Wednesday, August 19, 2009

[Level 2] ZFS and DTrace in Mac.

My friend who use Mac OS X, and told me that the Mac OS X support ZFS and DTrace. It surprises me! ZFS and DTrace is the new feature of Solaris 10, of course, OpenSolaris also has these two features, too. So I try the ZFS and DTrace on Mac immediately.
After test, I have the two following conclusions.
1. Mac support ZFS but for read only, so Mac cannot modify ZFS. That's so pity. The primary reason that I love ZFS just because I can have snapshot, clone. Readonly, is not the impression that I know about ZFS.
2. In OpenSolaris, it has over 60K probes(use command to check how many probes your system has, "/usr/sbin/dtrace -l | wc -l"), so I can detect the system behavior. But when I use the command, I found out the Mac OS only have just over 20K probes, just the 1/3 of OpenSolaris. And I try the D-script, it does work. That's a good gift that all Mac fans should to cherish. Really, because DTrace can be used for online debug, performance and you do not need to modify the source ( Sometimes, even we don't have!! ).
The following sample code is the D-script that exists in OpenSolaris, and to let Mac can run it, I modify the probe name from "open64" to "open" in the script. That you can try it on your Mac. Please save the source code on /tmp/opensnoop and use the following command to detect what processes are opening the file.

# /tmp/opensnoop


********************************************************************************************************


#!/usr/bin/sh
#
# opensnoop - snoop file opens as they occur.
# Written using DTrace (Solaris 10 3/05).
#
# $Id: opensnoop 3 2007-08-01 10:50:08Z brendan $
#
# USAGE: opensnoop [-a|-A|-ceghsvxZ] [-f pathname] [-n name] [-p PID]
#
# opensnoop # default output
#
# -a # print most data
# -A # dump all data, space delimited
# -c # print cwd of process
# -e # print errno value
# -g # print command arguments
# -s # print start time, us
# -v # print start time, string
# -x # only print failed opens
# -Z # print zonename
# -f pathname # file pathname to snoop
# -n name # command name to snoop
# -p PID # process ID to snoop
# eg,
# opensnoop -v # human readable timestamps
# opensnoop -e # see error codes
# opensnoop -f /etc/passwd # snoop this file only
#
# FIELDS:
# ZONE Zone name
# UID User ID
# PID Process ID
# PPID Parent Process ID
# FD file descriptor (-1 for error)
# ERR errno value (see /usr/include/sys/errno.h)
# CWD print current working directory of process
# PATH pathname for file open
# COMM command name for the process
# ARGS argument listing for the process
# TIME timestamp for the open event, us
# STRTIME timestamp for the open event, string
#
# SEE ALSO: truss, BSM auditing.
#
# COPYRIGHT: Copyright (c) 2006 Brendan Gregg.
#
# CDDL HEADER START
#
# The contents of this file are subject to the terms of the
# Common Development and Distribution License, Version 1.0 only
# (the "License"). You may not use this file except in compliance
# with the License.
#
# You can obtain a copy of the license at Docs/cddl1.txt
# or http://www.opensolaris.org/os/licensing.
# See the License for the specific language governing permissions
# and limitations under the License.
#
# CDDL HEADER END
#
# Author: Brendan Gregg [Sydney, Australia]
#
# 09-May-2004 Brendan Gregg Created this.
# 21-Jan-2005 " " Wrapped in sh to provide options.
# 08-May-2005 " " Rewritten for performance.
# 14-May-2005 " " Added errno.
# 28-Jun-2005 " " Added cwd, zonename.
# 17-Sep-2005 " " Increased switchrate, fixed page fault bug.
# 16-Jan-2006 " " Added -n, -p.
# 16-Jan-2006 " " Last update.
#




##############################
# --- Process Arguments ---
#


### Default variables
opt_dump=0; opt_file=0; opt_time=0; opt_timestr=0; opt_args=0
opt_zone=0; opt_cwd=0; opt_failonly=0; opt_err=0; filter=0; pathname=.
opt_name=0; opt_pid=0; pname=.; pid=0


### Process options
while getopts aAcef:ghn:p:svxZ name
do
case $name in
a) opt_time=1; opt_timestr=1; opt_args=1; opt_err=1 ;;
A) opt_dump=1 ;;
c) opt_cwd=1 ;;
e) opt_err=1 ;;
g) opt_args=1 ;;
f) opt_file=1; pathname=$OPTARG ;;
n) opt_name=1; pname=$OPTARG ;;
p) opt_pid=1; pid=$OPTARG ;;
s) opt_time=1 ;;
v) opt_timestr=1 ;;
x) opt_failonly=1 ;;
Z) opt_zone=1 ;;
h|?) cat <<-END >&2
USAGE: opensnoop [-a|-A|-ceghsvxZ] [-f pathname]
[-n name] [-p PID]
opensnoop # default output
-a # print most data
-A # dump all data, space delimited
-c # print cwd of process
-e # print errno value
-g # print command arguments
-s # print start time, us
-v # print start time, string
-x # only print failed opens
-Z # print zonename
-f pathname # pathname name to snoop
-n name # process name to snoop
-p PID # process ID to snoop
eg,
opensnoop -v # human readable timestamps
opensnoop -e # see error codes
opensnoop -f /etc/motd # snoop this file only
END
exit 1
esac
done


