A filesystem implemented in userspace (using FUSE), inspired by the way Git is designed. This is a very early implementation, and contains lots of bugs.
- Zlib (>= 1.2)
- FUSE (>= 2.6)
- pkg-config (>= 0.25)
- Linux kernel (>= 2.6.15)
For the first run, you need two directories:
-
A git directory where the data will be stored.
-
An empty directory to use as the mountpoint.
$ cd /tmp $ mkdir gitdir mountp $ phoneixfs mount gitdir mountp
Use the mountpoint as you see fit. Data will be written to the gitdir on umount: you can use it for subsequent mounts.
Now, everything in mountpoint is versioned. To access older revisions of FILE, use FILE@REV syntax, where REV is the number of revisions in the past you want to access. For example:
$ echo "hello" >file1
$ echo "goodbye" >file1
$ echo "another hello" >file1
$ cat file1
another hello
$ cat file1@1
goodbye
$ cat file1@2
hello
Finally, to umount:
$ fusermount -u mountp
Uses a B+ tree to keep track of the filesystem tree, and a modified version of packfile v3/ packfile index v2 for storing revision information.
-
gitdir/.git/loose/ contains zlib-deflated versions of content blobs, named by the SHA-1 digest of the content.
-
gitdir/fstree is a raw dump of the B+ tree in a custom format.
-
gitdir/master.pack and gitdir/master.idx are the packfile and packfile index respectively. During an unmount, the files in gitdir/.git/loose/ are packed up, and an index is generated.
-
/tmp/phoenixfs.log is the debug log
master.idx, master.pack, and fstree are enough to recreate the entire versioned filesystem. The files in gitdir/ and gitdir/.git/loose/ can be removed after unmount.
Notes on the filesystem tree:
(dr: directory record | fr: file record | vfr: versioned file record)
-
dr just contains a name and node pointer referencing vfrs. drs are inserted directly into the root node.
-
vfr contains the path of the file, a list of frs representing the various versions of the file (fixed at REV_TRUNCATE), and a HEAD pointer to keep track of the latest version of the file.
-
The B+ tree is keyed by the CRC32 hash of the path of the vfr/ dr, a design decision inspired by Btrfs.
-
An fr, vfr, and dr (corresponding to its path) are created when a new file is created on the filesystem. Empty directories are not tracked: no dr is created for empty directories.
The following invocations don't work:
$ cp file1@1 file1 # FILE@REV can't be treated like a file
$ echo "hello" >file1; echo "goodbye" >file1 # Race!
Simply fork the project on GitHub and send pull requests.