Painless Thumbdrive Backups

Exploit udev rules to back up your Flash drive daily or every time you insert it.

Raise your hand if you've ever lost (or worried you'd lost) a USB thumbdrive. You spent hours fruitlessly searching the house, and then as you opened the washing machine door, it suddenly dawned on you that perhaps you didn't check your pockets thoroughly when you put this load in.

Fortunately, you have a backup of all the data, right? You religiously mount the drive and copy the data to a backup directory on a regular schedule, no?

That sounds an awful lot like drudgery to me too, and I got into computers to avoid boring work. Naturally, it's a lot more fun to spend some time working out the perfect method for painless thumbdrive backups.

What do I mean by painless? How about a system where you can walk up to your Linux box, plug in the drive, wait for a “backup complete” sound, unplug and walk away? Perhaps a system that keeps its backups orderly (say, the last seven copies)? Oh, and it should handle encrypted thumbdrives as well. And, if you need to recover, it should do both whole-volume replacement and per-file restores.

Not a problem. The key to this system is using udev rules and a simple shell script. The tools already are on your system. In this example, I use a CentOS 4.3 system, though any Linux distribution with a 2.6 kernel should work.

udev to the Rescue

udev is the modern device manager for Linux, replacing the 2.4 kernel's devfs. udev handles all device mapping, including hot plugging of devices. One of its coolest features is it lets you write your own event rules. This article shows you how to craft a rule that automatically fires when you plug your USB thumbdrive in to the system.

These rules are stored in /etc/udev/rules.d (if you're using a different Linux distribution, check /etc/udev/udev.conf for the udev_rules= line, which should point to the rules directory). You can place whatever udev rules you want as text files in this directory, and udev picks them up immediately for use without requiring a reboot.

How to Identify Your Device

To write a udev event rule, you first need a unique way to identify the USB device. Most thumbdrives have serial numbers, though not all. Fortunately, even with thumbdrives that do not have a serial number, you can craft udev rules for them.

I use two thumbdrives as examples: a JetFlash JF110, encrypted with TrueCrypt, and a Corsair Flash Voyager. The JetFlash has a serial number; the Corsair does not.

Plug your thumbdrive in, and cat /proc/scsi/usb-storage/*. You should find an entry for it similar to this:

   Host scsi5: usb-storage
       Vendor: Unknown
      Product: USB Mass Storage Device
Serial Number: 85a5b1f2c96492
     Protocol: Transparent SCSI
    Transport: Bulk

If you have a serial number, skip forward to the “Writing the Rule” section of this article. If you see “None” for the Serial Number, you still can identify the device by using udevinfo. Follow these steps:

1) Look at dmesg's output. Typical output is as follows:

usb-storage: waiting for device to settle before scanning
  Vendor: Corsair   Model: Flash Voyager     Rev: 1.00
  Type:   Direct-Access
  ANSI SCSI SCSI device sde: 2031616 512-byte hdwr sectors (1040 MB)
sde: assuming drive cache: write through
 sde: sde1
Attached scsi removable disk sde at scsi12, channel 0, id 0, lun 0
Attached scsi generic sg4 at scsi12, channel 0, id 0, lun 0,  type 0

This tells you that /dev/sde is the device assigned.

2) Now, run:

udevinfo -a -p $(udevinfo -q path -n /dev/sde)

and examine the output. Look for these lines:

SYSFS{model}=="Flash Voyager  "
SYSFS{vendor}=="Corsair "
Writing the Rule

Now, with either the serial number or the vendor/model combo, you can write the rule. The rule creates a symlink for the device in the /dev tree, for example, /dev/corsair_drive, and then calls the script /usr/local/bin/, which I'll get to in a moment.

Become root (su -), and create a text file in /etc/udev/rules.d called 95.backup.rules. You can use a number other than 95, but keep in mind that udev processes rules in alphanumeric order, and it's better to have local rules processed last.

If you have a serial number, type a rule like this (all on one line) into the file, and save it:

BUS="usb", SYSFS{serial}="85a5b1f2c96492", SYMLINK="jet_drive",
RUN+="/usr/local/bin/ jet_drive "

If you're using vendor/model identification, your rule would look like this:

BUS="scsi", SYSFS{vendor}=="Corsair ", SYSFS{model}=="Flash Voyager  ",
SYMLINK="corsair_drive", RUN+="/usr/local/bin/

Note that you can string as many SYSFS{} entries together as you need to identify the drive uniquely. Your rule now fires every time you plug in your thumbdrive.

Note: if you have other rules for a device, udev executes the rules in sequence from top to bottom.


Andrew Fabbro is a senior technologist living in the Portland, Oregon, area. He's used Linux since Slackware came on floppies and presently works for Con-way, a Fortune 500 transportation company.


Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.

Please see my question above

Benjamin Cathey's picture

Please see my question above - no one ever responded with any help. I thought it was because no one read this - but there have been posts since.

