On Mon, Jun 10, 2024 at 3:48 PM Andrew Bartlett <abartlet@xxxxxxxxx> wrote: > > On Mon, 2024-06-10 at 15:05 -0500, Steve French wrote: > > Yes - the reproducer helps. The bug is easy to reproduce. > > > I wanted to verify that the succeeding cases are the same that I see: > > - works with "cache=none" > > and > > - works with "nobrl" > > and > > - works with "vers=1.0" > > > All other combinations fail ... > > > Should be straightforward to fix in cifs.ko. Will look at a fix for > > this later today. > > > Awesome, thanks. > > > Note that the problem with SMB3.1.1 POSIX extensions is a Samba bug - > > a serious regression in the server (but trivial fix). We are waiting > > on someone to merge the oneline fix to the server (which we tested out > > ok) from David. > > > Is there an MR for this? I couldn't find it. I was puzzled why there wasn't a fix immediately applied since it was tested (and could add my Reviewed-by if that helped), and obvious server bug (regression), but I think David (who wrote the fix) was busy with other tasks. I would have done one but I am out of date with the Samba merge process. There are two other fairly simple problems (with basic SMB3.1.1 behavior), e.g. ctime not being updated in a few obvious cases (which is not really a POSIX only issue, as it also affects default mounts from Windows to Samba) which we hit in SMB3.1.1 POSIX - that I would love to see fixes for so we could restart our regular test automation to Samba from Linux (cifs.ko SMB3.1.1 POSIX mounts) -- Thanks, Steve
