fusefs broken on 8-stable?

[David P. Discher]

On Dec 30, 2011, at 1:02 AM, Gleb Kurtsou wrote:

> On (29/12/2011 14:06), David P Discher wrote:
>> Is this a straight dump for p4 ?  
>> 
>> Earlier this week, I checked out //depot/projects/soc2011/ilya_fuse/...  out of p4.  But only used the kmod, not the libs. Unfortunately, it was 10x slower than the 0.3.9 version currently in the ports tree. I never got it to the same point ... after a few days, had only written about 6-10GB of data.
> 
> It's the same as p4 (+ few minor tweaks). fuse-0.4 is more strict on
> resizing files which is very likely to be slow for encrypted file
> system (encfs). Try setting sysctl vfs.fuse.sync_resize=0.

Got this port started, however its still around 10-60x slower than the 0.3.9 version in the ports tree.  vfs.fuse.sync_resize seemed to make no difference.  Granted this is a custom fusefs that I'm comparing, based off of bbfs (http://www.cs.nmsu.edu/~pfeiffer/fuse-tutorial/).

> 
> I don't know all the details but there was memory leak and arbitrary
> memory corruption in old fuse. It's believed to be fixed in fuse-0.4. My
> guess you are hitting the same bug.

I've compared the functions in the stack the trace, and have found no change between 0.4.4 and 0.3.9.  If you (or anyone else on the list) knows what patches are the memory leak and corruption fixes are - I'd like to attempt to back port them into 0.3.9.  I'd like the stability and speed. 


---
David P. Discher
dpd at bitgravity.com * AIM: bgDavidDPD
BITGRAVITY * http://www.bitgravity.com