Wednesday, February 24, 2010

Monitoring File Changes on Linux

This is a copy of article from (Copyright Anthony Lawrence, used by permission) In case the original link dead, I can still look at it.

Many modern systems provide a way to watch a directory for events (new files, reading the directory, modification of a file in the directory, etc.). This facility can be done in various ways, from providing hooks in the filesystem code itself to something that watches for inode changes. Linux and BSD have several possibilities in that regard, including dnotify ( replaced by inotify), changedfiles, watch, Gamin and fam.

Mac OS X has FileMonitor (shareware), FSEvents and it's certainly possible that some of the BSD tools might compile on OS X. But what do you do when there is no support in the OS?

Brute Force

Typically you are interested in new files in a particular directory. You can do something like this:
touch /tmp/testdirb.$$
while true
ls /testdir > /tmp/testdira.$$
diff /tmp/testdira.$$ /tmp/testdirb.$$ || echo "changed"
cp /tmp/testdira.$$ /tmp/testdirb.$$
sleep 300
If you were looking for a particular file to change, you'd use "ls -l", and if your interest was in if the file was being used or executed, "ls -lut" would give you that. This example justs echoes when something changes, but you would more likely call some other script that did more testing. One obvious issue that comes up if a file has been created is waiting for the creating program to have finished up: lsof or fuser can help you with that.

But this is all pretty crude. Sometimes crude is fine, but if you need to know more often, there's a fair amount of overhead in this that you really don't need.

If it is just one file, and your interest is additions to it, the mailchecking utility in your shell can give you alerts. For example, in Linux bash:
You can watch multiple files, each with its own message, by using MAILPATH instead of MAIL. You need to "unset MAIL", and set and export MAILPATH:
unset MAIL
MAILPATH='/tmp/foo/h?"H is changed":/tmp/foo/a?"A has changed"'
export MAILPATH 

Linux systems have a command line "stat" that can make checking changes a little less intensive. "stat -t testdir" looks something like this:
testdir 4096 8 41ed 5001 5001 303 82022 2 2b 1d 1070127877 1070128608 1070128608
which is everything you want to know about the file or directory for this purpose. Security Enhanced Linux versions add another field which we need to strip out for our purposes here:
OLD=`stat -t testdir`
# OLD=`stat -t testdir | sed 's/[0-9][0-9]*$//'`  # if SE linux stat
while true
NEW=`stat -t testdir`
# NEW=`stat -t testdir | sed 's/[0-9][0-9]*$//'`  if SE linux stat
[ "$NEW" == "$OLD" ] || echo "changed!"
sleep 3
You could get fancier by splitting out the fields into separate variables. That isn't a lot of fun at the shell level, so we'll move up a notch.

Perl or C

From Perl or C (and of course many other languages), you can get access to the stat information a bit more easily. Here's a simple Perl example like those already given:
@info=stat("testdir") or die "Can't stat testdir $!";
while (1) {
@newinfo=stat("testdir") or die "Can't stat testdir $!";
@what=qw(Device Inum Mode Links Owner Group Rdev Size Atime Mtime Ctime PBlock Blocks);
while ($info[$x]) {
  system("/bin/echo $what[$x] $info[$x] $newinfo[$x]") if ($info[$x] ne $newinfo[$x]);
sleep 1;

No comments: