0, file system for test is at wd1. Size of test file system is 40 Gb,
number of cg=5324, size of cg is 8m, block size = 8k, fragment size = 1k
OpenBSD-current from Dec 2000 with BUFCACHEPERCENT=50
You can get more info about the test systems and methods at:
http://www.ptci.ru/gluk/dirpref/old/dirpref.html
Test Results
tar -xzf ports.tar.gz rm -rf ports
mode old dirpref new dirpref speedup old dirprefnew dirpref speedup
First system
normal 667 472 1.41 477 331 1.44
async 285 144 1.98 130 14 9.29
sync 768 616 1.25 477 334 1.43
softdep 413 252 1.64 241 38 6.34
Second system
normal 329 81 4.06 263.5 93.5 2.81
async 302 25.7 11.75 112 2.26 49.56
sync 281 57.0 4.93 263 90.5 2.9
softdep 341 40.6 8.4 284 4.76 59.66
"old dirpref" and "new dirpref" columns give a test time in seconds.
speedup - speed increasement in times, ie. old dirpref / new dirpref.
------
Algorithm description
The old dirpref algorithm is described in comments:
/*
* Find a cylinder to place a directory.
*
* The policy implemented by this algorithm is to select from
* among those cylinder groups with above the average number of
* free inodes, the one with the smallest number of directories.
*/
A new directory is allocated in a different cylinder groups than its
parent directory resulting in a directory tree that is spreaded across
all the cylinder groups. This spreading out results in a non-optimal
access to the directories and files. When we have a small filesystem
it is not a problem but when the filesystem is big then perfomance
degradation becomes very apparent.
What I mean by a big file system ?
1. A big filesystem is a filesystem which occupy 20-30 or more percent
of total drive space, i.e. first and last cylinder are physically
located relatively far from each other.
2. It has a relatively large number of cylinder groups, for example
more cylinder groups than 50% of the buffers in the buffer cache.
The first results in long access times, while the second results in
many buffers being used by metadata operations. Such operations use
cylinder group blocks and on-disk inode blocks. The cylinder group
block (fs->fs_cblkno) contains struct cg, inode and block bit maps.
It is 2k in size for the default filesystem parameters. If new and
parent directories are located in different cylinder groups then the
system performs more input/output operations and uses more buffers.
On filesystems with many cylinder groups, lots of cache buffers are
used for metadata operations.
My solution for this problem is very simple. I allocate many directories
in one cylinder group. I also do some things, so that the new allocation
method does not cause excessive fragmentation and all directory inodes
will not be located at a location far from its file's inodes and data.
The algorithm is:
/*
* Find a cylinder group to place a directory.
*
* The policy implemented by this algorithm is to allocate a
* directory inode in the same cylinder group as its parent
* directory, but also to reserve space for its files inodes
* and data. Restrict the number of directories which may be
* allocated one after another in the same cylinder group
* without intervening allocation of files.
*
* If we allocate a first level directory then force allocation
* in another cylinder group.
*/
My early versions of dirpref give me a good results for a wide range of
file operations and different filesystem capacities except one case:
those applications that create their entire directory structure first
and only later fill this structure with files.
My solution for such and similar cases is to limit a number of
directories which may be created one after another in the same cylinder
group without intervening file creations. For this purpose, I allocate
an array of counters at mount time. This array is linked to the superblock
fs->fs_contigdirs[cg]. Each time a directory is created the counter
increases and each time a file is created the counter decreases. A 60Gb
filesystem with 8mb/cg requires 10kb of memory for the counters array.
The maxcontigdirs is a maximum number of directories which may be created
without an intervening file creation. I found in my tests that the best
performance occurs when I restrict the number of directories in one cylinder
group such that all its files may be located in the same cylinder group.
There may be some deterioration in performance if all the file inodes
are in the same cylinder group as its containing directory, but their
data partially resides in a different cylinder group. The maxcontigdirs
value is calculated to try to prevent this condition. Since there is
no way to know how many files and directories will be allocated later
I added two optimization parameters in superblock/tunefs. They are:
int32_t fs_avgfilesize; /* expected average file size */
int32_t fs_avgfpdir; /* expected # of files per directory */
These parameters have reasonable defaults but may be tweeked for special
uses of a filesystem. They are only necessary in rare cases like better
tuning a filesystem being used to store a squid cache.
I have been using this algorithm for about 3 months. I have done
a lot of testing on filesystems with different capacities, average
filesize, average number of files per directory, and so on. I think
this algorithm has no negative impact on filesystem perfomance. It
works better than the default one in all cases. The new dirpref
will greatly improve untarring/removing/coping of big directories,
decrease load on cvs servers and much more. The new dirpref doesn't
speedup a compilation process, but also doesn't slow it down.
Obtained from: Grigoriy Orlov <gluk@ptci.ru>
=====
=====
iedowse 2001/04/23 17:37:17 PDT
Pre-dirpref versions of fsck may zero out the new superblock fields
fs_contigdirs, fs_avgfilesize and fs_avgfpdir. This could cause
panics if these fields were zeroed while a filesystem was mounted
read-only, and then remounted read-write.
Add code to ffs_reload() which copies the fs_contigdirs pointer
from the previous superblock, and reinitialises fs_avgf* if necessary.
Reviewed by: mckusick
=====
=====
nik 2001/04/10 03:36:44 PDT
Add information about the new options to newfs and tunefs which set the
expected average file size and number of files per directory. Could do
with some fleshing out.
=====
2001-09-06 06:16:00 +04:00
|
|
|
int avgfilesize = AVFILESIZ;/* expected average file size */
|
|
|
|
int avgfpdir = AFPDIR; /* expected number of files per directory */
|
1993-03-21 12:45:37 +03:00
|
|
|
int bbsize = BBSIZE; /* boot block size */
|
|
|
|
int sbsize = SBSIZE; /* superblock size */
|
1994-06-08 23:27:32 +04:00
|
|
|
int mntflags = MNT_ASYNC; /* flags to be passed to mount */
|
1993-03-21 12:45:37 +03:00
|
|
|
u_long memleft; /* virtual memory available */
|
|
|
|
caddr_t membase; /* start address of memory based filesystem */
|
2001-07-29 13:55:22 +04:00
|
|
|
int needswap; /* Filesystem not in native byte order */
|
1993-03-21 12:45:37 +03:00
|
|
|
#ifdef COMPAT
|
|
|
|
char *disktype;
|
|
|
|
int unlabeled;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
char device[MAXPATHLEN];
|
|
|
|
|
1994-06-08 23:27:32 +04:00
|
|
|
int
|
2000-12-01 14:52:54 +03:00
|
|
|
main(int argc, char *argv[])
|
1993-03-21 12:45:37 +03:00
|
|
|
{
|
1997-07-01 02:20:30 +04:00
|
|
|
struct partition *pp;
|
|
|
|
struct disklabel *lp;
|
1994-12-18 08:09:39 +03:00
|
|
|
struct disklabel mfsfakelabel;
|
1993-03-21 12:45:37 +03:00
|
|
|
struct partition oldpartition;
|
|
|
|
struct stat st;
|
1994-06-08 23:27:32 +04:00
|
|
|
struct statfs *mp;
|
2001-07-29 13:55:22 +04:00
|
|
|
int ch, fsi, fso, len, maxpartitions, n, Fflag, Zflag;
|
|
|
|
char *cp, *endp, *s1, *s2, *special;
|
|
|
|
const char *opstring;
|
|
|
|
long long llsize;
|
1997-11-01 21:25:46 +03:00
|
|
|
#ifdef MFS
|
|
|
|
char mountfromname[100];
|
|
|
|
pid_t pid, res;
|
|
|
|
struct statfs sf;
|
|
|
|
int status;
|
|
|
|
#endif
|
1993-03-21 12:45:37 +03:00
|
|
|
|
2001-07-29 13:55:22 +04:00
|
|
|
cp = NULL;
|
|
|
|
fsi = fso = -1;
|
|
|
|
Fflag = Zflag = 0;
|
2001-02-20 01:48:57 +03:00
|
|
|
if (strstr(getprogname(), "mfs")) {
|
1993-03-21 12:45:37 +03:00
|
|
|
mfs = 1;
|
|
|
|
Nflag++;
|
|
|
|
}
|
|
|
|
|
1995-06-28 06:21:02 +04:00
|
|
|
maxpartitions = getmaxpartitions();
|
|
|
|
if (maxpartitions > 26)
|
1997-07-01 02:20:30 +04:00
|
|
|
errx(1, "insane maxpartitions value %d", maxpartitions);
|
1995-06-28 06:21:02 +04:00
|
|
|
|
1994-06-08 23:27:32 +04:00
|
|
|
opstring = mfs ?
|
Incorporate the enhanced ffs_dirpref() by Grigoriy Orlov, as found in
FreeBSD (three commits; the initial work, man page updates, and a fix
to ffs_reload()), with the following differences:
- Be consistent between newfs(8) and tunefs(8) as to the options which
set and control the tuning parameters for this work (avgfilesize & avgfpdir)
- Use u_int16_t instead of u_int8_t to keep track of the number of
contiguous directories (suggested by Chuck Silvers)
- Work within our FFS_EI framework
- Ensure that fs->fs_maxclusters and fs->fs_contigdirs don't point to
the same area of memory
The new algorithm has a marked performance increase, especially when
performing tasks such as untarring pkgsrc.tar.gz, etc.
