#file-integrity #file-tree #backup #hash #metadata #update #capture

bin+lib rsure

Rsure captures hashes and metadata about a tree of files, and can later verify these hashes and metadata to determine if the files have changed. It supports incremental updates, and includes an standalone executable for scanning and checking trees.

11 unstable releases

0.9.4 Feb 16, 2022
0.9.3 Mar 21, 2021
0.9.0 Sep 8, 2019
0.8.1 May 19, 2018
0.5.0 Jul 31, 2016

#1096 in Filesystem

40 downloads per month

MIT license

140KB
3K SLoC

Contains (Zip file, 10KB) timing-parellel.xlsx

Rsure file integrity

Build Status

It has been said that backups aren't useful unless you've tested them. But, how does one know that a test restore actually worked? Rsure is designed to help with this.

History

The md5sum program captures the MD5 hash of a set of files. It can also read this output and compare the hashes against the files. By capturing the hashes before the backup, and comparing them after a test restore, you can gain a bit of confidence that the contents of files is at least correct.

However, this doesn't capture the permissions and other attributes of the files. Sometimes a restore can fail for this kind of reason.

Intrusion detection

There have been several similar solutions focused on intrusion detection. Tripwire and FreeVeracity (or Veracity) come to mind. The idea is that the files are compared in place to verify that nobody has modified them.

Unfortunately, at least tripwire seems to focus so heavily on this intrusion detection problem, that the tool doesn't work very well for verifying backups. It really wants a central database, and to use files by absolute pathname. FreeVeracity was quite useful for verifying backups, however, it appears to have vanished entirely (it was under an unusual license).

Incremental updates

One thing that none of these solutions addressed was that of incremental updates, probably because of the focus on intrusion detection. In a normal running system, the POSIX ctime field can be reliably used to determine if a file has been modified. By making use of this, the integrity program can avoid recomputing hashes of files that haven't changed. This strategy is similar to what most backup software does as well. This is important, because taking the time to hash every file can make the integrity update take so long that people avoid running it. Full hashing is impractical for the same reasons that regular full backups are usually impractical.

Using rsure

Getting it

Rsure is written in Rust. It began as an exercise to determine how useful Rust is for a systems-type program, and has shown to be the easiest implementation to develop and maintain.

Once you have installed rust (and cargo) using either the rust installer, rustup, or your distro's packaging system, building it is as easy as:

$ cargo build --release

within the Rsure directory. The --release flag is important, otherwise the performance is poor. You can install or link to ./target/release/rsure for the executable. It may also be possible to use cargo install to install sure directly.

Basic usage

Change to a directory you wish to keep integrity for, for example, my home directory:

$ cd
$ rsure scan

This will scan the filesystem (possibly showing progress), and leave a 2sure.dat.gz (the 2sure is historical, FreeVeracity used a name starting with a 0, and having the digit makes it near the beginning of a directory listing). You can view this file if you'd like. The format is somewhat readable.

Then you can do:

$ rsure check

to verify the directory. This will show any differences. If you back up this file with your data, you can run rsure after a restore to check if the backup is correct.

Later, you can run:

$ rsure update

which will update the 2sure.dat.gz file with the new data. Rust uses a "weave" format to hold multiple revisions efficiently in the same file. The update command will refresh the hashes of any files that have changed. After this, you can run:

$ rsure signoff

to compare the old scan with the current, and report on what has changed between them.

Dependencies

~41MB
~667K SLoC