hjfs

hjfs

[Julius Schmidt]

[-- Attachment #1: Type: text/plain, Size: 1392 bytes --]

I just committed a new file system for 9front, called hjfs after the 
infamous #pragma.
It can be already used as a root filesystem and now needs to be reviewed 
and tested.
Please note that it's highly experimental and expect that any data you put 
on it may be lost.
(It just crashed here a few minutes after the commit)

I would be happy if someone could review the code and/or test it and report 
any (possible) bugs here.

There are no really exciting features (it wouldn't have made a good phd 
thesis and it won't make a good reddit post either); it's for the most part 
a simplified version of cwfs (about four times smaller in terms of SLOC), 
differences being in particular
1. it uses one partition for everything
2. it has refcounting for blocks
3. as a result, dump is very quick
4. it uses lib9p
5. deleted files which are still open continue to exist as "zombies" (with 
the DGONE flag set) until they are closed

Planned future features:
1. file system checking (this is vital really)
2. move command to move file trees quickly
3. (perhaps) copy command to copy file trees quickly (copy-on-write)
4. separate attach for commands
5. (perhaps) support for multiple file systems similar to the other file 
servers (most of the underlying code already supports this)
6. "pruning" the dump (removing old data)
7. (perhaps) copying the dump in parts from and to other storage media

[-- Attachment #2: Type: text/html, Size: 1608 bytes --]

hjfs

[Ethan Gardener]

On Sun, Oct 20, 2019, at 10:13 AM, Steve Simon wrote:
> hi all,
> 
> can anyone give me an opinion on hjfs,
> i am interested in how it compares to, say, cwfs.

You can use hjfs in a 1GB disk image, where 12GB is the minimum recommended for cwfs because it needs several partitions.

hjfs dump is very fast because it simply marks files copy-on-write.  cwfs copies files to another partition on dump. Not a big deal since cwfs normally dumps at 5am. hjfs doesn't have a built-in scheduler, you'd have to trigger dumps with cron if you want that. 

as far as i know, hjfs tends to fragmentation, and i don't think it has a defragmenter. cwfs dump can't fragment, and its cache can be defragmented with a dump-restore cycle.

> Also, i would like to know whether hjfs is ready for “production” use - 
> its manpage indicates it it exoerimental, is that still the case?

I've barely used hjfs, but have heard good things about it; lots of people use it. It's not like cwfs is perfect anyway; it will crash if the cache partition fills, potentially losing a day's work. its design makes it hard to determine free space in the cache. hjfs has only one partition and can correctly compute free space. I never filled my cache, I'm quite a light user of disks, but did see a corrupted file once, in 10 years use. I got it back from the dump, but didn't try to find the last good version.