1994-05-26 06:35:07 +00:00
|
|
|
/*
|
|
|
|
* Copyright (c) 1983, 1993
|
|
|
|
* The Regents of the University of California. All rights reserved.
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
* modification, are permitted provided that the following conditions
|
|
|
|
* are met:
|
|
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer.
|
|
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
* documentation and/or other materials provided with the distribution.
|
|
|
|
* 4. Neither the name of the University nor the names of its contributors
|
|
|
|
* may be used to endorse or promote products derived from this software
|
|
|
|
* without specific prior written permission.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
|
|
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
|
|
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
|
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
|
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
|
|
* SUCH DAMAGE.
|
|
|
|
*/
|
|
|
|
|
2003-05-03 18:41:59 +00:00
|
|
|
#if 0
|
1994-05-26 06:35:07 +00:00
|
|
|
#ifndef lint
|
1998-08-03 06:41:20 +00:00
|
|
|
static const char copyright[] =
|
1994-05-26 06:35:07 +00:00
|
|
|
"@(#) Copyright (c) 1983, 1993\n\
|
|
|
|
The Regents of the University of California. All rights reserved.\n";
|
|
|
|
#endif /* not lint */
|
|
|
|
|
|
|
|
#ifndef lint
|
|
|
|
static char sccsid[] = "@(#)tunefs.c 8.2 (Berkeley) 4/19/94";
|
|
|
|
#endif /* not lint */
|
2003-05-03 18:41:59 +00:00
|
|
|
#endif
|
|
|
|
#include <sys/cdefs.h>
|
|
|
|
__FBSDID("$FreeBSD$");
|
1994-05-26 06:35:07 +00:00
|
|
|
|
|
|
|
/*
|
2002-08-21 18:11:48 +00:00
|
|
|
* tunefs: change layout parameters to an existing file system.
|
1994-05-26 06:35:07 +00:00
|
|
|
*/
|
|
|
|
#include <sys/param.h>
|
1999-01-20 01:22:39 +00:00
|
|
|
#include <sys/mount.h>
|
2002-05-12 21:37:08 +00:00
|
|
|
#include <sys/disklabel.h>
|
1994-05-26 06:35:07 +00:00
|
|
|
#include <sys/stat.h>
|
|
|
|
|
1999-01-20 01:22:39 +00:00
|
|
|
#include <ufs/ufs/ufsmount.h>
|
2002-06-21 06:18:05 +00:00
|
|
|
#include <ufs/ufs/dinode.h>
|
|
|
|
#include <ufs/ffs/fs.h>
|
2010-04-24 07:05:35 +00:00
|
|
|
#include <ufs/ufs/dir.h>
|
1994-05-26 06:35:07 +00:00
|
|
|
|
2003-02-01 04:17:10 +00:00
|
|
|
#include <ctype.h>
|
1994-05-26 06:35:07 +00:00
|
|
|
#include <err.h>
|
|
|
|
#include <fcntl.h>
|
|
|
|
#include <fstab.h>
|
2003-01-20 21:15:02 +00:00
|
|
|
#include <libufs.h>
|
1994-05-26 06:35:07 +00:00
|
|
|
#include <paths.h>
|
1998-08-03 06:41:20 +00:00
|
|
|
#include <stdio.h>
|
1994-05-26 06:35:07 +00:00
|
|
|
#include <stdlib.h>
|
2010-04-24 07:05:35 +00:00
|
|
|
#include <stdint.h>
|
2000-03-14 07:44:32 +00:00
|
|
|
#include <string.h>
|
2011-10-11 19:03:57 +00:00
|
|
|
#include <time.h>
|
1994-05-26 06:35:07 +00:00
|
|
|
#include <unistd.h>
|
|
|
|
|
|
|
|
/* the optimization warning string template */
|
|
|
|
#define OPTWARN "should optimize for %s with minfree %s %d%%"
|
|
|
|
|
2011-11-04 13:36:02 +00:00
|
|
|
static struct uufsd disk;
|
2003-01-20 21:15:02 +00:00
|
|
|
#define sblock disk.d_fs
|
1994-05-26 06:35:07 +00:00
|
|
|
|
2002-03-21 13:20:49 +00:00
|
|
|
void usage(void);
|
|
|
|
void printfs(void);
|
2010-04-24 07:05:35 +00:00
|
|
|
int journal_alloc(int64_t size);
|
|
|
|
void journal_clear(void);
|
|
|
|
void sbdirty(void);
|
1994-05-26 06:35:07 +00:00
|
|
|
|
|
|
|
int
|
2003-01-20 21:15:02 +00:00
|
|
|
main(int argc, char *argv[])
|
1994-05-26 06:35:07 +00:00
|
|
|
{
|
2010-04-24 07:05:35 +00:00
|
|
|
char *avalue, *jvalue, *Jvalue, *Lvalue, *lvalue, *Nvalue, *nvalue;
|
2010-12-29 12:31:18 +00:00
|
|
|
char *tvalue;
|
2003-01-28 02:42:01 +00:00
|
|
|
const char *special, *on;
|
2001-07-15 05:47:47 +00:00
|
|
|
const char *name;
|
2004-03-26 16:11:13 +00:00
|
|
|
int active;
|
2010-04-24 07:05:35 +00:00
|
|
|
int Aflag, aflag, eflag, evalue, fflag, fvalue, jflag, Jflag, Lflag;
|
|
|
|
int lflag, mflag, mvalue, Nflag, nflag, oflag, ovalue, pflag, sflag;
|
2010-12-29 12:31:18 +00:00
|
|
|
int tflag;
|
2012-01-10 02:58:52 +00:00
|
|
|
int svalue, Svalue;
|
2004-03-26 08:39:36 +00:00
|
|
|
int ch, found_arg, i;
|
2001-07-15 05:47:47 +00:00
|
|
|
const char *chg[2];
|
1999-01-20 01:22:39 +00:00
|
|
|
struct ufs_args args;
|
|
|
|
struct statfs stfs;
|
1994-05-26 06:35:07 +00:00
|
|
|
|
2004-03-26 08:39:36 +00:00
|
|
|
if (argc < 3)
|
|
|
|
usage();
|
2010-04-24 07:05:35 +00:00
|
|
|
Aflag = aflag = eflag = fflag = jflag = Jflag = Lflag = lflag = 0;
|
2010-12-29 12:31:18 +00:00
|
|
|
mflag = Nflag = nflag = oflag = pflag = sflag = tflag = 0;
|
2010-04-24 07:05:35 +00:00
|
|
|
avalue = jvalue = Jvalue = Lvalue = lvalue = Nvalue = nvalue = NULL;
|
|
|
|
