Summain—deterministic file manifests

Lars Wirzenius

2021-09-26 08:16

1 Introduction

A file manifest lists files, with their metadata.

To verify a backup has been restored correctly, one can compare a manifest of the data before the backup and after it has been restored. If the manifests are identical, the data has been restored correctly.

This requires a way to produce manifests that is deterministic: if run twice on the same input files, without the files having changed, the result should be identical. The Summain program does this.

This version of Summain has been written in Rust for the Obnam project.

1.1 Why not mtree?

mtree is a tool included in NetBSD Unix since version 1.2, released in 1996. It produces a manifest, and can check a manifest against the file system. It is, in principle, a tool that solves the same problem Summain. Why not use an existing tool. Some reasons:

In principle, there is no reason why mtree couldn’t be extended to support everything I need for Obnam. In practice, since I’m working on this in my free time in order to have fun, I prefer to write a new tool in Rust.

1.2 Why not use the old Python version of Summain

I don’t like Python anymore. The old tool would need updates to work with current Python, and I’d rather use Rust.

2 Usage

Summain is given one or more files or directories on the command line, and it outputs to its standard output a manifest. If the command line arguments are the same, and the files haven’t changed, the manifest is the same.

The output is YAML. Each file gets its own YAML document, delimieted by --- and ... as usual.

Summain does not itself traverse directories. Instead, a tool like find(1) should be used. Summain will, however, sort its command line arguments so that it doesn’t matter if they’re always in the same order.

3 Acceptance criteria

These scenarios verify that Summain handles the various kinds of file system objects it may encounter, with two exceptions: block and character devices. To create those, one needs to be the root user, and we don’t want to have to run the test suite as root. Instead, we blithely rely on the output being correct for those anyway. Testing manually indicates that it works, and the only difference from, say, regular files is that the mode starts with a b or c, which is exactly correct.

3.1 Directory

given an installed summain
given directory empty
and mtime for empty is 456
when I run chmod a=rx empty
when I run summain empty
then output matches file empty.yaml

File: empty.yaml

---
path: empty
mode: dr-xr-xr-x
mtime: 456
mtime_nsec: 0
nlink: 2
size: ~
sha256: ~
target: ~

3.2 Writeable file

given an installed summain
given file foo
and mtime for foo is 22
when I run chmod a=rw foo
when I run summain foo
then output matches file foo.yaml

File: foo.yaml

---
path: foo
mode: "-rw-rw-rw-"
mtime: 22
mtime_nsec: 0
nlink: 1
size: 0
sha256: e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
target: ~

3.3 Read-only file

given an installed summain
given file foo
and mtime for foo is 44
when I run chmod a=r foo
when I run summain foo
then output matches file readonly.yaml

File: readonly.yaml

---
path: foo
mode: "-r--r--r--"
mtime: 44
mtime_nsec: 0
nlink: 1
size: 0
sha256: e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
target: ~

3.4 Two files sorted

given an installed summain
given file aaa
and mtime for aaa is 44
given file bbb
and mtime for bbb is 44
when I run chmod a=r aaa bbb
when I run summain bbb aaa
then output matches file aaabbb.yaml

File: aaabbb.yaml

---
path: aaa
mode: "-r--r--r--"
mtime: 44
mtime_nsec: 0
nlink: 1
size: 0
sha256: e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
target: ~
---
path: bbb
mode: "-r--r--r--"
mtime: 44
mtime_nsec: 0
nlink: 1
size: 0
sha256: e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
target: ~
given an installed summain
given symlink ccc pointing at aaa
and mtime for ccc is 44
when I run summain ccc
then output matches file ccc.yaml

File: ccc.yaml

---
path: ccc
mode: lrwxrwxrwx
mtime: 44
mtime_nsec: 0
nlink: 1
size: 3
sha256: ~
target: aaa

3.6 Unix domain socket

given an installed summain
given socket aaa
and file aaa has mode 0700
and mtime for aaa is 44
when I run summain aaa
then output matches file socket.yaml

File: socket.yaml

---
path: aaa
mode: srwx------
mtime: 44
mtime_nsec: 0
nlink: 1
size: 0
sha256: ~
target: ~

3.7 Named pipe

given an installed summain
given named pipe aaa
and file aaa has mode 0700
and mtime for aaa is 44
when I run summain aaa
then output matches file fifo.yaml

File: fifo.yaml

---
path: aaa
mode: prwx------
mtime: 44
mtime_nsec: 0
nlink: 1
size: 0
sha256: ~
target: ~