Automatically backup any USB storage device

Thomas Damgaard's picture


I have a server that only serves as backup server.
I've been trying to make a udev rule that would automatically backup any USB storage device connected.
This way, I can just plug in my USB devices to my server, and it is automatically backed up.

However, what I have made so far does not work. I hope you can help me.

Here is my udev rule:

RUN+="/usr/local/bin/ %k"

I hope you can help.

Cannot get this script working

Benjamin Cathey's picture

I read you magazine regularly and was glad when I found this article. However I cannot get it working at all --

I am running Ubuntu 8.04

For starters, when I run the udev check I do not get any values that speficy SYSFS, they are all ATTRS (although there is a line for serial and model) - also there is no BUS line at all.

I tried writing the rule using ATTRS and nothing, I also tried writing the rule using SYSFS (even though that parameter did not appear) and nothing -

They symlink is not even created.

HELP please

So I never heard back on this???

Benjamin Cathey's picture

Well, I asked for help and I never heard back on this - the output of udevinfo -a -p $(udevinfo -q path -n /dev/sdf) does not result in anything similar to what you are suggesting. There is no sysctl line or bus line - i see similar values in here but they are labelled attrs and the udev script just won't work

This is what I ended up making

root@lighthouse:/etc/udev/rules.d# cat 96-backuphome.rules
BUS=="usb", SYSFS{serial}=="0010101640150EE9W", SYMLINK=="tosh_ext", RUN+="/home/benito/scripts/ tosh_ext"

Although usb is listed as


NOT the BUS (although I know that it is) ... and serial looks like ATTRS{serial}== not SYSFS{serial}== as suggested in this article. I figured the reason I hadn't heard back is that no one read this. I read your magazine monthly - maybe I shouldn't bother if I can't get a reply?



How to recover

Derk Tattersall's picture

At the end of your article, you state that you can recover files from the image usong mount like so:

mkdir /mnt/thumb
mount -o loop corsair_drive.backup.0 /mnt/thumb

My own thumb drive (and most such drives, I think) has the data partition on a partition within the drive. You have to use a different mount command:

mount -o loop, offset=xxxxx corsair_drive.backup.0 /mnt/thumb

Determining the value of the offset is a pain. I found a script at that makes it much easier:

offset=$1; shift
limit=$1; shift
while [ $offset -le $limit ]
do if mount -o ro,loop,offset=$offset $* 2> /dev/null
then echo " Successfully mounted starting from offset $offset."
exit 0
[ $(($offset % 1000)) == 0 ] && echo -n . # Progress indicator
echo "No filesystem found up to $offset."
exit 1

I found the article very useful. Thanks.

Derek Tattersall

usb key partitions

Jeff Pipkins's picture

I found it instructive to write the run rule like this:
RUN+="/usr/local/bin/ myserialnum %k"

Then in the script I added echo $0 $@ >>/tmp/log.txt
I found that the script was called several times, with different device names. Then I removed the echo and added an if [ "$2" = "sdb2" ]
so I could mount only the partition I wanted.

I added a mount line in /etc/fstab and used the uid= and gid= to set myself as the owner. I have the script mount the drive, and luckily enough, when I remove the key, the mount goes away.

BTW, I don't use the key for backup, but I've found that the "unison" utility is very useful for syncing the data on the key with the data on either of two systems.

What I'd really like to do is to pop up a window, like gnome-terminal or xterm or something, and then execute an optionally interactive script. Anybody know how to do that? I tried sudo -u jpipkins gnome-terminal, but that didn't work.

Correction/Diff that worked for me

will's picture

Great article. This is something I've had to manually do and now I'm free of that task. Yahooo!

I struggled a little at first because it just didn't work straight away. I'm running Ubuntu Edgy. Then I followed the link to Daniel Drake's "Writing udev Rules" and noticed that his examples all used "==" instead of "=". Each declaration in the udev rule seems to need 2 math symbols. They should be "==" or "+=". Here is my rule and it worked great. Oh the joy!

BUS=="usb", SYSFS{serial}=="00176F962D19E", SYMLINK=="cruzer", RUN+="/usr/local/bin/ cruzer "

Now I just need new laptop with USB2.0 as a gig thumbdrive takes 30 mins to backup.


Luis Sismeiro's picture

Why not use rsync to backup only the modified files? It isn't difficult if the flash isn't encryped.

Luis Sismeiro


Bill Arlofski's picture

Rsync is a great solution for keeping files and/or directories in sync, and is much faster than copying the whole thing each time.

But, rsync is not so great if you sync, then realize that you need a specific version of a file from 2 days ago.

Bill Arlofski
Reverse Polarity

rsync snapshots

Chris's picture

You can get the best of both worlds though (speed of rsync + multiple versions), with the added bonus of consuming less space than multiple full copies.

The solution described there gives the illusion of multiple full copies, while only requiring the space of one copy plus the sum of the deltas.