evalue = fvalue = mvalue = ovalue = svalue = Svalue = 0;
|
2004-03-26 16:11:13 +00:00
|
|
|
active = 0;
|
2004-03-26 08:39:36 +00:00
|
|
|
found_arg = 0; /* At least one arg is required. */
|
2010-12-29 12:31:18 +00:00
|
|
|
while ((ch = getopt(argc, argv, "Aa:e:f:j:J:L:l:m:N:n:o:ps:S:t:"))
|
|
|
|
!= -1)
|
2004-03-26 08:39:36 +00:00
|
|
|
switch (ch) {
|
|
|
|
|
|
|
|
case 'A':
|
|
|
|
found_arg = 1;
|
|
|
|
Aflag++;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 'a':
|
|
|
|
found_arg = 1;
|
2009-12-21 19:39:10 +00:00
|
|
|
name = "POSIX.1e ACLs";
|
2004-03-26 08:39:36 +00:00
|
|
|
avalue = optarg;
|
|
|
|
if (strcmp(avalue, "enable") &&
|
|
|
|
strcmp(avalue, "disable")) {
|
|
|
|
errx(10, "bad %s (options are %s)",
|
|
|
|
name, "`enable' or `disable'");
|
|
|
|
}
|
|
|
|
aflag = 1;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 'e':
|
|
|
|
found_arg = 1;
|
|
|
|
name = "maximum blocks per file in a cylinder group";
|
|
|
|
evalue = atoi(optarg);
|
|
|
|
if (evalue < 1)
|
|
|
|
errx(10, "%s must be >= 1 (was %s)",
|
|
|
|
name, optarg);
|
|
|
|
eflag = 1;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 'f':
|
|
|
|
found_arg = 1;
|
|
|
|
name = "average file size";
|
|
|
|
fvalue = atoi(optarg);
|
|
|
|
if (fvalue < 1)
|
|
|
|
errx(10, "%s must be >= 1 (was %s)",
|
|
|
|
name, optarg);
|
|
|
|
fflag = 1;
|
|
|
|
break;
|
|
|
|
|
2010-04-24 07:05:35 +00:00
|
|
|
case 'j':
|
|
|
|
found_arg = 1;
|
|
|
|
name = "softdep journaled file system";
|
|
|
|
jvalue = optarg;
|
|
|
|
if (strcmp(jvalue, "enable") &&
|
|
|
|
strcmp(jvalue, "disable")) {
|
|
|
|
errx(10, "bad %s (options are %s)",
|
|
|
|
name, "`enable' or `disable'");
|
|
|
|
}
|
|
|
|
jflag = 1;
|
|
|
|
break;
|
|
|
|
|
2006-10-31 21:52:28 +00:00
|
|
|
case 'J':
|
|
|
|
found_arg = 1;
|
|
|
|
name = "gjournaled file system";
|
|
|
|
Jvalue = optarg;
|
|
|
|
if (strcmp(Jvalue, "enable") &&
|
|
|
|
strcmp(Jvalue, "disable")) {
|
|
|
|
errx(10, "bad %s (options are %s)",
|
|
|
|
name, "`enable' or `disable'");
|
|
|
|
}
|
|
|
|
Jflag = 1;
|
|
|
|
break;
|
|
|
|
|
|
|
|
|
2004-03-26 08:39:36 +00:00
|
|
|
case 'L':
|
|
|
|
found_arg = 1;
|
|
|
|
name = "volume label";
|
|
|
|
Lvalue = optarg;
|
|
|
|
i = -1;
|
|
|
|
while (isalnum(Lvalue[++i]));
|
|
|
|
if (Lvalue[i] != '\0') {
|
|
|
|
errx(10,
|
|
|
|
"bad %s. Valid characters are alphanumerics.",
|
|
|
|
name);
|
|
|
|
}
|
|
|
|
if (strlen(Lvalue) >= MAXVOLLEN) {
|
|
|
|
errx(10, "bad %s. Length is longer than %d.",
|
|
|
|
name, MAXVOLLEN - 1);
|
|
|
|
}
|
|
|
|
Lflag = 1;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 'l':
|
|
|
|
found_arg = 1;
|
|
|
|
name = "multilabel MAC file system";
|
|
|
|
lvalue = optarg;
|
|
|
|
if (strcmp(lvalue, "enable") &&
|
|
|
|
strcmp(lvalue, "disable")) {
|
|
|
|
errx(10, "bad %s (options are %s)",
|
|
|
|
name, "`enable' or `disable'");
|
|
|
|
}
|
|
|
|
lflag = 1;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 'm':
|
|
|
|
found_arg = 1;
|
|
|
|
name = "minimum percentage of free space";
|
|
|
|
mvalue = atoi(optarg);
|
|
|
|
if (mvalue < 0 || mvalue > 99)
|
|
|
|
errx(10, "bad %s (%s)", name, optarg);
|
|
|
|
mflag = 1;
|
|
|
|
break;
|
|
|
|
|
2009-12-21 19:39:10 +00:00
|
|
|
case 'N':
|
|
|
|
found_arg = 1;
|
|
|
|
name = "NFSv4 ACLs";
|
|
|
|
Nvalue = optarg;
|
|
|
|
if (strcmp(Nvalue, "enable") &&
|
|
|
|
strcmp(Nvalue, "disable")) {
|
|
|
|
errx(10, "bad %s (options are %s)",
|
|
|
|
name, "`enable' or `disable'");
|
|
|
|
}
|
|
|
|
Nflag = 1;
|
|
|
|
break;
|
|
|
|
|
2004-03-26 08:39:36 +00:00
|
|
|
case 'n':
|
|
|
|
found_arg = 1;
|
|
|
|
name = "soft updates";
|
|
|
|
nvalue = optarg;
|
2004-03-26 16:11:13 +00:00
|
|
|
if (strcmp(nvalue, "enable") != 0 &&
|
|
|
|
strcmp(nvalue, "disable") != 0) {
|
2004-03-26 08:39:36 +00:00
|
|
|
errx(10, "bad %s (options are %s)",
|
|
|
|
name, "`enable' or `disable'");
|
|
|
|
}
|
|
|
|
nflag = 1;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 'o':
|
|
|
|
found_arg = 1;
|
|
|
|
name = "optimization preference";
|
2004-03-26 16:11:13 +00:00
|
|
|
if (strcmp(optarg, "space") == 0)
|
2004-03-26 08:39:36 +00:00
|
|
|
ovalue = FS_OPTSPACE;
|
2004-03-26 16:11:13 +00:00
|
|
|
else if (strcmp(optarg, "time") == 0)
|
2004-03-26 08:39:36 +00:00
|
|
|
ovalue = FS_OPTTIME;
|
|
|
|
else
|
|
|
|
errx(10,
|
|
|
|
"bad %s (options are `space' or `time')",
|
|
|
|
name);
|
|
|
|
oflag = 1;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 'p':
|
|
|
|
found_arg = 1;
|
|
|
|
pflag = 1;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 's':
|
|
|
|
found_arg = 1;
|
|
|
|
name = "expected number of files per directory";
|
|
|
|
svalue = atoi(optarg);
|
|
|
|
if (svalue < 1)
|
|
|
|
errx(10, "%s must be >= 1 (was %s)",
