[linux-cifs-client] Re: du reports wrong directory sizes on kernel 2.6(.1rc2) and cifs 0.9.4

[Steven French]

Yes - I fixed that one weeks ago although I found it amusing that most 
every filesystem at one time or other has made the same mistake - assuming 
that blocksize should be used in calculating number of blocks used :) but 
du requires that you calculate based on 512 byte blocks no matter what the 
reported blocksize is - very weird.

I also fixed various other bugs reported on the list, but I don't always 
remember to send notification to the list when they are fixed.   I do 
appreciate when people report these well defined problems though, the cifs 
code has made leaps forward in the last few months due to the reports on 
the list and via e-mail like this, sometimes with attached patch 
suggestions (and also the heavy 2.6 stress testing going that the IBM test 
labs has been doing recently has been helping). 

This will be much, much, much easier when cifs passes the whole set of 
standard filesystem tests which needless to say can be hard for a 
non-local filesystem much less an SMB/CIFS based protocol to service from 
Linux due to little quirks of POSIX file API :)   but this is huge 
progress and I am getting much more satisfied, and plan to take the 
"experimental" tag off the filesystem in Linux KConfig

Right now cifs seems to do quite well on:
1) fsx
2) connectathon "nfs" (posix file api) basic and general tests
3) almost completes the connectathon special and lock tests (this is where 
some of the hardest Windows -> POSIX mappings are)
4) fsstress works under light to medium stress
5) iozone and dbench and the performance centric tests that I have tried 
seem to work well

There are still half a dozen file system tests in the LTP which have not 
been traditionally tried on network filesystems, but which I would like to 
try on cifs as I get time so I can use the whole LTP file system test 
subsystem as a weekly regression suite and expand the verification to 
other server types (I would love to package a subset of ltp file tests 
that would autorun over cifs via a bootable knoppix iso image so server 
people could test their Linux network compatability eaiser).    I think 
the easier it is to run these regression tests - with no special windows 
compensations that the user has to care about .     jra in the Samba team 
has been looking at minor Samba changes when it sees the new cifs "posix" 
dialect which will also be a big help.

Finishing the last 1/2 of the xattr support and directory notification 
should be the next new features, although the spnego/kerberos ioctl should 
not be long behind.


Steve French
Senior Software Engineer
Linux Technology Center - IBM Austin
phone: 512-838-2294
email: sfrench at-sign us dot ibm dot com



Michael Pflüger <empx at gmx.de> 
Sent by: linux-cifs-client-bounces+sfrench=us.ibm.com at lists.samba.org
06/17/2004 10:01 AM

To
linux-cifs-client at lists.samba.org
cc

Subject
[linux-cifs-client] Re: du reports wrong directory sizes on kernel 
2.6(.1rc2) and cifs 0.9.4






The issue seems to be fixed now as of Kernel 2.6.7 from the cifs BK tree.
I think the fix happened between now and the 2.6.6 release.

Great work :)

Michael

_______________________________________________
linux-cifs-client mailing list
linux-cifs-client at lists.samba.org
http://lists.samba.org/mailman/listinfo/linux-cifs-client

-------------- next part --------------
HTML attachment scrubbed and removed