### Option logic
if [ $opt_dump -eq 1 ]; then
opt_zone=0; opt_cwd=0; opt_time=0; opt_timestr=0; opt_args=2
fi
if [ $opt_name -eq 1 -o $opt_pid -eq 1 ]; then
filter=1
fi




#################################
# --- Main Program, DTrace ---
#
/usr/sbin/dtrace -n '
/*
* Command line arguments
*/
inline int OPT_dump = '$opt_dump';
inline int OPT_file = '$opt_file';
inline int OPT_args = '$opt_args';
inline int OPT_cwd = '$opt_cwd';
inline int OPT_err = '$opt_err';
inline int OPT_zone = '$opt_zone';
inline int OPT_time = '$opt_time';
inline int OPT_timestr = '$opt_timestr';
inline int OPT_failonly = '$opt_failonly';
inline int OPT_pid = '$opt_pid';
inline int OPT_name = '$opt_name';
inline int FILTER = '$filter';
inline int PID = '$pid';
inline string PATHNAME = "'$pathname'";
inline string NAME = "'$pname'";


#pragma D option quiet
#pragma D option switchrate=10hz


/*
* Print header
*/
dtrace:::BEGIN
{
/*
* ternary operators are used to improve performance.
* OPT_args is unusual in that it can have one of three values.
*/


/* print optional headers */
OPT_time ? printf("%-14s ", "TIME") : 1;
OPT_timestr ? printf("%-20s ", "STRTIME") : 1;
OPT_zone ? printf("%-10s ", "ZONE") : 1;


/* print dump headers */
OPT_dump ? printf("%s %s %s %s %s %s %s %s %s %s %s", "ZONE",
"TIME", "UID", "PID", "PPID", "COMM", "FD", "ERR", "CWD",
"PATH", "ARGS") : printf("%5s %6s ","UID","PID");


/* print main headers */
OPT_args == 0 ? printf("%-12s ", "COMM") : 1;
OPT_dump == 0 ? printf("%3s ", "FD") : 1;
OPT_err ? printf("%3s ", "ERR") : 1;
OPT_cwd ? printf("%-20s ", "CWD") : 1;
OPT_dump == 0 ? printf("%-20s ", "PATH") : 1;
OPT_args == 1 ? printf("%s", "ARGS") : 1;
printf("\n");
}


/*
* Print open event
*/
syscall::open:entry, syscall::open:entry
{
/* save pathname */
self->pathp = arg0;


/* default is to trace unless filtering */
self->ok = FILTER ? 0 : 1;


/* check each filter */
(OPT_name == 1 && NAME == execname) ? self->ok = 1 : 1;
(OPT_pid == 1 && PID == pid) ? self->ok = 1 : 1;
/* OPT_file is checked on return to ensure pathp is mapped */
}


syscall::open:return, syscall::open:return
/self->ok && (! OPT_failonly || (int)arg0 < 0) && ((OPT_file == 0) || (OPT_file == 1 && PATHNAME == copyinstr(self->pathp)))/
{
/* print optional fields */
OPT_time ? printf("%-14d ", timestamp/1000) : 1;
OPT_timestr ? printf("%-20Y ", walltimestamp) : 1;
OPT_zone ? printf("%-10s ", zonename) : 1;


/* print dump fields */
OPT_dump ? printf("%s %d %d %d %d %s %d %d %s %s %S", zonename,
timestamp/1000, uid, pid, ppid, execname, (int)arg0, errno,
cwd, copyinstr(self->pathp), curpsinfo->pr_psargs) :
printf("%5d %6d ", uid, pid);


/* print main fields */
OPT_args == 0 ? printf("%-12s ", execname) : 1;
OPT_dump == 0 ? printf("%3d ", (int)arg0) : 1;
OPT_err ? printf("%3d ", errno) : 1;
OPT_cwd ? printf("%-20s ", cwd) : 1;
OPT_dump == 0 ? printf("%-20s ", copyinstr(self->pathp)) : 1;
OPT_args == 1 ? printf("%S", curpsinfo->pr_psargs) : 1;
printf("\n");


/* cleanup */
self->pathp = 0;
self->ok = 0;
}


/*
* Cleanup
*/
syscall::open:return, syscall::open:return
/self->ok/
{
self->pathp = 0;
self->ok = 0;
}
'