Subject: Re: Log structured filesystem
To: None <metamatic!metamatic.denver.co.us!ediger@uucp-1.csn.net>
From: David Brownlee <david@mono.org>
List: port-sparc
Date: 06/30/1996 17:53:09
	AFAIR lfs is broken in NetBSD - & looking for someone to work
	on it :)

	It should be marked as 'still experimental' in the config file.

	The man page _really_ should note that its broken, care
	to submit a PR on this? :)


.---- I've been too drunk to love ----.--- I've been too drunk to remember -.
|          Too drunk to care          |     The hell of the night before    |
|  Looked like death, felt like Hell  |    I've been drinking myself blind  |
`------ Been the worse for wear ------'--- And still I'll drink some more --'


On Sun, 30 Jun 1996, Bruce Allen Ediger wrote:

> 
> 
> I'm running NetBSD 1.1 on a SPARCStation IPC, ROM Rev 2.9. I've tried
> to put an LFS filesystem on a SUN0207 disk (<CONNER, CP30200  SUN0207, 4234>).
> 
> I can put the filesystem itself on the 'g' partition of the disk
> using the newlfs command (newlfs -L /dev/rsd1g).  From der Mouse's
> sunlabel program, the disk looks like this:
> 
> 1:46am 4 # sunlabel /dev/rsd1a                                                /
> sunlabel> p
> a: start cyl =      0, size =    16848 (52/0/0 - 8.22656Mb)
> b: start cyl =     52, size =   178200 (550/0/0 - 87.0117Mb)
> c: start cyl =      0, size =   406296 (1254/0/0 - 198.387Mb)
> d: start cyl =      0, size =        0 (0/0/0 - 0Mb)
> e: start cyl =      0, size =        0 (0/0/0 - 0Mb)
> f: start cyl =      0, size =        0 (0/0/0 - 0Mb)
> g: start cyl =    602, size =   211248 (652/0/0 - 103.148Mb)  <=== LFS here
> h: start cyl =      0, size =        0 (0/0/0 - 0Mb)
> 
> I can mount the LFS filesystem on /lfs using either a vanilla 'mount' or
> a mount_lfs:
> 
> mount_lfs -d -s /dev/sd1g /lfs
> 
> It mounts, and I can 'cd' into /lfs and do listings.  I can even 'touch'
> and 'rm' files.  But if I do a 'cp' or a 'mkdir' and 'cd', the 'cp'
> process hangs.  Here's what a 'ps -algxww' says about a zsh shell that
> I used to cd into a newly created directory in /lfs:
> 
>     0   144   134   0  -5  0   740  164 biowai D+   p4    0:02.39 (zsh)
> 
> Apparently, any process that tries to write bytes to an LFS hangs.
> The lfs_cleanerd just fives messages like this:
> 
> Cleaner Not Running at Sun Jun 30 01:51:03 1996
> 
> I've tried both the binary distribution kernel, and a kernel I've built
> myself.  Both of these have 'options LFS' in the configuration file.
> I get the same hanging in 'D' state on the biowai WCHAN in every process
> that tries to write to the disk.
> 
> I've even recompiled the newlfs, lfs_cleanerd and mount_lfs executables,
> with no change in results.
> 
> I've put an ffs filesystem on the same disk, and that filesystem seems
> to work, so I'm pretty sure its not the disk hardware that's hosed.
> 
> I found Margo Seltzer's January, 1995 usenet article, and the source code
> for LFS that she used.  I compared the NetBSD lfs code with hers and found
> a few differences, but nothing more than trivialites (syscallarg and
> SCARG macros, format of credentials struct, timeval field name differences).
> 
> What's the scoop?  Does LFS work at all?  Is there some mystical trick
> I'm missing, like extra options in the config file?
> 
> Best regards,
> Bruce Ediger
>