|
|
|
|
name, optarg);
|
|
|
|
sflag = 1;
|
|
|
|
break;
|
|
|
|
|
2010-04-24 07:05:35 +00:00
|
|
|
case 'S':
|
|
|
|
found_arg = 1;
|
|
|
|
name = "Softdep Journal Size";
|
|
|
|
Svalue = atoi(optarg);
|
|
|
|
if (Svalue < SUJ_MIN)
|
|
|
|
errx(10, "%s must be >= %d (was %s)",
|
|
|
|
name, SUJ_MIN, optarg);
|
|
|
|
break;
|
|
|
|
|
2010-12-29 12:31:18 +00:00
|
|
|
case 't':
|
|
|
|
found_arg = 1;
|
|
|
|
name = "trim";
|
|
|
|
tvalue = optarg;
|
|
|
|
if (strcmp(tvalue, "enable") != 0 &&
|
|
|
|
strcmp(tvalue, "disable") != 0) {
|
|
|
|
errx(10, "bad %s (options are %s)",
|
|
|
|
name, "`enable' or `disable'");
|
|
|
|
}
|
|
|
|
tflag = 1;
|
|
|
|
break;
|
|
|
|
|
2004-03-26 08:39:36 +00:00
|
|
|
default:
|
|
|
|
usage();
|
2002-10-14 19:52:12 +00:00
|
|
|
}
|
2000-11-28 18:17:15 +00:00
|
|
|
argc -= optind;
|
|
|
|
argv += optind;
|
|
|
|
if (found_arg == 0 || argc != 1)
|
2004-03-26 08:39:36 +00:00
|
|
|
usage();
|
2000-11-28 18:17:15 +00:00
|
|
|
|
2003-01-28 02:42:01 +00:00
|
|
|
on = special = argv[0];
|
|
|
|
if (ufs_disk_fillout(&disk, special) == -1)
|
|
|
|
goto err;
|
|
|
|
if (disk.d_name != special) {
|
2010-04-30 04:21:22 +00:00
|
|
|
if (statfs(special, &stfs) != 0)
|
|
|
|
warn("Can't stat %s", special);
|
|
|
|
if (strcmp(special, stfs.f_mntonname) == 0)
|
2000-12-10 20:59:30 +00:00
|
|
|
active = 1;
|
2003-01-20 21:15:02 +00:00
|
|
|
}
|
2000-12-10 20:59:30 +00:00
|
|
|
|
|
|
|
if (pflag) {
|
|
|
|
printfs();
|
|
|
|
exit(0);
|
|
|
|
}
|
2003-02-01 04:17:10 +00:00
|
|
|
if (Lflag) {
|
|
|
|
name = "volume label";
|
|
|
|
strlcpy(sblock.fs_volname, Lvalue, MAXVOLLEN);
|
|
|
|
}
|
2002-10-14 19:52:12 +00:00
|
|
|
if (aflag) {
|
2009-12-21 19:39:10 +00:00
|
|
|
name = "POSIX.1e ACLs";
|
2002-10-14 19:52:12 +00:00
|
|
|
if (strcmp(avalue, "enable") == 0) {
|
|
|
|
if (sblock.fs_flags & FS_ACLS) {
|
|
|
|
warnx("%s remains unchanged as enabled", name);
|
2009-12-21 19:39:10 +00:00
|
|
|
} else if (sblock.fs_flags & FS_NFS4ACLS) {
|
|
|
|
warnx("%s and NFSv4 ACLs are mutually "
|
|
|
|
"exclusive", name);
|
2002-10-14 19:52:12 +00:00
|
|
|
} else {
|
|
|
|
sblock.fs_flags |= FS_ACLS;
|
|
|
|
warnx("%s set", name);
|
|
|
|
}
|
|
|
|
} else if (strcmp(avalue, "disable") == 0) {
|
|
|
|
if ((~sblock.fs_flags & FS_ACLS) ==
|
|
|
|
FS_ACLS) {
|
|
|
|
warnx("%s remains unchanged as disabled",
|
|
|
|
name);
|
|
|
|
} else {
|
|
|
|
sblock.fs_flags &= ~FS_ACLS;
|
2002-10-16 05:03:40 +00:00
|
|
|
warnx("%s cleared", name);
|
2002-10-14 19:52:12 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2000-12-10 20:59:30 +00:00
|
|
|
if (eflag) {
|
|
|
|
name = "maximum blocks per file in a cylinder group";
|
2004-03-26 16:11:13 +00:00
|
|
|
if (sblock.fs_maxbpg == evalue)
|
2000-12-10 20:59:30 +00:00
|
|
|
warnx("%s remains unchanged as %d", name, evalue);
|
|
|
|
else {
|
|
|
|
warnx("%s changes from %d to %d",
|
2004-03-26 16:11:13 +00:00
|
|
|
name, sblock.fs_maxbpg, evalue);
|
2000-12-10 20:59:30 +00:00
|
|
|
sblock.fs_maxbpg = evalue;
|
|
|
|
}
|
|
|
|
}
|
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>
2001-04-10 08:38:59 +00:00
|
|
|
if (fflag) {
|
|
|
|
name = "average file size";
|
2010-02-11 06:33:35 +00:00
|
|
|
if (sblock.fs_avgfilesize == (unsigned)fvalue) {
|
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>
2001-04-10 08:38:59 +00:00
|
|
|
warnx("%s remains unchanged as %d", name, fvalue);
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
warnx("%s changes from %d to %d",
|
|
|
|
name, sblock.fs_avgfilesize, fvalue);
|
|
|
|
sblock.fs_avgfilesize = fvalue;
|
|
|
|
}
|
|
|
|
}
|
2010-04-24 07:05:35 +00:00
|
|
|
if (jflag) {
|
|
|
|
name = "soft updates journaling";
|
|
|
|
if (strcmp(jvalue, "enable") == 0) {
|
|
|
|
if ((sblock.fs_flags & (FS_DOSOFTDEP | FS_SUJ)) ==
|
|
|
|
(FS_DOSOFTDEP | FS_SUJ)) {
|
|
|
|
warnx("%s remains unchanged as enabled", name);
|
|
|
|
} else if (sblock.fs_clean == 0) {
|
|
|
|
warnx("%s cannot be enabled until fsck is run",
|
|
|
|
name);
|
|
|
|
} else if (journal_alloc(Svalue) != 0) {
|
|
|
|
warnx("%s can not be enabled", name);
|
|
|
|
} else {
|
|
|
|
sblock.fs_flags |= FS_DOSOFTDEP | FS_SUJ;
|
|
|
|
warnx("%s set", name);
|
|
|
|
}
|
|
|
|
} else if (strcmp(jvalue, "disable") == 0) {
|
|
|
|
if ((~sblock.fs_flags & FS_SUJ) == FS_SUJ) {
|
|
|
|
warnx("%s remains unchanged as disabled", name);
|
|
|
|
} else {
|
|
|
|
journal_clear();
|
2010-05-18 01:45:28 +00:00
|
|
|
sblock.fs_flags &= ~FS_SUJ;
|
2010-04-24 07:05:35 +00:00
|
|
|
sblock.fs_sujfree = 0;
|
2010-05-18 01:45:28 +00:00
|
|
|
warnx("%s cleared but soft updates still set.",
|
|
|
|
name);
|
|
|
|
|
|
|
|
warnx("remove .sujournal to reclaim space");
|
