[linux-cifs-client] 2.4 bugs - vi failure - O_CREAT | O_EXCL problem fixed now, caching bug remains (only in 2.4)

Steve French smfltc at us.ibm.com
Thu Feb 5 02:49:14 GMT 2004


On Wed, 2004-02-04 at 10:41, Ian wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Steve,
> ~    Is this the cause of the odd vi behaviour? I ran some more tests and
> discovered that vi will behave oddly whether the file being edited is
> new or not.
> 

The vi problem was caused (again same story as one of the earlier bugs)
by the odd multistage behavior unique to 2.4 file creation causing:
	lookup/create/close/setattributes/lookup/getattributes/open/ 
that was broken (Only on 2.4) when one specified O_CREAT | O_EXCL (the
connecathon posix "nfs" special test flagged it nicer than vi).  The fix
for this is checked into samba cvs and will be released later as 1.0.2b
(I will move to 1.0.3 when I can fix the problem Howard reported below).

The differences in how invalidate_inode_pages works in 2.4 and 2.6 is
still puzzling (2.6 caching seems to work as expected - see previous
posts last week, but 2.4 kernel seems to ignore the invalidate of the
cached pages - or at least do something different than I expect). 
Hopefullly someone of linux-fsdevel will have a magic solution to how to
get past this on 2.4 kernel.



More information about the linux-cifs-client mailing list