The original FreeBSD commit messages are attached:
=====
mckusick 2001/04/10 01:39:00 PDT
Directory layout preference improvements from Grigoriy Orlov <gluk@ptci.ru>.
His description of the problem and solution follow. My own tests show
speedups on typical filesystem intensive workloads of 5% to 12% which
is very impressive considering the small amount of code change involved.
------
One day I noticed that some file operations run much faster on
small file systems then on big ones. I've looked at the ffs
algorithms, thought about them, and redesigned the dirpref algorithm.
First I want to describe the results of my tests. These results are old
and I have improved the algorithm after these tests were done. Nevertheless
they show how big the perfomance speedup may be. I have done two file/directory
intensive tests on a two OpenBSD systems with old and new dirpref algorithm.
The first test is "tar -xzf ports.tar.gz", the second is "rm -rf ports".
The ports.tar.gz file is the ports collection from the OpenBSD 2.8 release.
It contains 6596 directories and 13868 files. The test systems are:
1. Celeron-450, 128Mb, two IDE drives, the system at wd0, file system for
test is at wd1. Size of test file system is 8 Gb, number of cg=991,
size of cg is 8m, block size = 8k, fragment size = 1k OpenBSD-current
from Dec 2000 with BUFCACHEPERCENT=35
2. PIII-600, 128Mb, two IBM DTLA-307045 IDE drives at i815e, the system
at wd0, file system for test is at wd1. Size of test file system is 40 Gb,
number of cg=5324, size of cg is 8m, block size = 8k, fragment size = 1k
OpenBSD-current from Dec 2000 with BUFCACHEPERCENT=50
You can get more info about the test systems and methods at:
http://www.ptci.ru/gluk/dirpref/old/dirpref.html
Test Results
tar -xzf ports.tar.gz rm -rf ports
mode old dirpref new dirpref speedup old dirprefnew dirpref speedup
First system
normal 667 472 1.41 477 331 1.44
async 285 144 1.98 130 14 9.29
sync 768 616 1.25 477 334 1.43
softdep 413 252 1.64 241 38 6.34
Second system
normal 329 81 4.06 263.5 93.5 2.81
async 302 25.7 11.75 112 2.26 49.56
sync 281 57.0 4.93 263 90.5 2.9
softdep 341 40.6 8.4 284 4.76 59.66
"old dirpref" and "new dirpref" columns give a test time in seconds.
speedup - speed increasement in times, ie. old dirpref / new dirpref.
------
Algorithm description
The old dirpref algorithm is described in comments:
/*
* Find a cylinder to place a directory.
*
* The policy implemented by this algorithm is to select from
* among those cylinder groups with above the average number of
* free inodes, the one with the smallest number of directories.
*/
A new directory is allocated in a different cylinder groups than its
parent directory resulting in a directory tree that is spreaded across
all the cylinder groups. This spreading out results in a non-optimal
access to the directories and files. When we have a small filesystem
it is not a problem but when the filesystem is big then perfomance
degradation becomes very apparent.
What I mean by a big file system ?
1. A big filesystem is a filesystem which occupy 20-30 or more percent
of total drive space, i.e. first and last cylinder are physically
located relatively far from each other.
2. It has a relatively large number of cylinder groups, for example
more cylinder groups than 50% of the buffers in the buffer cache.
The first results in long access times, while the second results in
many buffers being used by metadata operations. Such operations use
cylinder group blocks and on-disk inode blocks. The cylinder group
block (fs->fs_cblkno) contains struct cg, inode and block bit maps.
It is 2k in size for the default filesystem parameters. If new and
parent directories are located in different cylinder groups then the
system performs more input/output operations and uses more buffers.
On filesystems with many cylinder groups, lots of cache buffers are
used for metadata operations.
My solution for this problem is very simple. I allocate many directories
in one cylinder group. I also do some things, so that the new allocation
method does not cause excessive fragmentation and all directory inodes
will not be located at a location far from its file's inodes and data.
The algorithm is:
/*
* Find a cylinder group to place a directory.
*
* The policy implemented by this algorithm is to allocate a
* directory inode in the same cylinder group as its parent
* directory, but also to reserve space for its files inodes
* and data. Restrict the number of directories which may be
* allocated one after another in the same cylinder group
* without intervening allocation of files.
*
* If we allocate a first level directory then force allocation
* in another cylinder group.
*/
My early versions of dirpref give me a good results for a wide range of
file operations and different filesystem capacities except one case:
those applications that create their entire directory structure first
and only later fill this structure with files.
My solution for such and similar cases is to limit a number of
directories which may be created one after another in the same cylinder
group without intervening file creations. For this purpose, I allocate
an array of counters at mount time. This array is linked to the superblock
fs->fs_contigdirs[cg]. Each time a directory is created the counter
increases and each time a file is created the counter decreases. A 60Gb
filesystem with 8mb/cg requires 10kb of memory for the counters array.
The maxcontigdirs is a maximum number of directories which may be created
without an intervening file creation. I found in my tests that the best
performance occurs when I restrict the number of directories in one cylinder
group such that all its files may be located in the same cylinder group.
There may be some deterioration in performance if all the file inodes
are in the same cylinder group as its containing directory, but their
data partially resides in a different cylinder group. The maxcontigdirs
value is calculated to try to prevent this condition. Since there is
no way to know how many files and directories will be allocated later
I added two optimization parameters in superblock/tunefs. They are:
int32_t fs_avgfilesize; /* expected average file size */
int32_t fs_avgfpdir; /* expected # of files per directory */
These parameters have reasonable defaults but may be tweeked for special
uses of a filesystem. They are only necessary in rare cases like better
tuning a filesystem being used to store a squid cache.
I have been using this algorithm for about 3 months. I have done
a lot of testing on filesystems with different capacities, average
filesize, average number of files per directory, and so on. I think
this algorithm has no negative impact on filesystem perfomance. It
works better than the default one in all cases. The new dirpref
will greatly improve untarring/removing/coping of big directories,
decrease load on cvs servers and much more. The new dirpref doesn't
speedup a compilation process, but also doesn't slow it down.
Obtained from: Grigoriy Orlov <gluk@ptci.ru>
=====
=====
iedowse 2001/04/23 17:37:17 PDT
Pre-dirpref versions of fsck may zero out the new superblock fields
fs_contigdirs, fs_avgfilesize and fs_avgfpdir. This could cause
panics if these fields were zeroed while a filesystem was mounted
read-only, and then remounted read-write.
Add code to ffs_reload() which copies the fs_contigdirs pointer
from the previous superblock, and reinitialises fs_avgf* if necessary.
Reviewed by: mckusick
=====
=====
nik 2001/04/10 03:36:44 PDT
Add information about the new options to newfs and tunefs which set the
expected average file size and number of files per directory. Could do
with some fleshing out.
