From: David S. <ds...@ja...> - 2005-04-08 16:46:18
|
On Fri, Apr 08, 2005 at 06:39:25PM +0200, Miklos Szeredi wrote: > > > In testing, I've found what seems to be a problem with NFS and the > > > direct_io option. Basically, without direct_io, a regular fuse mount > > > works as expected over NFS. When flipping direct_io on, the sequence > > > of events for a cp onto an already-existing file is: > > > > > > GETATTR /file > > > TRUNCATE /file to 0 > > > OPEN /file with flags 100001 > > > > > > then nothing. The cp just blocks at that point and the filesystem > > > program gets no further calls. > > > > We found another oops behind this problem: > > Great. This fix will just return an error for direct_io writes over > nfs. I'm not sure it's feasable to fix it any further. Thanks for the patch. Can you explain further what the problem is here? For my usage, I do need to have a direct_io mount, since I do not know the size of files in advance. This is just a read issue, of course. Is there some way to use direct_io for reads, but not for writes? David |