2010-04-24 07:05:35 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2006-10-31 21:52:28 +00:00
|
|
|
if (Jflag) {
|
|
|
|
name = "gjournal";
|
|
|
|
if (strcmp(Jvalue, "enable") == 0) {
|
|
|
|
if (sblock.fs_flags & FS_GJOURNAL) {
|
|
|
|
warnx("%s remains unchanged as enabled", name);
|
|
|
|
} else {
|
|
|
|
sblock.fs_flags |= FS_GJOURNAL;
|
|
|
|
warnx("%s set", name);
|
|
|
|
}
|
|
|
|
} else if (strcmp(Jvalue, "disable") == 0) {
|
|
|
|
if ((~sblock.fs_flags & FS_GJOURNAL) ==
|
|
|
|
FS_GJOURNAL) {
|
|
|
|
warnx("%s remains unchanged as disabled",
|
|
|
|
name);
|
|
|
|
} else {
|
|
|
|
sblock.fs_flags &= ~FS_GJOURNAL;
|
|
|
|
warnx("%s cleared", name);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2002-10-14 19:52:12 +00:00
|
|
|
if (lflag) {
|
|
|
|
name = "multilabel";
|
|
|
|
if (strcmp(lvalue, "enable") == 0) {
|
|
|
|
if (sblock.fs_flags & FS_MULTILABEL) {
|
|
|
|
warnx("%s remains unchanged as enabled", name);
|
|
|
|
} else {
|
|
|
|
sblock.fs_flags |= FS_MULTILABEL;
|
|
|
|
warnx("%s set", name);
|
|
|
|
}
|
|
|
|
} else if (strcmp(lvalue, "disable") == 0) {
|
|
|
|
if ((~sblock.fs_flags & FS_MULTILABEL) ==
|
|
|
|
FS_MULTILABEL) {
|
|
|
|
warnx("%s remains unchanged as disabled",
|
|
|
|
name);
|
|
|
|
} else {
|
|
|
|
sblock.fs_flags &= ~FS_MULTILABEL;
|
2002-10-16 05:03:40 +00:00
|
|
|
warnx("%s cleared", name);
|
2002-10-14 19:52:12 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2000-12-10 20:59:30 +00:00
|
|
|
if (mflag) {
|
|
|
|
name = "minimum percentage of free space";
|
2004-03-26 16:11:13 +00:00
|
|
|
if (sblock.fs_minfree == mvalue)
|
2000-12-10 20:59:30 +00:00
|
|
|
warnx("%s remains unchanged as %d%%", name, mvalue);
|
|
|
|
else {
|
|
|
|
warnx("%s changes from %d%% to %d%%",
|
|
|
|
name, sblock.fs_minfree, mvalue);
|
|
|
|
sblock.fs_minfree = mvalue;
|
|
|
|
if (mvalue >= MINFREE && sblock.fs_optim == FS_OPTSPACE)
|
|
|
|
warnx(OPTWARN, "time", ">=", MINFREE);
|
|
|
|
if (mvalue < MINFREE && sblock.fs_optim == FS_OPTTIME)
|
|
|
|
warnx(OPTWARN, "space", "<", MINFREE);
|
|
|
|
}
|
|
|
|
}
|
2009-12-21 19:39:10 +00:00
|
|
|
if (Nflag) {
|
|
|
|
name = "NFSv4 ACLs";
|
|
|
|
if (strcmp(Nvalue, "enable") == 0) {
|
|
|
|
if (sblock.fs_flags & FS_NFS4ACLS) {
|
|
|
|
warnx("%s remains unchanged as enabled", name);
|
|
|
|
} else if (sblock.fs_flags & FS_ACLS) {
|
|
|
|
warnx("%s and POSIX.1e ACLs are mutually "
|
|
|
|
"exclusive", name);
|
|
|
|
} else {
|
|
|
|
sblock.fs_flags |= FS_NFS4ACLS;
|
|
|
|
warnx("%s set", name);
|
|
|
|
}
|
|
|
|
} else if (strcmp(Nvalue, "disable") == 0) {
|
|
|
|
if ((~sblock.fs_flags & FS_NFS4ACLS) ==
|
|
|
|
FS_NFS4ACLS) {
|
|
|
|
warnx("%s remains unchanged as disabled",
|
|
|
|
name);
|
|
|
|
} else {
|
|
|
|
sblock.fs_flags &= ~FS_NFS4ACLS;
|
|
|
|
warnx("%s cleared", name);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2000-12-10 20:59:30 +00:00
|
|
|
if (nflag) {
|
|
|
|
name = "soft updates";
|
|
|
|
if (strcmp(nvalue, "enable") == 0) {
|
2004-03-26 16:11:13 +00:00
|
|
|
if (sblock.fs_flags & FS_DOSOFTDEP)
|
2000-12-10 20:59:30 +00:00
|
|
|
warnx("%s remains unchanged as enabled", name);
|
2004-03-26 16:11:13 +00:00
|
|
|
else if (sblock.fs_clean == 0) {
|
2001-04-13 23:54:49 +00:00
|
|
|
warnx("%s cannot be enabled until fsck is run",
|
|
|
|
name);
|
2000-12-10 20:59:30 +00:00
|
|
|
} else {
|
|
|
|
sblock.fs_flags |= FS_DOSOFTDEP;
|
|
|
|
warnx("%s set", name);
|
|
|
|
}
|
|
|
|
} else if (strcmp(nvalue, "disable") == 0) {
|
2004-03-26 16:11:13 +00:00
|
|
|
if ((~sblock.fs_flags & FS_DOSOFTDEP) == FS_DOSOFTDEP)
|
2000-12-10 20:59:30 +00:00
|
|
|
warnx("%s remains unchanged as disabled", name);
|
2004-03-26 16:11:13 +00:00
|
|
|
else {
|
2000-12-10 20:59:30 +00:00
|
|
|
sblock.fs_flags &= ~FS_DOSOFTDEP;
|
|
|
|
warnx("%s cleared", name);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (oflag) {
|
|
|
|
name = "optimization preference";
|
|
|
|
chg[FS_OPTSPACE] = "space";
|
|
|
|
chg[FS_OPTTIME] = "time";
|
2004-03-26 16:11:13 +00:00
|
|
|
if (sblock.fs_optim == ovalue)
|
2000-12-10 20:59:30 +00:00
|
|
|
warnx("%s remains unchanged as %s", name, chg[ovalue]);
|
|
|
|
else {
|
|
|
|
warnx("%s changes from %s to %s",
|
|
|
|
name, chg[sblock.fs_optim], chg[ovalue]);
|
|
|
|
sblock.fs_optim = ovalue;
|
|
|
|
if (sblock.fs_minfree >= MINFREE &&
|
2004-03-26 16:11:13 +00:00
|
|
|
ovalue == FS_OPTSPACE)
|
2000-12-10 20:59:30 +00:00
|
|
|
warnx(OPTWARN, "time", ">=", MINFREE);
|
2004-03-26 16:11:13 +00:00
|
|
|
if (sblock.fs_minfree < MINFREE && ovalue == FS_OPTTIME)
|
2000-12-10 20:59:30 +00:00
|
|
|
warnx(OPTWARN, "space", "<", MINFREE);
|
|
|
|
}
|
|
|
|
}
|
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>
2001-04-10 08:38:59 +00:00