=====
2001-09-06 06:16:00 +04:00
|
|
|
"NT:a:b:c:d:e:f:g:h:i:m:o:s:" :
|
|
|
|
"B:FNOS:T:Za:b:c:d:e:f:g:h:i:k:l:m:n:o:p:r:s:t:u:x:";
|
1997-09-15 10:23:03 +04:00
|
|
|
while ((ch = getopt(argc, argv, opstring)) != -1)
|
1994-06-08 23:27:32 +04:00
|
|
|
switch (ch) {
|
1998-03-18 20:10:15 +03:00
|
|
|
case 'B':
|
|
|
|
if (strcmp(optarg, "be") == 0) {
|
|
|
|
#if BYTE_ORDER == LITTLE_ENDIAN
|
|
|
|
needswap = 1;
|
|
|
|
#endif
|
|
|
|
} else if (strcmp(optarg, "le") == 0) {
|
|
|
|
#if BYTE_ORDER == BIG_ENDIAN
|
|
|
|
needswap = 1;
|
|
|
|
#endif
|
|
|
|
} else
|
|
|
|
usage();
|
|
|
|
break;
|
2001-07-29 13:55:22 +04:00
|
|
|
case 'F':
|
|
|
|
Fflag = 1;
|
|
|
|
break;
|
1993-03-21 12:45:37 +03:00
|
|
|
case 'N':
|
1994-06-08 23:27:32 +04:00
|
|
|
Nflag = 1;
|
|
|
|
break;
|
|
|
|
case 'O':
|
|
|
|
Oflag = 1;
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 'S':
|
2001-07-29 13:55:22 +04:00
|
|
|
sectorsize = strsuftoi("sector size",
|
|
|
|
optarg, 1, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
#ifdef COMPAT
|
|
|
|
case 'T':
|
|
|
|
disktype = optarg;
|
|
|
|
break;
|
|
|
|
#endif
|
2001-07-29 13:55:22 +04:00
|
|
|
case 'Z':
|
|
|
|
Zflag = 1;
|
|
|
|
break;
|
1993-03-21 12:45:37 +03:00
|
|
|
case 'a':
|
2001-07-29 13:55:22 +04:00
|
|
|
maxcontig = strsuftoi("maximum contiguous blocks",
|
|
|
|
optarg, 1, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 'b':
|
2001-07-29 13:55:22 +04:00
|
|
|
bsize = strsuftoi("block size",
|
|
|
|
optarg, MINBSIZE, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 'c':
|
2001-07-29 13:55:22 +04:00
|
|
|
cpg = strsuftoi("cylinders per group",
|
|
|
|
optarg, 1, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
cpgflg++;
|
|
|
|
break;
|
|
|
|
case 'd':
|
2001-07-29 13:55:22 +04:00
|
|
|
rotdelay = strsuftoi("rotational delay",
|
|
|
|
optarg, 0, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 'e':
|
2001-07-29 13:55:22 +04:00
|
|
|
maxbpg = strsuftoi(
|
|
|
|
"blocks per file in a cylinder group",
|
|
|
|
optarg, 1, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 'f':
|
2001-07-29 13:55:22 +04:00
|
|
|
fsize = strsuftoi("fragment size",
|
|
|
|
optarg, 1, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
Incorporate the enhanced ffs_dirpref() by Grigoriy Orlov, as found in
FreeBSD (three commits; the initial work, man page updates, and a fix
to ffs_reload()), with the following differences:
- Be consistent between newfs(8) and tunefs(8) as to the options which
set and control the tuning parameters for this work (avgfilesize & avgfpdir)
- Use u_int16_t instead of u_int8_t to keep track of the number of
contiguous directories (suggested by Chuck Silvers)
- Work within our FFS_EI framework
- Ensure that fs->fs_maxclusters and fs->fs_contigdirs don't point to
the same area of memory
The new algorithm has a marked performance increase, especially when
performing tasks such as untarring pkgsrc.tar.gz, etc.
The original FreeBSD commit messages are attached:
=====
mckusick 2001/04/10 01:39:00 PDT
Directory layout preference improvements from Grigoriy Orlov <gluk@ptci.ru>.
His description of the problem and solution follow. My own tests show
speedups on typical filesystem intensive workloads of 5% to 12% which
is very impressive considering the small amount of code change involved.
------
One day I noticed that some file operations run much faster on
small file systems then on big ones. I've looked at the ffs
algorithms, thought about them, and redesigned the dirpref algorithm.
First I want to describe the results of my tests. These results are old
and I have improved the algorithm after these tests were done. Nevertheless
they show how big the perfomance speedup may be. I have done two file/directory
intensive tests on a two OpenBSD systems with old and new dirpref algorithm.
The first test is "tar -xzf ports.tar.gz", the second is "rm -rf ports".
The ports.tar.gz file is the ports collection from the OpenBSD 2.8 release.
It contains 6596 directories and 13868 files. The test systems are:
1. Celeron-450, 128Mb, two IDE drives, the system at wd0, file system for
test is at wd1. Size of test file system is 8 Gb, number of cg=991,
size of cg is 8m, block size = 8k, fragment size = 1k OpenBSD-current
from Dec 2000 with BUFCACHEPERCENT=35
2. PIII-600, 128Mb, two IBM DTLA-307045 IDE drives at i815e, the system
at wd0, file system for test is at wd1. Size of test file system is 40 Gb,
number of cg=5324, size of cg is 8m, block size = 8k, fragment size = 1k
OpenBSD-current from Dec 2000 with BUFCACHEPERCENT=50
You can get more info about the test systems and methods at:
http://www.ptci.ru/gluk/dirpref/old/dirpref.html
Test Results
tar -xzf ports.tar.gz rm -rf ports
mode old dirpref new dirpref speedup old dirprefnew dirpref speedup
First system
normal 667 472 1.41 477 331 1.44
async 285 144 1.98 130 14 9.29
sync 768 616 1.25 477 334 1.43
softdep 413 252 1.64 241 38 6.34
Second system
normal 329 81 4.06 263.5 93.5 2.81
async 302 25.7 11.75 112 2.26 49.56
sync 281 57.0 4.93 263 90.5 2.9
softdep 341 40.6 8.4 284 4.76 59.66
"old dirpref" and "new dirpref" columns give a test time in seconds.
speedup - speed increasement in times, ie. old dirpref / new dirpref.
------
Algorithm description
The old dirpref algorithm is described in comments:
/*
* Find a cylinder to place a directory.
*
* The policy implemented by this algorithm is to select from
* among those cylinder groups with above the average number of
* free inodes, the one with the smallest number of directories.
*/
A new directory is allocated in a different cylinder groups than its
parent directory resulting in a directory tree that is spreaded across
all the cylinder groups. This spreading out results in a non-optimal
access to the directories and files. When we have a small filesystem
it is not a problem but when the filesystem is big then perfomance
degradation becomes very apparent.
What I mean by a big file system ?
1. A big filesystem is a filesystem which occupy 20-30 or more percent
of total drive space, i.e. first and last cylinder are physically
located relatively far from each other.
2. It has a relatively large number of cylinder groups, for example
more cylinder groups than 50% of the buffers in the buffer cache.
The first results in long access times, while the second results in
many buffers being used by metadata operations. Such operations use
cylinder group blocks and on-disk inode blocks. The cylinder group
block (fs->fs_cblkno) contains struct cg, inode and block bit maps.
It is 2k in size for the default filesystem parameters. If new and
parent directories are located in different cylinder groups then the
system performs more input/output operations and uses more buffers.
On filesystems with many cylinder groups, lots of cache buffers are
used for metadata operations.
My solution for this problem is very simple. I allocate many directories
in one cylinder group. I also do some things, so that the new allocation
method does not cause excessive fragmentation and all directory inodes
will not be located at a location far from its file's inodes and data.
The algorithm is:
/*
* Find a cylinder group to place a directory.
*
* The policy implemented by this algorithm is to allocate a
* directory inode in the same cylinder group as its parent
* directory, but also to reserve space for its files inodes
* and data. Restrict the number of directories which may be
* allocated one after another in the same cylinder group
* without intervening allocation of files.
*
* If we allocate a first level directory then force allocation
* in another cylinder group.
*/
My early versions of dirpref give me a good results for a wide range of
file operations and different filesystem capacities except one case:
those applications that create their entire directory structure first
and only later fill this structure with files.
My solution for such and similar cases is to limit a number of
directories which may be created one after another in the same cylinder
group without intervening file creations. For this purpose, I allocate
an array of counters at mount time. This array is linked to the superblock
fs->fs_contigdirs[cg]. Each time a directory is created the counter
increases and each time a file is created the counter decreases. A 60Gb
filesystem with 8mb/cg requires 10kb of memory for the counters array.
The maxcontigdirs is a maximum number of directories which may be created
without an intervening file creation. I found in my tests that the best
performance occurs when I restrict the number of directories in one cylinder
group such that all its files may be located in the same cylinder group.
There may be some deterioration in performance if all the file inodes
are in the same cylinder group as its containing directory, but their
data partially resides in a different cylinder group. The maxcontigdirs
value is calculated to try to prevent this condition. Since there is
no way to know how many files and directories will be allocated later
I added two optimization parameters in superblock/tunefs. They are:
int32_t fs_avgfilesize; /* expected average file size */
int32_t fs_avgfpdir; /* expected # of files per directory */
These parameters have reasonable defaults but may be tweeked for special
uses of a filesystem. They are only necessary in rare cases like better
tuning a filesystem being used to store a squid cache.
I have been using this algorithm for about 3 months. I have done
a lot of testing on filesystems with different capacities, average
filesize, average number of files per directory, and so on. I think
this algorithm has no negative impact on filesystem perfomance. It
works better than the default one in all cases. The new dirpref
will greatly improve untarring/removing/coping of big directories,
decrease load on cvs servers and much more. The new dirpref doesn't
speedup a compilation process, but also doesn't slow it down.
Obtained from: Grigoriy Orlov <gluk@ptci.ru>
=====
=====
iedowse 2001/04/23 17:37:17 PDT
Pre-dirpref versions of fsck may zero out the new superblock fields
fs_contigdirs, fs_avgfilesize and fs_avgfpdir. This could cause
panics if these fields were zeroed while a filesystem was mounted
read-only, and then remounted read-write.
Add code to ffs_reload() which copies the fs_contigdirs pointer
from the previous superblock, and reinitialises fs_avgf* if necessary.
Reviewed by: mckusick
=====
=====
nik 2001/04/10 03:36:44 PDT
Add information about the new options to newfs and tunefs which set the
expected average file size and number of files per directory. Could do
with some fleshing out.
