On 04/05/2011 04:38 PM, Jeff Layton wrote: > On Tue, 05 Apr 2011 16:34:17 -0700 > Ben Greear<greearb@candelatech.com> wrote: > >> On 04/05/2011 04:26 PM, Jeff Layton wrote: >> >>> Does the attached patch fix it? It's probably stable material if so... >> >> I'll test your patch. I verified that cifs_file is NULL in cifsFileInfo_put >> and protecting against that also solves the problem. >> > > I'd probably prefer the patch I'm suggesting. I worry that turning > cifsFileInfo_put into a no-op for NULL pointers might paper over future > bugs. Fine by me. Your patch appears to work fine, though I had to apply it manually with patch, as 'git am' didn't understand it's format. I agree this should go to stable. Tested-by: Ben Greear <greearb@candelatech.com> > >>> >>> FWIW, cifs doesn't handle O_DIRECT at all. >> >> I know..and I warn my users when the open() call fails, but I still >> let them attempt the file open. Seems a decent test case after all :) >> > > Definitely. It's a cifs bug for sure, just pointing out that it > probably isn't going to give you what you expect. As Steve points out > forcedirectio might give you something closer to this, but proper > O_DIRECT support is still on the to-do list. I was using just 'directio'. On Fedora 13, there is no mention of 'forcedirectio' in the mount.cifs man page. Are they actually the same thing, or is the man page just missing stuff? Thanks, Ben