|
|
|
if (sflag) {
|
|
|
|
name = "expected number of files per directory";
|
2010-02-11 06:33:35 +00:00
|
|
|
if (sblock.fs_avgfpdir == (unsigned)svalue) {
|
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>
2001-04-10 08:38:59 +00:00
|
|
|
warnx("%s remains unchanged as %d", name, svalue);
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
warnx("%s changes from %d to %d",
|
|
|
|
name, sblock.fs_avgfpdir, svalue);
|
|
|
|
sblock.fs_avgfpdir = svalue;
|
|
|
|
}
|
|
|
|
}
|
2010-12-29 12:31:18 +00:00
|
|
|
if (tflag) {
|
|
|
|
name = "issue TRIM to the disk";
|
|
|
|
if (strcmp(tvalue, "enable") == 0) {
|
|
|
|
if (sblock.fs_flags & FS_TRIM)
|
|
|
|
warnx("%s remains unchanged as enabled", name);
|
|
|
|
else {
|
|
|
|
sblock.fs_flags |= FS_TRIM;
|
|
|
|
warnx("%s set", name);
|
|
|
|
}
|
|
|
|
} else if (strcmp(tvalue, "disable") == 0) {
|
|
|
|
if ((~sblock.fs_flags & FS_TRIM) == FS_TRIM)
|
|
|
|
warnx("%s remains unchanged as disabled", name);
|
|
|
|
else {
|
|
|
|
sblock.fs_flags &= ~FS_TRIM;
|
|
|
|
warnx("%s cleared", name);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2000-12-10 20:59:30 +00:00
|
|
|
|
2003-01-28 02:42:01 +00:00
|
|
|
if (sbwrite(&disk, Aflag) == -1)
|
|
|
|
goto err;
|
|
|
|
ufs_disk_close(&disk);
|
1999-01-20 01:22:39 +00:00
|
|
|
if (active) {
|
|
|
|
bzero(&args, sizeof(args));
|
2003-01-28 02:42:01 +00:00
|
|
|
if (mount("ufs", on,
|
1999-01-20 01:22:39 +00:00
|
|
|
stfs.f_flags | MNT_UPDATE | MNT_RELOAD, &args) < 0)
|
|
|
|
err(9, "%s: reload", special);
|
2002-08-21 18:11:48 +00:00
|
|
|
warnx("file system reloaded");
|
1999-01-20 01:22:39 +00:00
|
|
|
}
|
1994-05-26 06:35:07 +00:00
|
|
|
exit(0);
|
2003-01-28 02:42:01 +00:00
|
|
|
err:
|
|
|
|
if (disk.d_error != NULL)
|
|
|
|
errx(11, "%s: %s", special, disk.d_error);
|
|
|
|
else
|
|
|
|
err(12, "%s", special);
|
1994-05-26 06:35:07 +00:00
|
|
|
}
|
|
|
|
|
2010-04-24 07:05:35 +00:00
|
|
|
void
|
|
|
|
sbdirty(void)
|
|
|
|
{
|
|
|
|
disk.d_fs.fs_flags |= FS_UNCLEAN | FS_NEEDSFSCK;
|
|
|
|
disk.d_fs.fs_clean = 0;
|
|
|
|
}
|
|
|
|
|
2011-11-04 13:36:02 +00:00
|
|
|
static int blocks;
|
2010-04-24 07:05:35 +00:00
|
|
|
static char clrbuf[MAXBSIZE];
|
|
|
|
|
|
|
|
static ufs2_daddr_t
|
|
|
|
journal_balloc(void)
|
|
|
|
{
|
|
|
|
ufs2_daddr_t blk;
|
|
|
|
struct cg *cgp;
|
|
|
|
int valid;
|
|
|
|
static int contig = 1;
|
|
|
|
|
|
|
|
cgp = &disk.d_cg;
|
|
|
|
for (;;) {
|
|
|
|
blk = cgballoc(&disk);
|
|
|
|
if (blk > 0)
|
|
|
|
break;
|
|
|
|
/*
|
|
|
|
* If we failed to allocate a block from this cg, move to
|
|
|
|
* the next.
|
|
|
|
*/
|
|
|
|
if (cgwrite(&disk) < 0) {
|
|
|
|
warn("Failed to write updated cg");
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
while ((valid = cgread(&disk)) == 1) {
|
|
|
|
/*
|
|
|
|
* Try to minimize fragmentation by requiring a minimum
|
|
|
|
* number of blocks present.
|
|
|
|
*/
|
2010-05-18 01:45:28 +00:00
|
|
|
if (cgp->cg_cs.cs_nbfree > 256 * 1024)
|
2010-04-24 07:05:35 +00:00
|
|
|
break;
|
|
|
|
if (contig == 0 && cgp->cg_cs.cs_nbfree)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (valid)
|
|
|
|
continue;
|
|
|
|
/*
|
|
|
|
* Try once through looking only for large contiguous regions
|
|
|
|
* and again taking any space we can find.
|
|
|
|
*/
|
|
|
|
if (contig) {
|
|
|
|
contig = 0;
|
|
|
|
disk.d_ccg = 0;
|
|
|
|
warnx("Journal file fragmented.");
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
warnx("Failed to find sufficient free blocks for the journal");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
if (bwrite(&disk, fsbtodb(&sblock, blk), clrbuf,
|
|
|
|
sblock.fs_bsize) <= 0) {
|
|
|
|
warn("Failed to initialize new block");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
return (blk);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Search a directory block for the SUJ_FILE.
|
|
|
|
*/
|
|
|
|
static ino_t
|
|
|
|
dir_search(ufs2_daddr_t blk, int bytes)
|
|
|
|
{
|
|
|
|
char block[MAXBSIZE];
|
|
|
|
struct direct *dp;
|
|
|
|
int off;
|
|
|
|
|
|
|
|
if (bread(&disk, fsbtodb(&sblock, blk), block, bytes) <= 0) {
|
|
|
|
warn("Failed to read dir block");
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
for (off = 0; off < bytes; off += dp->d_reclen) {
|
|
|
|
dp = (struct direct *)&block[off];
|
|
|
|
if (dp->d_reclen == 0)
|
|
|
|
break;
|
|
|
|
if (dp->d_ino == 0)
|
|
|
|
continue;
|
|
|
|
if (dp->d_namlen != strlen(SUJ_FILE))
|
|
|
|
continue;
|
|
|
|
if (bcmp(dp->d_name, SUJ_FILE, dp->d_namlen) != 0)
|
|
|
|
continue;
|
|
|
|
return (dp->d_ino);
|
|
|
|
}
|
|
|
|
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Search in the ROOTINO for the SUJ_FILE. If it exists we can not enable
|
|
|
|
* journaling.