=====
2001-09-06 06:16:00 +04:00
|
|
|
case 'g':
|
|
|
|
avgfilesize = strsuftoi("average file size",
|
|
|
|
optarg, 1, INT_MAX);
|
|
|
|
break;
|
|
|
|
case 'h':
|
|
|
|
avgfpdir = strsuftoi("expected files per directory",
|
|
|
|
optarg, 1, INT_MAX);
|
|
|
|
break;
|
1993-03-21 12:45:37 +03:00
|
|
|
case 'i':
|
2001-07-29 13:55:22 +04:00
|
|
|
density = strsuftoi("bytes per inode",
|
|
|
|
optarg, 1, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 'k':
|
2001-07-29 13:55:22 +04:00
|
|
|
trackskew = strsuftoi("track skew",
|
|
|
|
optarg, 0, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 'l':
|
2001-07-29 13:55:22 +04:00
|
|
|
interleave = strsuftoi("interleave",
|
|
|
|
optarg, 1, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 'm':
|
2001-07-29 13:55:22 +04:00
|
|
|
minfree = strsuftoi("free space %",
|
|
|
|
optarg, 0, 99);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 'n':
|
2001-07-29 13:55:22 +04:00
|
|
|
nrpos = strsuftoi("rotational layout count",
|
|
|
|
optarg, 1, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 'o':
|
1994-06-08 23:27:32 +04:00
|
|
|
if (mfs)
|
1997-09-16 18:05:39 +04:00
|
|
|
getmntopts(optarg, mopts, &mntflags, 0);
|
1994-06-08 23:27:32 +04:00
|
|
|
else {
|
|
|
|
if (strcmp(optarg, "space") == 0)
|
|
|
|
opt = FS_OPTSPACE;
|
|
|
|
else if (strcmp(optarg, "time") == 0)
|
|
|
|
opt = FS_OPTTIME;
|
|
|
|
else
|
1997-07-01 02:20:30 +04:00
|
|
|
errx(1, "%s %s",
|
|
|
|
"unknown optimization preference: ",
|
|
|
|
"use `space' or `time'.");
|
1994-06-08 23:27:32 +04:00
|
|
|
}
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 'p':
|
2001-07-29 13:55:22 +04:00
|
|
|
trackspares = strsuftoi("spare sectors per track",
|
|
|
|
optarg, 0, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 'r':
|
2001-07-29 13:55:22 +04:00
|
|
|
rpm = strsuftoi("revolutions per minute",
|
|
|
|
optarg, 1, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 's':
|
2001-07-29 13:55:22 +04:00
|
|
|
llsize = strtoll(optarg, &endp, 10);
|
|
|
|
if (endp[0] != '\0' && endp[1] != '\0')
|
|
|
|
llsize = -1;
|
|
|
|
else {
|
|
|
|
int ssiz;
|
|
|
|
|
|
|
|
ssiz = (sectorsize ? sectorsize : DFL_SECSIZE);
|
|
|
|
switch (tolower((unsigned char)endp[0])) {
|
|
|
|
case 'b':
|
|
|
|
llsize /= ssiz;
|
|
|
|
break;
|
|
|
|
case 'k':
|
|
|
|
llsize *= 1024 / ssiz;
|
|
|
|
break;
|
|
|
|
case 'm':
|
|
|
|
llsize *= 1024 * 1024 / ssiz;
|
|
|
|
break;
|
|
|
|
case 'g':
|
|
|
|
llsize *= 1024 * 1024 * 1024 / ssiz;
|
|
|
|
break;
|
|
|
|
case '\0':
|
|
|
|
case 's':
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
llsize = -1;
|
2000-12-01 15:48:09 +03:00
|
|
|
}
|
|
|
|
}
|
2001-07-29 13:55:22 +04:00
|
|
|
if (llsize > INT_MAX)
|
|
|
|
errx(1, "file system size `%s' is too large.",
|
|
|
|
optarg);
|
|
|
|
if (llsize <= 0)
|
|
|
|
errx(1,
|
|
|
|
"`%s' is not a valid number for file system size.",
|
|
|
|
optarg);
|
|
|
|
fssize = (int)llsize;
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 't':
|
2001-07-29 13:55:22 +04:00
|
|
|
ntracks = strsuftoi("total tracks",
|
|
|
|
optarg, 1, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 'u':
|
2001-07-29 13:55:22 +04:00
|
|
|
nsectors = strsuftoi("sectors per track",
|
|
|
|
optarg, 1, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case 'x':
|
2001-07-29 13:55:22 +04:00
|
|
|
cylspares = strsuftoi("spare sectors per cylinder",
|
|
|
|
optarg, 0, INT_MAX);
|
1993-03-21 12:45:37 +03:00
|
|
|
break;
|
|
|
|
case '?':
|
|
|
|
default:
|
|
|
|
usage();
|
|
|
|
}
|
|
|
|
argc -= optind;
|
|
|
|
argv += optind;
|
|
|
|
|
|
|
|
if (argc != 2 && (mfs || argc != 1))
|
|
|
|
usage();
|
|
|
|
|
|
|
|
special = argv[0];
|
2001-07-29 13:55:22 +04:00
|
|
|
if ((mfs && !strcmp(special, "swap")) || Fflag) {
|
1994-12-18 08:09:39 +03:00
|
|
|
/*
|
2001-07-29 13:55:22 +04:00
|
|
|
* it's an MFS mounted on "swap" or a file system image;
|
|
|
|
* fake up a label. XXX
|
1994-12-18 08:09:39 +03:00
|
|
|
*/
|
2001-07-29 13:55:22 +04:00
|
|
|
if (!sectorsize)
|
|
|
|
sectorsize = DFL_SECSIZE;
|
|
|
|
|
|
|
|
if (Fflag && (stat(special, &st) != -1 && !S_ISREG(st.st_mode)))
|
|
|
|
errx(1, "%s is not a regular file", special);
|
|
|
|
if (Fflag && !Nflag) { /* creating image in a regular file */
|
2001-07-30 11:45:08 +04:00
|
|
|
if (fssize == 0)
|
|
|
|
errx(1, "need to specify size when using -F");
|
2001-07-29 13:55:22 +04:00
|
|
|
fso = open(special, O_RDWR | O_CREAT | O_TRUNC, 0777);
|
|
|
|
if (fso == -1)
|
|
|
|
err(1, "can't open file %s", special);
|
2001-07-30 11:45:08 +04:00
|
|
|
if ((fsi = dup(fso)) == -1)
|
|
|
|
err(1, "can't dup(2) image fd");
|
2001-07-29 13:55:22 +04:00
|
|
|
if (ftruncate(fso, (off_t)fssize * sectorsize) == -1)
|
|
|
|
err(1, "can't resize %s to %d",
|
|
|
|
special, fssize);
|
|
|
|
|
|
|
|
if (Zflag) { /* pre-zero the file */
|
|
|
|
char *buf;
|
2001-08-08 11:34:53 +04:00
|
|
|
int bufsize, i;
|
|
|
|
off_t bufrem;
|
|
|
|
struct statfs sfs;
|
2001-07-29 13:55:22 +04:00
|
|
|
|
2001-08-08 11:34:53 +04:00
|
|
|
if (fstatfs(fso, &sfs) == -1) {
|
|
|
|
warn("can't fstatfs `%s'", special);
|
|
|
|
bufsize = 8192;
|
|
|
|
} else
|
|
|
|
bufsize = sfs.f_iosize;
|
|
|
|
|
|
|
|
if ((buf = calloc(1, bufsize)) == NULL)
|
2001-07-29 13:55:22 +04:00
|
|
|
err(1, "can't malloc buffer of %d",
|
2001-08-08 11:34:53 +04:00
|
|
|
bufsize);
|
|
|
|
bufrem = fssize * sectorsize;
|
2001-07-29 13:55:22 +04:00
|
|
|
printf(
|
2001-08-08 11:34:53 +04:00
|
|
|
"Creating file system image in `%s', size %lld bytes, in %d byte chunks.\n",
|
|
|
|
special, (long long)bufrem, bufsize);
|
|
|
|
while (bufrem > 0) {
|
|
|
|
i = write(fso, buf,
|
|
|
|
MIN(bufsize, bufrem));
|
|
|
|
if (i == -1)
|
|
|
|
err(1, "writing image");
|
|
|
|
bufrem -= i;
|
2001-07-29 13:55:22 +04:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
}
|
1994-12-18 08:09:39 +03:00
|
|
|
|
|
|
|
memset(&mfsfakelabel, 0, sizeof(mfsfakelabel));
|
2001-07-29 13:55:22 +04:00
|
|
|
mfsfakelabel.d_secsize = sectorsize;
|
|
|
|
mfsfakelabel.d_nsectors = 64; /* these 3 add up to 16MB */
|
1994-12-18 08:09:39 +03:00
|
|
|
mfsfakelabel.d_ntracks = 16;
|
|
|
|
mfsfakelabel.d_ncylinders = 16;
|
2001-07-29 13:55:22 +04:00
|
|
|
mfsfakelabel.d_secpercyl =
|
|
|
|
mfsfakelabel.d_nsectors * mfsfakelabel.d_ntracks;
|
|
|
|
mfsfakelabel.d_secperunit =
|
|
|
|
mfsfakelabel.d_ncylinders * mfsfakelabel.d_secpercyl;
|
1994-12-18 08:09:39 +03:00
|
|
|
mfsfakelabel.d_rpm = 3600;
|
|
|
|
mfsfakelabel.d_interleave = 1;
|
|
|
|
mfsfakelabel.d_npartitions = 1;
|
2001-07-29 13:55:22 +04:00
|
|
|
mfsfakelabel.d_partitions[0].p_size = mfsfakelabel.d_secperunit;
|
1994-12-18 08:09:39 +03:00
|
|
|
mfsfakelabel.d_partitions[0].p_fsize = 1024;
|
|
|
|
mfsfakelabel.d_partitions[0].p_frag = 8;
|
|
|
|
mfsfakelabel.d_partitions[0].p_cpg = 16;
|
|
|
|
|
|
|
|
lp = &mfsfakelabel;
|
|
|
|
pp = &mfsfakelabel.d_partitions[0];
|
|
|
|
|
|
|
|
goto havelabel;
|
|
|
|
}
|
1994-09-23 18:26:58 +04:00
|
|
|
cp = strrchr(special, '/');
|
2001-07-29 13:55:22 +04:00
|
|
|
if (cp == NULL) {
|
1993-03-21 12:45:37 +03:00
|
|
|
/*
|
|
|
|
* No path prefix; try /dev/r%s then /dev/%s.
|
|
|
|
*/
|
2001-07-29 13:55:22 +04:00
|
|
|
(void)snprintf(device, sizeof(device), "%sr%s",
|
|
|
|
_PATH_DEV, special);
|
1993-03-21 12:45:37 +03:00
|
|
|
if (stat(device, &st) == -1)
|
1998-07-28 23:22:54 +04:00
|
|
|
(void)snprintf(device, sizeof(device), "%s%s",
|
|
|
|
_PATH_DEV, special);
|
1993-03-21 12:45:37 +03:00
|
|
|
special = device;
|
|
|
|
}
|
1994-06-08 23:27:32 +04:00
|
|
|
if (Nflag) {
|
|
|
|
fso = -1;
|
|
|
|
} else {
|
1993-03-21 12:45:37 +03:00
|
|
|
fso = open(special, O_WRONLY);
|
|
|
|
if (fso < 0)
|
1997-09-29 09:18:00 +04:00
|
|
|
err(1, "%s: open", special);
|
1994-06-08 23:27:32 +04:00
|
|
|
|
|
|
|
/* Bail if target special is mounted */
|
|
|
|
n = getmntinfo(&mp, MNT_NOWAIT);
|
|
|
|
if (n == 0)
|
1997-09-29 09:18:00 +04:00
|
|
|
err(1, "%s: getmntinfo", special);
|
1994-06-08 23:27:32 +04:00
|
|
|
|
|
|
|
len = sizeof(_PATH_DEV) - 1;
|
|
|
|
s1 = special;
|
|
|
|
if (strncmp(_PATH_DEV, s1, len) == 0)
|
|
|
|
s1 += len;
|
|
|
|
|
|
|
|
while (--n >= 0) {
|
|
|
|
s2 = mp->f_mntfromname;
|
|
|
|
if (strncmp(_PATH_DEV, s2, len) == 0) {
|
|
|
|
s2 += len - 1;
|
|
|
|
*s2 = 'r';
|
|
|
|
}
|
|
|
|
if (strcmp(s1, s2) == 0 || strcmp(s1, &s2[1]) == 0)
|
1997-07-01 02:20:30 +04:00
|
|
|
errx(1, "%s is mounted on %s",
|
1994-06-08 23:27:32 +04:00
|
|
|
special, mp->f_mntonname);
|
|
|
|
++mp;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (mfs && disktype != NULL) {
|
|
|
|
lp = (struct disklabel *)getdiskbyname(disktype);
|
|
|
|
if (lp == NULL)
|
1997-07-01 02:20:30 +04:00
|
|
|
errx(1, "%s: unknown disk type", disktype);
|
1994-06-08 23:27:32 +04:00
|
|
|
pp = &lp->d_partitions[1];
|
|
|
|
} else {
|
|
|
|
fsi = open(special, O_RDONLY);
|
|
|
|
if (fsi < 0)
|
1997-09-29 09:18:00 +04:00
|
|
|
err(1, "%s: open", special);
|
1994-06-08 23:27:32 +04:00
|
|
|
if (fstat(fsi, &st) < 0)
|
1997-09-29 09:18:00 +04:00
|
|
|
err(1, "%s: fstat", special);
|
1995-01-30 22:39:36 +03:00
|
|
|
if (!S_ISCHR(st.st_mode) && !mfs)
|
1997-07-01 02:20:30 +04:00
|
|
|
warnx("%s: not a character-special device", special);
|
1994-09-23 18:26:58 +04:00
|
|
|
cp = strchr(argv[0], '\0') - 1;
|
1997-07-01 02:20:30 +04:00
|
|
|
if (cp == 0 || ((*cp < 'a' || *cp > ('a' + maxpartitions - 1))
|
|
|
|
&& !isdigit(*cp)))
|
|
|
|
errx(1, "can't figure out file system partition");
|
1993-03-21 12:45:37 +03:00
|
|
|
#ifdef COMPAT
|
1994-06-08 23:27:32 +04:00
|
|
|
if (!mfs && disktype == NULL)
|
|
|
|
disktype = argv[1];
|
1993-03-21 12:45:37 +03:00
|
|
|
#endif
|
1994-06-08 23:27:32 +04:00
|
|
|
lp = getdisklabel(special, fsi);
|
|
|
|
if (isdigit(*cp))
|
|
|
|
pp = &lp->d_partitions[0];
|
|
|
|
else
|
|
|
|
pp = &lp->d_partitions[*cp - 'a'];
|
|
|
|
if (pp->p_size == 0)
|
1997-07-01 02:20:30 +04:00
|
|
|
errx(1, "`%c' partition is unavailable", *cp);
|
1994-06-08 23:27:32 +04:00
|
|
|
if (pp->p_fstype == FS_BOOT)
|
1997-07-01 02:20:30 +04:00
|
|
|
errx(1, "`%c' partition overlaps boot program", *cp);
|
1994-06-08 23:27:32 +04:00
|
|
|
}
|
2001-07-29 13:55:22 +04:00
|
|
|
|
|
|
|
havelabel:
|
1993-03-21 12:45:37 +03:00
|
|
|
if (fssize == 0)
|
|
|
|
fssize = pp->p_size;
|
2001-07-29 13:55:22 +04:00
|
|
|
if (fssize > pp->p_size && !mfs && !Fflag)
|
1997-07-01 02:20:30 +04:00
|
|
|
errx(1, "maximum file system size on the `%c' partition is %d",
|
|
|
|
*cp, pp->p_size);
|
1993-03-21 12:45:37 +03:00
|
|
|
if (rpm == 0) {
|
|
|
|
rpm = lp->d_rpm;
|
|
|
|
if (rpm <= 0)
|
|
|
|
rpm = 3600;
|
|
|
|
}
|
|
|
|
if (ntracks == 0) {
|
|
|
|
ntracks = lp->d_ntracks;
|
|
|
|
if (ntracks <= 0)
|
1997-07-01 02:20:30 +04:00
|
|
|
errx(1, "no default #tracks");
|
1993-03-21 12:45:37 +03:00
|
|
|
}
|
|
|
|
if (nsectors == 0) {
|
|
|
|
nsectors = lp->d_nsectors;
|
|
|
|
if (nsectors <= 0)
|
1997-07-01 02:20:30 +04:00
|
|
|
errx(1, "no default #sectors/track");
|
1993-03-21 12:45:37 +03:00
|
|
|
}
|
|
|
|
if (sectorsize == 0) {
|
|
|
|
sectorsize = lp->d_secsize;
|
|
|
|
if (sectorsize <= 0)
|
1997-07-01 02:20:30 +04:00
|
|
|
errx(1, "no default sector size");
|
1993-03-21 12:45:37 +03:00
|
|
|
}
|
|
|
|
if (trackskew == -1) {
|
|
|
|
trackskew = lp->d_trackskew;
|
|
|
|
if (trackskew < 0)
|
|
|
|
trackskew = 0;
|
|
|
|
}
|
|
|
|
if (interleave == 0) {
|
|
|
|
interleave = lp->d_interleave;
|
|
|
|
if (interleave <= 0)
|
|
|
|
interleave = 1;
|
|
|
|
}
|
|
|
|
if (fsize == 0) {
|
|
|
|
fsize = pp->p_fsize;
|
|
|
|
if (fsize <= 0)
|
|
|
|
fsize = MAX(DFL_FRAGSIZE, lp->d_secsize);
|
|
|
|
}
|
|
|
|
if (bsize == 0) {
|
|
|
|
bsize = pp->p_frag * pp->p_fsize;
|
|
|
|
if (bsize <= 0)
|
|
|
|
bsize = MIN(DFL_BLKSIZE, 8 * fsize);
|
|
|
|
}
|
1999-05-19 04:25:59 +04:00
|
|
|
if (cpgflg == 0) {
|
|
|
|
if (pp->p_cpg != 0)
|
|
|
|
cpg = pp->p_cpg;
|
|
|
|
}
|
1994-06-08 23:27:32 +04:00
|
|
|
/*
|
|
|
|
* Maxcontig sets the default for the maximum number of blocks
|
|
|
|
* that may be allocated sequentially. With filesystem clustering
|
|
|
|
* it is possible to allocate contiguous blocks up to the maximum
|
|
|
|
* transfer size permitted by the controller or buffering.
