Pages

Tuesday, February 2, 2010

Episode #80: Time Bandits

Tim stomps in:

I have always wanted to time travel. Since it isn't possible to go back and kill Hitler I thought, maybe we can go back in time and change some files. Obviously, we don't have the technology to actually go back in time and make changes. However, what if we could make it appear that we went back in time by altering timestamps.

First, let's create a few files for our time warp. By the way, this next set of commands are all functionally equivalent.

PS C:\> Write-Output aaaa | Out-File a.txt
PS C:\> Write bbbb | Out-File b.txt
PS C:\> echo cccc | Out-File c.txt
PS C:\> echo dddd > d.txt


Now to see the time related properties available to us for the file object.

PS C:\> Get-ChildItem | Get-Member -MemberType Property | Where-Object { $_.Name -like "*time*" }

TypeName: System.IO.FileInfo

Name MemberType Definition
---- ---------- ----------
CreationTime Property System.DateTime CreationTime {get;set;}
CreationTimeUtc Property System.DateTime CreationTimeUtc {get;set;}
LastAccessTime Property System.DateTime LastAccessTime {get;set;}
LastAccessTimeUtc Property System.DateTime LastAccessTimeUtc {get;set;}
LastWriteTime Property System.DateTime LastWriteTime {get;set;}
LastWriteTimeUtc Property System.DateTime LastWriteTimeUtc {get;set;}


The Get-Member cmdlet (alias gm) is used to get the properties and methods available for an object that has been sent down the pipeline. In our case, the object sent down the pipeline is the file object. We just want to look at the properties (not methods, scriptproperties, etc) of the object, so we use the MemberType parameter for filtering. Then the Where-Object cmdlet (alias ?) is used to filter for properties with "time" in the name. The properties above are read/write, as shown by {get;set;}. Well, lookey there, we can set the timestamps!

An important side note: The Get-Member cmdlet is an extremely useful command. I can't begin to say how often I've used this command to find properties and methods available for an object. There are other MemberTypes, but we will have to cover those at a later time. For a full description check out the MemberType parameter on this Microsoft help page.

The xxxxxTime and xxxxxTimeUtc properties are actually the same property, the only difference is how they display the date in regards to the UTC (coordinated universal time) offset. In my case, the difference is 6 hours. For the sake of brevity the UTC times will be ignored since they are essentially redundant.

Let's take a look at our files.

PS C:\> gci | select Name, LastWriteTime, CreationTime, LastAccessTime

Name LastWriteTime CreationTime LastAccessTime
---- ------------- ------------ --------------
a.txt 1/23/2010 11:16:24 AM 1/23/2010 11:16:24 AM 1/23/2010 11:16:24 AM
b.txt 1/23/2010 11:16:24 AM 1/23/2010 11:16:24 AM 1/23/2010 11:16:24 AM
c.txt 1/23/2010 11:16:24 AM 1/23/2010 11:16:24 AM 1/23/2010 11:16:24 AM
d.txt 1/23/2010 11:16:24 AM 1/23/2010 11:16:24 AM 1/23/2010 11:16:24 AM


Now let's go back in time.

(gci a.txt).LastAccessTime = Get-Date ("1/1/2010")
(gci b.txt).CreationTime = Get-Date ("1/1/2010")
(gci c.txt).LastWriteTime = Get-Date ("1/1/2010")

Since there isn't a cmdlet for setting the time properties, we need to access properties in a different manner. To do this, we get the object and use the dot notation to access the property. The Get-Date cmdlet creates a valid date/time object for our new timestamp. Let's see how that worked.

PS C:\> gci | select Name, LastWriteTime, CreationTime, LastAccessTime

Name LastWriteTime CreationTime LastAccessTime
---- ------------- ------------ --------------
a.txt 1/23/2010 11:16:24 AM 1/23/2010 12:07:05 PM 1/1/2010 12:00:00 AM
b.txt 1/23/2010 11:16:24 AM 1/1/2010 12:00:00 AM 1/23/2010 12:07:05 PM
c.txt 1/1/2010 12:00:00 AM 1/23/2010 12:07:05 PM 1/23/2010 12:07:05 PM
d.txt 1/23/2010 11:16:24 AM 1/23/2010 12:07:05 PM 1/23/2010 12:07:05 PM


Interesting, the times have been changed, but can we forensically find a difference? After taking an image of the drive and using the istat tool from the SluethKit.org guys it is very obvious that something weird has happened. Let's take a look at an istat output snippet to see where the problem lies.


$STANDARD_INFORMATION Attribute Values:
Flags: Archive
Owner ID: 0
Security ID: 408 ()
Created: Fri Jan 01 00:00:00 2010
File Modified: Fri Jan 23 11:16:24 2010
MFT Modified: Fri Jan 23 11:20:43 2010
Accessed: Fri Jan 23 11:16:24 2010

$FILE_NAME Attribute Values:
Flags: Archive
Name: a.txt
Parent MFT Entry: 9946 Sequence: 10
Allocated Size: 0 Actual Size: 0
Created: Fri Jan 23 11:16:24 2010
File Modified: Fri Jan 23 11:16:24 2010
MFT Modified: Fri Jan 23 11:16:24 2010
Accessed: Fri Jan 23 11:16:24 2010


The PowerShell commands modify the STANDARD_INFO's created, file modified and accessed times. As you can see, there is a discrepancy in the Creation times between the FILE_NAME and STANDARD_INFO attribute values. Also, if you look at the STANDARD_INFO's MFT Modified date you can take a good guess as to when this change was made. The MFT Modified stamp is updated to the current system time whenever any of our PowerShell commands make a change to the file. The MFT Modified timestamp only marks the last change, so we could change all the dates to make it more confusing as to what change happened at that time.