|
|
|
|
*/
|
|
|
|
static ino_t
|
|
|
|
journal_findfile(void)
|
|
|
|
{
|
|
|
|
struct ufs1_dinode *dp1;
|
|
|
|
struct ufs2_dinode *dp2;
|
|
|
|
ino_t ino;
|
|
|
|
int mode;
|
|
|
|
void *ip;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
if (getino(&disk, &ip, ROOTINO, &mode) != 0) {
|
|
|
|
warn("Failed to get root inode");
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
dp2 = ip;
|
|
|
|
dp1 = ip;
|
|
|
|
if (sblock.fs_magic == FS_UFS1_MAGIC) {
|
|
|
|
if ((off_t)dp1->di_size >= lblktosize(&sblock, NDADDR)) {
|
|
|
|
warnx("ROOTINO extends beyond direct blocks.");
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
for (i = 0; i < NDADDR; i++) {
|
|
|
|
if (dp1->di_db[i] == 0)
|
|
|
|
break;
|
|
|
|
if ((ino = dir_search(dp1->di_db[i],
|
|
|
|
sblksize(&sblock, (off_t)dp1->di_size, i))) != 0)
|
|
|
|
return (ino);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if ((off_t)dp1->di_size >= lblktosize(&sblock, NDADDR)) {
|
|
|
|
warnx("ROOTINO extends beyond direct blocks.");
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
for (i = 0; i < NDADDR; i++) {
|
|
|
|
if (dp2->di_db[i] == 0)
|
|
|
|
break;
|
|
|
|
if ((ino = dir_search(dp2->di_db[i],
|
|
|
|
sblksize(&sblock, (off_t)dp2->di_size, i))) != 0)
|
|
|
|
return (ino);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
2011-02-12 13:12:45 +00:00
|
|
|
static void
|
|
|
|
dir_clear_block(char *block, off_t off)
|
|
|
|
{
|
|
|
|
struct direct *dp;
|
|
|
|
|
|
|
|
for (; off < sblock.fs_bsize; off += DIRBLKSIZ) {
|
|
|
|
dp = (struct direct *)&block[off];
|
|
|
|
dp->d_ino = 0;
|
|
|
|
dp->d_reclen = DIRBLKSIZ;
|
|
|
|
dp->d_type = DT_UNKNOWN;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-04-24 07:05:35 +00:00
|
|
|
/*
|
|
|
|
* Insert the journal at inode 'ino' into directory blk 'blk' at the first
|
|
|
|
* free offset of 'off'. DIRBLKSIZ blocks after off are initialized as
|
|
|
|
* empty.
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
dir_insert(ufs2_daddr_t blk, off_t off, ino_t ino)
|
|
|
|
{
|
|
|
|
struct direct *dp;
|
|
|
|
char block[MAXBSIZE];
|
|
|
|
|
|
|
|
if (bread(&disk, fsbtodb(&sblock, blk), block, sblock.fs_bsize) <= 0) {
|
|
|
|
warn("Failed to read dir block");
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
bzero(&block[off], sblock.fs_bsize - off);
|
|
|
|
dp = (struct direct *)&block[off];
|
|
|
|
dp->d_ino = ino;
|
|
|
|
dp->d_reclen = DIRBLKSIZ;
|
|
|
|
dp->d_type = DT_REG;
|
|
|
|
dp->d_namlen = strlen(SUJ_FILE);
|
|
|
|
bcopy(SUJ_FILE, &dp->d_name, strlen(SUJ_FILE));
|
2011-02-12 13:12:45 +00:00
|
|
|
dir_clear_block(block, off + DIRBLKSIZ);
|
2010-04-24 07:05:35 +00:00
|
|
|
if (bwrite(&disk, fsbtodb(&sblock, blk), block, sblock.fs_bsize) <= 0) {
|
|
|
|
warn("Failed to write dir block");
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Extend a directory block in 'blk' by copying it to a full size block
|
|
|
|
* and inserting the new journal inode into .sujournal.
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
dir_extend(ufs2_daddr_t blk, ufs2_daddr_t nblk, off_t size, ino_t ino)
|
|
|
|
{
|
|
|
|
char block[MAXBSIZE];
|
|
|
|
|
2011-02-12 13:12:45 +00:00
|
|
|
if (bread(&disk, fsbtodb(&sblock, blk), block,
|
|
|
|
roundup(size, sblock.fs_fsize)) <= 0) {
|
2010-04-24 07:05:35 +00:00
|
|
|
warn("Failed to read dir block");
|
|
|
|
return (-1);
|
|
|
|
}
|
2011-02-12 13:12:45 +00:00
|
|
|
dir_clear_block(block, size);
|
|
|
|
if (bwrite(&disk, fsbtodb(&sblock, nblk), block, sblock.fs_bsize)
|
|
|
|
<= 0) {
|
2010-04-24 07:05:35 +00:00
|
|
|
warn("Failed to write dir block");
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
|
2011-02-12 13:12:45 +00:00
|
|
|
return (dir_insert(nblk, size, ino));
|
2010-04-24 07:05:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Insert the journal file into the ROOTINO directory. We always extend the
|
|
|
|
* last frag
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
journal_insertfile(ino_t ino)
|
|
|
|
{
|
|
|
|
struct ufs1_dinode *dp1;
|
|
|
|
struct ufs2_dinode *dp2;
|
|
|
|
void *ip;
|
|
|
|
ufs2_daddr_t nblk;
|
|
|
|
ufs2_daddr_t blk;
|
|
|
|
ufs_lbn_t lbn;
|
|
|
|
int size;
|
|
|
|
int mode;
|
|
|
|
int off;
|
|
|
|
|
|
|
|
if (getino(&disk, &ip, ROOTINO, &mode) != 0) {
|
|
|
|
warn("Failed to get root inode");
|
|
|
|
sbdirty();
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
dp2 = ip;
|
|
|
|
dp1 = ip;
|
|
|
|
blk = 0;
|
|
|
|
size = 0;
|
|
|
|
nblk = journal_balloc();
|
|
|
|
if (nblk <= 0)
|
|
|
|
return (-1);
|
|
|
|
/*
|
|
|
|
* For simplicity sake we aways extend the ROOTINO into a new
|
|
|
|
* directory block rather than searching for space and inserting
|
|
|
|
* into an existing block. However, if the rootino has frags
|
|
|
|
* have to free them and extend the block.