|
|
|
|
*/
|
|
|
|
if (maxcontig == 0)
|
1997-09-16 18:05:39 +04:00
|
|
|
maxcontig = MAX(1, MIN(MAXPHYS, MAXBSIZE) / bsize);
|
1993-03-21 12:45:37 +03:00
|
|
|
if (density == 0)
|
|
|
|
density = NFPI * fsize;
|
1994-02-06 11:19:56 +03:00
|
|
|
if (minfree < MINFREE && opt != FS_OPTSPACE) {
|
1997-07-01 02:20:30 +04:00
|
|
|
warnx("%s %s %d%%", "Warning: changing optimization to space",
|
|
|
|
"because minfree is less than", MINFREE);
|
1993-03-21 12:45:37 +03:00
|
|
|
opt = FS_OPTSPACE;
|
|
|
|
}
|
|
|
|
if (trackspares == -1) {
|
|
|
|
trackspares = lp->d_sparespertrack;
|
|
|
|
if (trackspares < 0)
|
|
|
|
trackspares = 0;
|
|
|
|
}
|
|
|
|
nphyssectors = nsectors + trackspares;
|
|
|
|
if (cylspares == -1) {
|
|
|
|
cylspares = lp->d_sparespercyl;
|
|
|
|
if (cylspares < 0)
|
|
|
|
cylspares = 0;
|
|
|
|
}
|
|
|
|
secpercyl = nsectors * ntracks - cylspares;
|
|
|
|
if (secpercyl != lp->d_secpercyl)
|
1997-07-01 02:20:30 +04:00
|
|
|
warnx("%s (%d) %s (%u)\n",
|
1993-03-21 12:45:37 +03:00
|
|
|
"Warning: calculated sectors per cylinder", secpercyl,
|
|
|
|
"disagrees with disk label", lp->d_secpercyl);
|
|
|
|
if (maxbpg == 0)
|
|
|
|
maxbpg = MAXBLKPG(bsize);
|
|
|
|
#ifdef notdef /* label may be 0 if faked up by kernel */
|
|
|
|
bbsize = lp->d_bbsize;
|
|
|
|
sbsize = lp->d_sbsize;
|
|
|
|
#endif
|
|
|
|
oldpartition = *pp;
|
|
|
|
mkfs(pp, special, fsi, fso);
|
2001-07-29 13:55:22 +04:00
|
|
|
if (!Nflag && memcmp(pp, &oldpartition, sizeof(oldpartition)) && !Fflag)
|
1993-03-21 12:45:37 +03:00
|
|
|
rewritelabel(special, fso, lp);
|
|
|
|
if (!Nflag)
|
|
|
|
close(fso);
|
|
|
|
close(fsi);
|
|
|
|
#ifdef MFS
|
|
|
|
if (mfs) {
|
|
|
|
struct mfs_args args;
|
|
|
|
|
1997-11-01 21:25:46 +03:00
|
|
|
switch (pid = fork()) {
|
|
|
|
case -1:
|
|
|
|
perror("mfs");
|
|
|
|
exit(10);
|
|
|
|
case 0:
|
1998-07-28 23:22:54 +04:00
|
|
|
(void)snprintf(mountfromname, sizeof(mountfromname),
|
|
|
|
"mfs:%d", getpid());
|
1997-11-01 21:25:46 +03:00
|
|
|
break;
|
|
|
|
default:
|
1998-07-28 23:22:54 +04:00
|
|
|
(void)snprintf(mountfromname, sizeof(mountfromname),
|
|
|
|
"mfs:%d", pid);
|
1997-11-01 21:25:46 +03:00
|
|
|
for (;;) {
|
|
|
|
/*
|
|
|
|
* spin until the mount succeeds
|
|
|
|
* or the child exits
|
|
|
|
*/
|
|
|
|
usleep(1);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* XXX Here is a race condition: another process
|
|
|
|
* can mount a filesystem which hides our
|
|
|
|
* ramdisk before we see the success.
|
|
|
|
*/
|
|
|
|
if (statfs(argv[1], &sf) < 0)
|
|
|
|
err(88, "statfs %s", argv[1]);
|
|
|
|
if (!strcmp(sf.f_mntfromname, mountfromname) &&
|
|
|
|
!strncmp(sf.f_mntonname, argv[1],
|
|
|
|
MNAMELEN) &&
|
|
|
|
!strcmp(sf.f_fstypename, "mfs"))
|
|
|
|
exit(0);
|
|
|
|
|
|
|
|
res = waitpid(pid, &status, WNOHANG);
|
|
|
|
if (res == -1)
|
|
|
|
err(11, "waitpid");
|
|
|
|
if (res != pid)
|
|
|
|
continue;
|
1997-11-19 12:48:52 +03:00
|
|
|
if (WIFEXITED(status)) {
|
|
|
|
if (WEXITSTATUS(status) == 0)
|
|
|
|
exit(0);
|
1997-11-01 21:25:46 +03:00
|
|
|
errx(1, "%s: mount: %s", argv[1],
|
|
|
|
strerror(WEXITSTATUS(status)));
|
1997-11-19 12:48:52 +03:00
|
|
|
} else
|
1997-11-01 21:25:46 +03:00
|
|
|
errx(11, "abnormal termination");
|
|
|
|
}
|
|
|
|
/* NOTREACHED */
|
|
|
|
}
|
|
|
|
|
|
|
|
(void) setsid();
|
|
|
|
(void) close(0);
|
|
|
|
(void) close(1);
|
|
|
|
(void) close(2);
|
|
|
|
(void) chdir("/");
|
|
|
|
|
|
|
|
args.fspec = mountfromname;
|
1994-06-08 23:27:32 +04:00
|
|
|
args.export.ex_root = -2;
|
|
|
|
if (mntflags & MNT_RDONLY)
|
|
|
|
args.export.ex_flags = MNT_EXRDONLY;
|
|
|
|
else
|
|
|
|
args.export.ex_flags = 0;
|
1993-03-21 12:45:37 +03:00
|
|
|
args.base = membase;
|
|
|
|
args.size = fssize * sectorsize;
|
|
|
|
if (mount(MOUNT_MFS, argv[1], mntflags, &args) < 0)
|
1997-11-01 21:25:46 +03:00
|
|
|
exit(errno); /* parent prints message */
|
1993-03-21 12:45:37 +03:00
|
|
|
}
|
|
|
|
#endif
|
|
|
|
exit(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
#ifdef COMPAT
|
2000-10-11 00:24:49 +04:00
|
|
|
const char lmsg[] = "%s: can't read disk label; disk type must be specified";
|
1993-03-21 12:45:37 +03:00
|
|
|
#else
|
2000-10-11 00:24:49 +04:00
|
|
|
const char lmsg[] = "%s: can't read disk label";
|
1993-03-21 12:45:37 +03:00
|
|
|
#endif
|
|
|
|
|
1997-07-01 02:20:30 +04:00
|
|
|
static struct disklabel *
|
2000-12-01 14:52:54 +03:00
|
|
|
getdisklabel(char *s, volatile int fd)
|
|
|
|
/* XXX why is fs volatile?! */
|
1993-03-21 12:45:37 +03:00
|
|
|
{
|
|
|
|
static struct disklabel lab;
|
|
|
|
|
1997-06-30 12:09:22 +04:00
|
|
|
if (ioctl(fd, DIOCGDINFO, &lab) < 0) {
|
1993-03-21 12:45:37 +03:00
|
|
|
#ifdef COMPAT
|
|
|
|
if (disktype) {
|
1997-07-01 02:20:30 +04:00
|
|
|
struct disklabel *lp;
|
1993-03-21 12:45:37 +03:00
|
|
|
|
|
|
|
unlabeled++;
|
|
|
|
lp = getdiskbyname(disktype);
|
|
|
|
if (lp == NULL)
|
1997-07-01 02:20:30 +04:00
|
|
|
errx(1, "%s: unknown disk type", disktype);
|
1993-03-21 12:45:37 +03:00
|
|
|
return (lp);
|
|
|
|
}
|
|
|
|
#endif
|
1994-06-08 23:27:32 +04:00
|
|
|
warn("ioctl (GDINFO)");
|
1997-07-01 02:20:30 +04:00
|
|
|
errx(1, lmsg, s);
|
1993-03-21 12:45:37 +03:00
|
|
|
}
|
|
|
|
return (&lab);
|
|
|
|
}
|
|
|
|
|
1997-07-01 02:20:30 +04:00
|
|
|
static void
|
2000-12-01 14:52:54 +03:00
|
|
|
rewritelabel(char *s, volatile int fd, struct disklabel *lp)
|
|
|
|
/* XXX why is fd volatile?! */
|
1993-03-21 12:45:37 +03:00
|
|
|
{
|
|
|
|
#ifdef COMPAT
|
|
|
|
if (unlabeled)
|
|
|
|
return;
|
|
|
|
#endif
|
|
|
|
lp->d_checksum = 0;
|
|
|
|
lp->d_checksum = dkcksum(lp);
|
|
|
|
if (ioctl(fd, DIOCWDINFO, (char *)lp) < 0) {
|
1994-06-08 23:27:32 +04:00
|
|
|
warn("ioctl (WDINFO)");
|
1997-07-01 02:20:30 +04:00
|
|
|
errx(1, "%s: can't rewrite disk label", s);
|
1993-03-21 12:45:37 +03:00
|
|
|
}
|
1998-10-20 05:48:14 +04:00
|
|
|
#if __vax__
|
1993-03-21 12:45:37 +03:00
|
|
|
if (lp->d_type == DTYPE_SMD && lp->d_flags & D_BADSECT) {
|
1997-07-01 02:20:30 +04:00
|
|
|
int i;
|
1993-03-21 12:45:37 +03:00
|
|
|
int cfd;
|
|
|
|
daddr_t alt;
|
|
|
|
char specname[64];
|
|
|
|
char blk[1024];
|
|
|
|
char *cp;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Make name for 'c' partition.