While making the changes to the timestamps in PowerShell is effective when looking at the file system via the GUI or command line, not everything is hidden when looking at it with forensic tools.

While we can't go back in time and bump off Hitler, we did go back in time and take this functionality out of the standard Windows command line. OK, that's not really true. But what is true is that there is no way in cmd.exe to manipulate time stamps. So Ed won't be joining us with any cmd fu this week.

Hal has the touch:

Changing atimes and mtimes on Unix files is easy because we have the touch command. If you simply "touch somefile", then the atime and the mtime on that file will be updated to the current date and time (assuming you're the file owner or root).

But if you're root, you can also specify an arbitrary time stamp using the "-t" flag:

# touch -t 200901010000 /tmp/test
# alltimes /tmp/test
/tmp/test
atime: Thu Jan 1 00:00:00 2009
mtime: Thu Jan 1 00:00:00 2009
ctime: Sun Jan 24 05:33:56 2010

The ctime value is always set to the current date and time, because the touch command is updating the atime and mtime meta-data in the inode and ctime tracks the last meta-data update. By the way, don't bother looking for the alltimes command in your Unix OS. It's a little Perl script I wrote just for this Episode (download the script here).

A couple of our loyal readers wrote in to remind me of the stat command as an alternative to my alltimes script. stat has a different output format on different Unix-like OSes, but on Linux I could have done:

# stat /tmp/test | tail -3
Access: 2009-01-01 00:00:00.000000000 -0800
Modify: 2009-01-01 00:00:00.000000000 -0800
Change: 2010-02-03 15:26:33.279577981 -0800

Anyway, thenks everybody for the stat reminder!


The touch command also has "-a" and "-m" flags that allow you to selectively update only the atime or the mtime:

# touch -a -t 200909090909 /tmp/test
# touch -m -t 201010101010 /tmp/test
# alltimes /tmp/test
/tmp/test
atime: Wed Sep 9 09:09:00 2009
mtime: Sun Oct 10 10:10:00 2010
ctime: Sun Jan 24 05:49:29 2010

As you can see from the above example, touch is perfectly willing to set timestamps into the future as well as the past.

OK, so what about tweaking the ctime value? In general, setting the ctime on a file to an arbitrary value requires specialized, file system dependent tools. The good news(?) is that for Linux EXT file systems, the debugfs command will let us muck with inode meta-data. If you're dealing with other file system types or other operating systems, however, all I can say is good luck with your Google searching.

debugfs has a huge number of options that we don't have time to get into here. I'm just going to show you how to use set_inode_field to update the ctime value:

# debugfs -w -R 'set_inode_field /tmp/test ctime 200901010101' /dev/mapper/elk-root
debugfs 1.41.9 (22-Aug-2009)


The "-w" option specifies that the file system should be opened read-write so that we can actually make changes-- by default debugfs will open the file system in read-only mode for safety. We also need to specify the file system we want to open as the last argument. Sometimes this will be a disk partition device name like "/dev/sda1", but in my case I'm using LVM, so my disk devices have the "/dev/mapper/" prefix. If you're not sure what device name to use you can always run a command like "df -h /tmp/test" and look for the device name in the first column.

The "-R" option can be used to specify a single debugfs command to run in non-interactive mode. Note that there's also a "-f" option that allows you to specify a file of commands you want to run. If you leave off both "-R" and "-f" you'll end up in an interactive mode where you can run different commands at will.

In this case, however, we're going to use "-R" and run set_inode_field to set the ctime on /tmp/test. As you can see, you use a time stamp specification that's very similar to the one used by the touch command. And speaking of touch, we could use debugfs to "set_inode_field ... atime ..." or "set_inode_field ... mtime ..." instead of touch if we wanted to. This would allow us to update the atime/mtime values for a file without updating the ctime like touch does.

Anyway, now our ctime value should be updated, right? Let's check:

# alltimes /tmp/test
/tmp/test
atime: Wed Sep 9 09:09:00 2009
mtime: Sun Oct 10 10:10:00 2010
ctime: Sun Jan 24 05:49:29 2010

That doesn't look right! What's going on?

What's happening is that we've run afoul of the Linux disk cache. We actually have updated the information in the inode, but we've done it in such a way as to do an end-run around the normal Linux disk access routines, so our changes are not reflected in the in-memory file system cache. The good news is that (at least as of Linux kernel 2.6.16) there's a simple way to flush the inode cache:

# echo 2 > /proc/sys/vm/drop_caches
# alltimes /tmp/test
/tmp/test
atime: Wed Sep 9 09:09:00 2009
mtime: Sun Oct 10 10:10:00 2010
ctime: Thu Jan 1 01:01:00 2009

That looks better!

By the way, if you "echo 1 > /proc/sys/vm/drop_caches", that flushes the page cache. If you "echo 3 > /proc/sys/vm/drop_caches" it flushes both the page cache and the inode/dentry cache.