I am writing a shell script that will be deployed to multiple machines and connect to a central server. When connecting, the script should identify the machine it is running on. (This is used to implement some rudimentary locking but that's not important for the question.)
I know that I could use the host-name, as reported by hostname -f, to identify the machine. But many personal devices have far from unique host-names, such as my-laptop or workstation so I'm wary of using this.
I might be able to add entropy by hashing the host-name together with some other host-specific information that will not change.
echo $(hostname -f) $(uname -snmpio) | md5sum
But the added entropy is very low. I'm having a hard time thinking of other system properties that can be hashed and are guaranteed not to change. (For example, I don't want to add any properties of the file system or other system configuration1 because it might legitimately change at any time.)
Finally, I thought about generating a random string the first time the script is run and store it in some configuration file. This wold be extremely likely to be unique and guaranteed not to change. But if possible, I'd prefer not having to manage persistent state.
Ideally, there would exist a utility to obtain a deterministic non-volatile UUID for the local system (like blkid for block-devices). It is not required that the UUIDs be hard to forge. This is not an authentication mechanism and I'm trusting all parties that run the script.
Are there any superior options I have overlooked?
1 Technically, the host-name is a system configuration, too. But if we change it, we expect the system to no longer be identified as the one it was before.
The easiest way to generate a unique identifier is to use a UUID. The most common type of UUID for this purpose is UUID v4, which is generally the correct choice unless you have some specific circumstances (e.g. namespacing requirements or poor sources of entropy) that would lead you to using one of the other UUID types.
You can use uuidgen on Linux, which can be found in the "uuid-runtime" package on Debian-based systems. The uuidgen tool is also installed by default OS X, should you need it. On Linux, the tool relies on libuuid and /dev/random to generate Version 4 UUIDs. If a high-quality random number generator isn't available, uuidgen will fall back on Version 1 time-based UUIDs.
UUIDGEN(1) says:
There are two types of UUIDs which uuidgen can generate: time-based UUIDs and random-based UUIDs. By default uuidgen will generate a random-based UUID if a high-quality random number generator is present. Otherwise, it will choose a time-based UUID. It is possible to force the generation of one of these two UUID types by using the -r or -t options.
As a general rule, if you're able to do so you should definitely stick with Version 4 as Version 1 has known security risks and limitations to its uniqueness properties. However, specific use cases may vary.
In the interest of providing options, you could use the RSA fingerprint of the localhost. Your machines most likely have all the components necessary already configured.
hostkey=$(ssh-keygen -l -f /path/to/host_key.pub)
The output will contain spaces and whatnot, but you could parse those out if it is a problem. The host key is usually in /etc/ssh/ssh_host_* but may depend on the distribution of linux.