|
|
|
|
*/
|
|
|
|
strcpy(specname, s);
|
|
|
|
cp = specname + strlen(specname) - 1;
|
|
|
|
if (!isdigit(*cp))
|
|
|
|
*cp = 'c';
|
|
|
|
cfd = open(specname, O_WRONLY);
|
|
|
|
if (cfd < 0)
|
1997-09-29 09:18:00 +04:00
|
|
|
err(1, "%s: open", specname);
|
1994-09-23 18:26:58 +04:00
|
|
|
memset(blk, 0, sizeof(blk));
|
1993-03-21 12:45:37 +03:00
|
|
|
*(struct disklabel *)(blk + LABELOFFSET) = *lp;
|
|
|
|
alt = lp->d_ncylinders * lp->d_secpercyl - lp->d_nsectors;
|
|
|
|
for (i = 1; i < 11 && i < lp->d_nsectors; i += 2) {
|
1995-03-18 10:02:29 +03:00
|
|
|
off_t offset;
|
|
|
|
|
|
|
|
offset = alt + i;
|
|
|
|
offset *= lp->d_secsize;
|
|
|
|
if (lseek(cfd, offset, SEEK_SET) == -1)
|
1997-07-01 02:20:30 +04:00
|
|
|
err(1, "lseek to badsector area: ");
|
1993-03-21 12:45:37 +03:00
|
|
|
if (write(cfd, blk, lp->d_secsize) < lp->d_secsize)
|
1994-06-08 23:27:32 +04:00
|
|
|
warn("alternate label %d write", i/2);
|
1993-03-21 12:45:37 +03:00
|
|
|
}
|
|
|
|
close(cfd);
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
2001-07-29 13:55:22 +04:00
|
|
|
static int
|
|
|
|
strsuftoi(const char *desc, const char *arg, int min, int max)
|
|
|
|
{
|
|
|
|
long long result;
|
|
|
|
char *ep;
|
|
|
|
|
|
|
|
errno = 0;
|
|
|
|
result = strtoll(arg, &ep, 10);
|
|
|
|
if (ep[0] != '\0' && ep[1] != '\0')
|
|
|
|
errx(1, "%s `%s' is not a valid number.", desc, arg);
|
|
|
|
switch (tolower((unsigned char)ep[0])) {
|
|
|
|
case '\0':
|
|
|
|
case 'b':
|
|
|
|
break;
|
|
|
|
case 'k':
|
|
|
|
result <<= 10;
|
|
|
|
break;
|
|
|
|
case 'm':
|
|
|
|
result <<= 20;
|
|
|
|
break;
|
|
|
|
case 'g':
|
|
|
|
result <<= 30;
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
errx(1, "`%s' is not a valid suffix for %s.", ep, desc);
|
|
|
|
}
|
|
|
|
if (result < min)
|
|
|
|
errx(1, "%s `%s' (%lld) is less than minimum (%d).",
|
|
|
|
desc, arg, result, min);
|
|
|
|
if (result > max)
|
|
|
|
errx(1, "%s `%s' (%lld) is greater than maximum (%d).",
|
|
|
|
desc, arg, result, max);
|
|
|
|
return ((int)result);
|
|
|
|
}
|
|
|
|
|
1997-07-01 02:20:30 +04:00
|
|
|
static void
|
2000-12-01 14:52:54 +03:00
|
|
|
usage(void)
|
1993-03-21 12:45:37 +03:00
|
|
|
{
|
2001-07-29 13:55:22 +04:00
|
|
|
|
1993-03-21 12:45:37 +03:00
|
|
|
if (mfs) {
|
|
|
|
fprintf(stderr,
|
2001-07-29 13:55:22 +04:00
|
|
|
"usage: %s [ fsoptions ] special-device mount-point\n",
|
2001-02-20 01:48:57 +03:00
|
|
|
getprogname());
|
1993-03-21 12:45:37 +03:00
|
|
|
} else
|
|
|
|
fprintf(stderr,
|
2001-07-29 13:55:22 +04:00
|
|
|
"usage: %s [ fsoptions ] special-device%s\n",
|
2001-02-20 01:48:57 +03:00
|
|
|
getprogname(),
|
1993-03-21 12:45:37 +03:00
|
|
|
#ifdef COMPAT
|
|
|
|
" [device-type]");
|
|
|
|
#else
|
|
|
|
"");
|
|
|
|
#endif
|
|
|
|
fprintf(stderr, "where fsoptions are:\n");
|
2001-07-29 13:55:22 +04:00
|
|
|
if (!mfs) {
|
|
|
|
fprintf(stderr,
|
|
|
|
"\t-B byteorder\tbyte order (`be' or `le')\n");
|
|
|
|
fprintf(stderr,
|
|
|
|
"\t-F\t\tcreate file system image in regular file\n");
|
|
|
|
}
|
|
|
|
fprintf(stderr, "\t-N\t\tdo not create file system, "
|
|
|
|
"just print out parameters\n");
|
|
|
|
if (!mfs) {
|
|
|
|
fprintf(stderr,
|
|
|
|
"\t-O\t\tcreate a 4.3BSD format filesystem\n");
|
|
|
|
fprintf(stderr,
|
|
|
|
"\t-S secsize\tsector size\n");
|
|
|
|
}
|
1993-03-21 12:45:37 +03:00
|
|
|
#ifdef COMPAT
|
2001-07-29 13:55:22 +04:00
|
|
|
fprintf(stderr, "\t-T disktype\tdisk type\n");
|
1993-03-21 12:45:37 +03:00
|
|
|
#endif
|
2001-07-29 13:55:22 +04:00
|
|
|
if (!mfs)
|
|
|
|
fprintf(stderr,
|
|
|
|
"\t-Z\t\tpre-zero the image file (with -F)\n");
|
|
|
|
fprintf(stderr, "\t-a maxcontig\tmaximum contiguous blocks\n");
|
|
|
|
fprintf(stderr, "\t-b bsize\tblock size\n");
|
|
|
|
fprintf(stderr, "\t-c cpg\t\tcylinders/group\n");
|
|
|
|
fprintf(stderr, "\t-d rotdelay\trotational delay between "
|
|
|
|
"contiguous blocks\n");
|
|
|
|
fprintf(stderr, "\t-e maxbpg\tmaximum blocks per file "
|
|
|
|
"in a cylinder group\n");
|
|
|
|
fprintf(stderr, "\t-f fsize\tfragment size\n");
|
Incorporate the enhanced ffs_dirpref() by Grigoriy Orlov, as found in
FreeBSD (three commits; the initial work, man page updates, and a fix
to ffs_reload()), with the following differences:
- Be consistent between newfs(8) and tunefs(8) as to the options which
set and control the tuning parameters for this work (avgfilesize & avgfpdir)
- Use u_int16_t instead of u_int8_t to keep track of the number of
contiguous directories (suggested by Chuck Silvers)
- Work within our FFS_EI framework
- Ensure that fs->fs_maxclusters and fs->fs_contigdirs don't point to
the same area of memory
The new algorithm has a marked performance increase, especially when
performing tasks such as untarring pkgsrc.tar.gz, etc.