|
|
|
|
*/
|
|
|
|
if (sblock.fs_magic == FS_UFS1_MAGIC) {
|
|
|
|
lbn = lblkno(&sblock, dp1->di_size);
|
|
|
|
off = blkoff(&sblock, dp1->di_size);
|
|
|
|
blk = dp1->di_db[lbn];
|
|
|
|
size = sblksize(&sblock, (off_t)dp1->di_size, lbn);
|
|
|
|
} else {
|
|
|
|
lbn = lblkno(&sblock, dp2->di_size);
|
|
|
|
off = blkoff(&sblock, dp2->di_size);
|
|
|
|
blk = dp2->di_db[lbn];
|
|
|
|
size = sblksize(&sblock, (off_t)dp2->di_size, lbn);
|
|
|
|
}
|
|
|
|
if (off != 0) {
|
|
|
|
if (dir_extend(blk, nblk, off, ino) == -1)
|
|
|
|
return (-1);
|
|
|
|
} else {
|
|
|
|
blk = 0;
|
|
|
|
if (dir_insert(nblk, 0, ino) == -1)
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
if (sblock.fs_magic == FS_UFS1_MAGIC) {
|
|
|
|
dp1->di_blocks += (sblock.fs_bsize - size) / DEV_BSIZE;
|
|
|
|
dp1->di_db[lbn] = nblk;
|
|
|
|
dp1->di_size = lblktosize(&sblock, lbn+1);
|
|
|
|
} else {
|
|
|
|
dp2->di_blocks += (sblock.fs_bsize - size) / DEV_BSIZE;
|
|
|
|
dp2->di_db[lbn] = nblk;
|
|
|
|
dp2->di_size = lblktosize(&sblock, lbn+1);
|
|
|
|
}
|
|
|
|
if (putino(&disk) < 0) {
|
|
|
|
warn("Failed to write root inode");
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
if (cgwrite(&disk) < 0) {
|
|
|
|
warn("Failed to write updated cg");
|
|
|
|
sbdirty();
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
if (blk) {
|
|
|
|
if (cgbfree(&disk, blk, size) < 0) {
|
|
|
|
warn("Failed to write cg");
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
indir_fill(ufs2_daddr_t blk, int level, int *resid)
|
|
|
|
{
|
|
|
|
char indirbuf[MAXBSIZE];
|
|
|
|
ufs1_daddr_t *bap1;
|
|
|
|
ufs2_daddr_t *bap2;
|
|
|
|
ufs2_daddr_t nblk;
|
|
|
|
int ncnt;
|
|
|
|
int cnt;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
bzero(indirbuf, sizeof(indirbuf));
|
|
|
|
bap1 = (ufs1_daddr_t *)indirbuf;
|
|
|
|
bap2 = (void *)bap1;
|
|
|
|
cnt = 0;
|
|
|
|
for (i = 0; i < NINDIR(&sblock) && *resid != 0; i++) {
|
|
|
|
nblk = journal_balloc();
|
|
|
|
if (nblk <= 0)
|
|
|
|
return (-1);
|
|
|
|
cnt++;
|
|
|
|
if (sblock.fs_magic == FS_UFS1_MAGIC)
|
|
|
|
*bap1++ = nblk;
|
|
|
|
else
|
|
|
|
*bap2++ = nblk;
|
|
|
|
if (level != 0) {
|
|
|
|
ncnt = indir_fill(nblk, level - 1, resid);
|
|
|
|
if (ncnt <= 0)
|
|
|
|
return (-1);
|
|
|
|
cnt += ncnt;
|
|
|
|
} else
|
|
|
|
(*resid)--;
|
|
|
|
}
|
|
|
|
if (bwrite(&disk, fsbtodb(&sblock, blk), indirbuf,
|
|
|
|
sblock.fs_bsize) <= 0) {
|
|
|
|
warn("Failed to write indirect");
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
return (cnt);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Clear the flag bits so the journal can be removed.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
journal_clear(void)
|
|
|
|
{
|
|
|
|
struct ufs1_dinode *dp1;
|
|
|
|
struct ufs2_dinode *dp2;
|
|
|
|
ino_t ino;
|
|
|
|
int mode;
|
|
|
|
void *ip;
|
|
|
|
|
|
|
|
ino = journal_findfile();
|
|
|
|
if (ino == (ino_t)-1 || ino == 0) {
|
|
|
|
warnx("Journal file does not exist");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
printf("Clearing journal flags from inode %d\n", ino);
|
|
|
|
if (getino(&disk, &ip, ino, &mode) != 0) {
|
|
|
|
warn("Failed to get journal inode");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
dp2 = ip;
|
|
|
|
dp1 = ip;
|
|
|
|
if (sblock.fs_magic == FS_UFS1_MAGIC)
|
|
|
|
dp1->di_flags = 0;
|
|
|
|
else
|
|
|
|
dp2->di_flags = 0;
|
|
|
|
if (putino(&disk) < 0) {
|
|
|
|
warn("Failed to write journal inode");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
int
|
|
|
|
journal_alloc(int64_t size)
|
|
|
|
{
|
|
|
|
struct ufs1_dinode *dp1;
|
|
|
|
struct ufs2_dinode *dp2;
|
|
|
|
ufs2_daddr_t blk;
|
|
|
|
void *ip;
|
|
|
|
struct cg *cgp;
|
|
|
|
int resid;
|
|
|
|
ino_t ino;
|
|
|
|
int blks;
|
|
|
|
int mode;
|
2011-10-11 19:03:57 +00:00
|
|
|
time_t utime;
|
2010-04-24 07:05:35 +00:00
|
|
|
int i;
|
|
|
|
|
|
|
|
cgp = &disk.d_cg;
|
|
|
|
ino = 0;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If the journal file exists we can't allocate it.
|
|
|
|
*/
|
|
|
|
ino = journal_findfile();
|
|
|
|
if (ino == (ino_t)-1)
|
|
|
|
return (-1);
|
|
|
|
if (ino > 0) {
|
|
|
|
warnx("Journal file %s already exists, please remove.",
|
|
|
|
SUJ_FILE);
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* If the user didn't supply a size pick one based on the filesystem
|
|
|
|
* size constrained with hardcoded MIN and MAX values. We opt for
|
|
|
|
* 1/1024th of the filesystem up to MAX but not exceeding one CG and
|
|
|
|
* not less than the MIN.
|
|
|
|
*/
|
|
|
|
if (size == 0) {
|
|
|
|
size = (sblock.fs_size * sblock.fs_bsize) / 1024;
|
|
|
|
size = MIN(SUJ_MAX, size);
|
|
|
|
if (size / sblock.fs_fsize > sblock.fs_fpg)
|
|
|
|
size = sblock.fs_fpg * sblock.fs_fsize;
|
|
|
|
size = MAX(SUJ_MIN, size);
|
2010-05-18 01:45:28 +00:00
|
|
|
/* fsck does not support fragments in journal files. */
|
|
|
|
size = roundup(size, sblock.fs_bsize);
|
2010-04-24 07:05:35 +00:00
|
|
|
}
|
|
|
|
resid = blocks = size / sblock.fs_bsize;
|
|
|
|
if (sblock.fs_cstotal.cs_nbfree < blocks) {
|
|
|
|
warn("Insufficient free space for %jd byte journal", size);
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* Find a cg with enough blocks to satisfy the journal
|
|
|
|
* size. Presently the journal does not span cgs.
|
|
|
|
*/
|
|
|
|
while (cgread(&disk) == 1) {
|
|
|
|
if (cgp->cg_cs.cs_nifree == 0)
|
|
|
|
continue;
|
|
|
|
ino = cgialloc(&disk);
|
|
|
|
if (ino <= 0)
|
|
|
|
break;
|
|
|
|
printf("Using inode %d in cg %d for %jd byte journal\n",
|
|
|
|
ino, cgp->cg_cgx, size);
|
|
|
|
if (getino(&disk, &ip, ino, &mode) != 0) {
|
|
|
|
warn("Failed to get allocated inode");
|
|
|
|
sbdirty();
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* We leave fields unrelated to the number of allocated
|
|
|
|
* blocks and size uninitialized. This causes legacy
|
|
|
|
* fsck implementations to clear the inode.