The original FreeBSD commit messages are attached:
=====
mckusick 2001/04/10 01:39:00 PDT
Directory layout preference improvements from Grigoriy Orlov <gluk@ptci.ru>.
His description of the problem and solution follow. My own tests show
speedups on typical filesystem intensive workloads of 5% to 12% which
is very impressive considering the small amount of code change involved.
------
One day I noticed that some file operations run much faster on
small file systems then on big ones. I've looked at the ffs
algorithms, thought about them, and redesigned the dirpref algorithm.
First I want to describe the results of my tests. These results are old
and I have improved the algorithm after these tests were done. Nevertheless
they show how big the perfomance speedup may be. I have done two file/directory
intensive tests on a two OpenBSD systems with old and new dirpref algorithm.
The first test is "tar -xzf ports.tar.gz", the second is "rm -rf ports".
The ports.tar.gz file is the ports collection from the OpenBSD 2.8 release.
It contains 6596 directories and 13868 files. The test systems are:
1. Celeron-450, 128Mb, two IDE drives, the system at wd0, file system for
test is at wd1. Size of test file system is 8 Gb, number of cg=991,
size of cg is 8m, block size = 8k, fragment size = 1k OpenBSD-current
from Dec 2000 with BUFCACHEPERCENT=35
2. PIII-600, 128Mb, two IBM DTLA-307045 IDE drives at i815e, the system
at wd0, file system for test is at wd1. Size of test file system is 40 Gb,
number of cg=5324, size of cg is 8m, block size = 8k, fragment size = 1k
OpenBSD-current from Dec 2000 with BUFCACHEPERCENT=50
You can get more info about the test systems and methods at:
http://www.ptci.ru/gluk/dirpref/old/dirpref.html
Test Results
tar -xzf ports.tar.gz rm -rf ports
mode old dirpref new dirpref speedup old dirprefnew dirpref speedup
First system
normal 667 472 1.41 477 331 1.44
async 285 144 1.98 130 14 9.29
sync 768 616 1.25 477 334 1.43
softdep 413 252 1.64 241 38 6.34
Second system
normal 329 81 4.06 263.5 93.5 2.81
async 302 25.7 11.75 112 2.26 49.56
sync 281 57.0 4.93 263 90.5 2.9
softdep 341 40.6 8.4 284 4.76 59.66
"old dirpref" and "new dirpref" columns give a test time in seconds.
speedup - speed increasement in times, ie. old dirpref / new dirpref.
------
Algorithm description
The old dirpref algorithm is described in comments:
/*
* Find a cylinder to place a directory.
*
* The policy implemented by this algorithm is to select from
* among those cylinder groups with above the average number of
* free inodes, the one with the smallest number of directories.
*/
A new directory is allocated in a different cylinder groups than its
parent directory resulting in a directory tree that is spreaded across
all the cylinder groups. This spreading out results in a non-optimal
access to the directories and files. When we have a small filesystem
it is not a problem but when the filesystem is big then perfomance
degradation becomes very apparent.
What I mean by a big file system ?
1. A big filesystem is a filesystem which occupy 20-30 or more percent
of total drive space, i.e. first and last cylinder are physically
located relatively far from each other.
2. It has a relatively large number of cylinder groups, for example
more cylinder groups than 50% of the buffers in the buffer cache.
The first results in long access times, while the second results in
many buffers being used by metadata operations. Such operations use
cylinder group blocks and on-disk inode blocks. The cylinder group
block (fs->fs_cblkno) contains struct cg, inode and block bit maps.
It is 2k in size for the default filesystem parameters. If new and
parent directories are located in different cylinder groups then the
system performs more input/output operations and uses more buffers.
On filesystems with many cylinder groups, lots of cache buffers are
used for metadata operations.
My solution for this problem is very simple. I allocate many directories
in one cylinder group. I also do some things, so that the new allocation
method does not cause excessive fragmentation and all directory inodes
will not be located at a location far from its file's inodes and data.
The algorithm is:
/*
* Find a cylinder group to place a directory.
*
* The policy implemented by this algorithm is to allocate a
* directory inode in the same cylinder group as its parent
* directory, but also to reserve space for its files inodes
* and data. Restrict the number of directories which may be
* allocated one after another in the same cylinder group
* without intervening allocation of files.
*
* If we allocate a first level directory then force allocation
* in another cylinder group.
*/
My early versions of dirpref give me a good results for a wide range of
file operations and different filesystem capacities except one case:
those applications that create their entire directory structure first
and only later fill this structure with files.
My solution for such and similar cases is to limit a number of
directories which may be created one after another in the same cylinder
group without intervening file creations. For this purpose, I allocate
an array of counters at mount time. This array is linked to the superblock
fs->fs_contigdirs[cg]. Each time a directory is created the counter
increases and each time a file is created the counter decreases. A 60Gb
filesystem with 8mb/cg requires 10kb of memory for the counters array.
The maxcontigdirs is a maximum number of directories which may be created
without an intervening file creation. I found in my tests that the best
performance occurs when I restrict the number of directories in one cylinder
group such that all its files may be located in the same cylinder group.
There may be some deterioration in performance if all the file inodes
are in the same cylinder group as its containing directory, but their
data partially resides in a different cylinder group. The maxcontigdirs
value is calculated to try to prevent this condition. Since there is
no way to know how many files and directories will be allocated later
I added two optimization parameters in superblock/tunefs. They are:
int32_t fs_avgfilesize; /* expected average file size */
int32_t fs_avgfpdir; /* expected # of files per directory */
These parameters have reasonable defaults but may be tweeked for special
uses of a filesystem. They are only necessary in rare cases like better
tuning a filesystem being used to store a squid cache.
I have been using this algorithm for about 3 months. I have done
a lot of testing on filesystems with different capacities, average
filesize, average number of files per directory, and so on. I think
this algorithm has no negative impact on filesystem perfomance. It
works better than the default one in all cases. The new dirpref
will greatly improve untarring/removing/coping of big directories,
decrease load on cvs servers and much more. The new dirpref doesn't
speedup a compilation process, but also doesn't slow it down.
Obtained from: Grigoriy Orlov <gluk@ptci.ru>
=====
=====
iedowse 2001/04/23 17:37:17 PDT
Pre-dirpref versions of fsck may zero out the new superblock fields
fs_contigdirs, fs_avgfilesize and fs_avgfpdir. This could cause
panics if these fields were zeroed while a filesystem was mounted
read-only, and then remounted read-write.
Add code to ffs_reload() which copies the fs_contigdirs pointer
from the previous superblock, and reinitialises fs_avgf* if necessary.
Reviewed by: mckusick
=====
=====
nik 2001/04/10 03:36:44 PDT
Add information about the new options to newfs and tunefs which set the
expected average file size and number of files per directory. Could do
with some fleshing out.
=====
2001-09-06 06:16:00 +04:00
|
|
|
fprintf(stderr, "\t-g average file size\n");
|
|
|
|
fprintf(stderr, "\t-h average files per directory\n");
|
2001-07-29 13:55:22 +04:00
|
|
|
fprintf(stderr, "\t-i density\tnumber of bytes per inode\n");
|
|
|
|
if (!mfs) {
|
|
|
|
fprintf(stderr,
|
|
|
|
"\t-k trackskew\tsector 0 skew, per track\n");
|
|
|
|
fprintf(stderr,
|
|
|
|
"\t-l interleave\thardware sector interleave\n");
|
|
|
|
}
|
|
|
|
fprintf(stderr, "\t-m minfree\tminimum free space %%\n");
|
|
|
|
if (!mfs)
|
|
|
|
fprintf(stderr,
|
|
|
|
"\t-n nrpos\tnumber of distinguished "
|
|
|
|
"rotational positions\n");
|
|
|
|
fprintf(stderr, "\t-o optim\toptimization preference "
|
|
|
|
"(`space' or `time')\n");
|
|
|
|
if (!mfs)
|
|
|
|
fprintf(stderr,
|
|
|
|
"\t-p trackspares\tspare sectors per track\n");
|
|
|
|
fprintf(stderr, "\t-s fssize\tfile system size (sectors)\n");
|
|
|
|
if (!mfs) {
|
|
|
|
fprintf(stderr,
|
|
|
|
"\t-r rpm\t\trevolutions/minute\n");
|
|
|
|
fprintf(stderr,
|
|
|
|
"\t-t ntracks\ttracks/cylinder\n");
|
|
|
|
fprintf(stderr,
|
|
|
|
"\t-u nsectors\tsectors/track\n");
|
|
|
|
fprintf(stderr,
|
|
|
|
"\t-x cylspares\tspare sectors per cylinder\n");
|
|
|
|
}
|
1993-03-21 12:45:37 +03:00
|
|
|
exit(1);
|
|
|
|
}
|