|
|
|
|
*/
|
|
|
|
dp2 = ip;
|
|
|
|
dp1 = ip;
|
2011-10-11 19:03:57 +00:00
|
|
|
time(&utime);
|
2010-04-24 07:05:35 +00:00
|
|
|
if (sblock.fs_magic == FS_UFS1_MAGIC) {
|
|
|
|
bzero(dp1, sizeof(*dp1));
|
|
|
|
dp1->di_size = size;
|
|
|
|
dp1->di_mode = IFREG | IREAD;
|
|
|
|
dp1->di_nlink = 1;
|
|
|
|
dp1->di_flags = SF_IMMUTABLE | SF_NOUNLINK | UF_NODUMP;
|
2011-10-11 19:03:57 +00:00
|
|
|
dp1->di_atime = utime;
|
|
|
|
dp1->di_mtime = utime;
|
|
|
|
dp1->di_ctime = utime;
|
2010-04-24 07:05:35 +00:00
|
|
|
} else {
|
|
|
|
bzero(dp2, sizeof(*dp2));
|
|
|
|
dp2->di_size = size;
|
|
|
|
dp2->di_mode = IFREG | IREAD;
|
|
|
|
dp2->di_nlink = 1;
|
|
|
|
dp2->di_flags = SF_IMMUTABLE | SF_NOUNLINK | UF_NODUMP;
|
2011-10-11 19:03:57 +00:00
|
|
|
dp2->di_atime = utime;
|
|
|
|
dp2->di_mtime = utime;
|
|
|
|
dp2->di_ctime = utime;
|
|
|
|
dp2->di_birthtime = utime;
|
2010-04-24 07:05:35 +00:00
|
|
|
}
|
|
|
|
for (i = 0; i < NDADDR && resid; i++, resid--) {
|
|
|
|
blk = journal_balloc();
|
|
|
|
if (blk <= 0)
|
|
|
|
goto out;
|
|
|
|
if (sblock.fs_magic == FS_UFS1_MAGIC) {
|
|
|
|
dp1->di_db[i] = blk;
|
|
|
|
dp1->di_blocks++;
|
|
|
|
} else {
|
|
|
|
dp2->di_db[i] = blk;
|
|
|
|
dp2->di_blocks++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
for (i = 0; i < NIADDR && resid; i++) {
|
|
|
|
blk = journal_balloc();
|
|
|
|
if (blk <= 0)
|
|
|
|
goto out;
|
|
|
|
blks = indir_fill(blk, i, &resid) + 1;
|
|
|
|
if (blks <= 0) {
|
|
|
|
sbdirty();
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (sblock.fs_magic == FS_UFS1_MAGIC) {
|
|
|
|
dp1->di_ib[i] = blk;
|
|
|
|
dp1->di_blocks += blks;
|
|
|
|
} else {
|
|
|
|
dp2->di_ib[i] = blk;
|
|
|
|
dp2->di_blocks += blks;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (sblock.fs_magic == FS_UFS1_MAGIC)
|
|
|
|
dp1->di_blocks *= sblock.fs_bsize / disk.d_bsize;
|
|
|
|
else
|
|
|
|
dp2->di_blocks *= sblock.fs_bsize / disk.d_bsize;
|
|
|
|
if (putino(&disk) < 0) {
|
|
|
|
warn("Failed to write inode");
|
|
|
|
sbdirty();
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
if (cgwrite(&disk) < 0) {
|
|
|
|
warn("Failed to write updated cg");
|
|
|
|
sbdirty();
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
if (journal_insertfile(ino) < 0) {
|
|
|
|
sbdirty();
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
sblock.fs_sujfree = 0;
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
warnx("Insufficient free space for the journal.");
|
|
|
|
out:
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
|
1994-05-26 06:35:07 +00:00
|
|
|
void
|
2003-01-20 21:15:02 +00:00
|
|
|
usage(void)
|
1994-05-26 06:35:07 +00:00
|
|
|
{
|
2010-12-29 12:31:18 +00:00
|
|
|
fprintf(stderr, "%s\n%s\n%s\n%s\n%s\n%s\n",
|
2003-02-23 01:50:07 +00:00
|
|
|
"usage: tunefs [-A] [-a enable | disable] [-e maxbpg] [-f avgfilesize]",
|
2010-05-01 09:05:06 +00:00
|
|
|
" [-J enable | disable] [-j enable | disable]",
|
|
|
|
" [-L volname] [-l enable | disable] [-m minfree]",
|
|
|
|
" [-N enable | disable] [-n enable | disable]",
|
2010-12-29 12:31:18 +00:00
|
|
|
" [-o space | time] [-p] [-s avgfpdir] [-t enable | disable]",
|
|
|
|
" special | filesystem");
|
1994-05-26 06:35:07 +00:00
|
|
|
exit(2);
|
|
|
|
}
|
|
|
|
|
1995-06-25 17:46:13 +00:00
|
|
|
void
|
2003-01-20 21:15:02 +00:00
|
|
|
printfs(void)
|
1995-06-25 17:46:13 +00:00
|
|
|
{
|
2009-12-21 19:39:10 +00:00
|
|
|
warnx("POSIX.1e ACLs: (-a) %s",
|
2002-10-15 18:14:05 +00:00
|
|
|
(sblock.fs_flags & FS_ACLS)? "enabled" : "disabled");
|
2009-12-21 19:39:10 +00:00
|
|
|
warnx("NFSv4 ACLs: (-N) %s",
|
|
|
|
(sblock.fs_flags & FS_NFS4ACLS)? "enabled" : "disabled");
|
2002-10-15 18:14:05 +00:00
|
|
|
warnx("MAC multilabel: (-l) %s",
|
|
|
|
(sblock.fs_flags & FS_MULTILABEL)? "enabled" : "disabled");
|
2003-01-18 06:29:15 +00:00
|
|
|
warnx("soft updates: (-n) %s",
|
1998-03-08 09:59:44 +00:00
|
|
|
(sblock.fs_flags & FS_DOSOFTDEP)? "enabled" : "disabled");
|
2010-04-24 07:05:35 +00:00
|
|
|
warnx("soft update journaling: (-j) %s",
|
|
|
|
(sblock.fs_flags & FS_SUJ)? "enabled" : "disabled");
|
2006-10-31 21:52:28 +00:00
|
|
|
warnx("gjournal: (-J) %s",
|
|
|
|
(sblock.fs_flags & FS_GJOURNAL)? "enabled" : "disabled");
|
2010-12-29 12:31:18 +00:00
|
|
|
warnx("trim: (-t) %s",
|
|
|
|
(sblock.fs_flags & FS_TRIM)? "enabled" : "disabled");
|
1995-06-25 17:46:13 +00:00
|
|
|
warnx("maximum blocks per file in a cylinder group: (-e) %d",
|
|
|
|
sblock.fs_maxbpg);
|
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>
2001-04-10 08:38:59 +00:00
|
|
|
warnx("average file size: (-f) %d",
|
|
|
|
sblock.fs_avgfilesize);
|
|
|
|
warnx("average number of files in a directory: (-s) %d",
|
|
|
|
sblock.fs_avgfpdir);
|
1995-06-25 17:46:13 +00:00
|
|
|
warnx("minimum percentage of free space: (-m) %d%%",
|
|
|
|
sblock.fs_minfree);
|
|
|
|
warnx("optimization preference: (-o) %s",
|
|
|
|
sblock.fs_optim == FS_OPTSPACE ? "space" : "time");
|
|
|
|
if (sblock.fs_minfree >= MINFREE &&
|
|
|
|
sblock.fs_optim == FS_OPTSPACE)
|
|
|
|
warnx(OPTWARN, "time", ">=", MINFREE);
|
|
|
|
if (sblock.fs_minfree < MINFREE &&
|
|
|
|
sblock.fs_optim == FS_OPTTIME)
|
|
|
|
warnx(OPTWARN, "space", "<", MINFREE);
|
2003-02-01 04:17:10 +00:00
|
|
|
warnx("volume label: (-L) %s",
|
|
|
|
sblock.fs_volname);
|
1995-06-25 17:46:13 +00:00
|
|
|
}
|