On Fri, Nov 11, 2022 at 04:38:26PM -0500, Mike Marshall wrote:
> Hello... I have applied your v5 patch series to 6.1-rc2. I get
> numerous acl related xfstests failures with your patch.
>
> generic/105 is the first one that fails, so I got started dee-ciphering
> what it does.
>
> I hoped I could find what it is in the patch that causes the
> regression, but I have not yet, and we're already up to rc4.
>
> I have a sequence of events that is part of generic/105 where
> I believe "the bad thing" happens, I thought I'd show you, you
> might know right away what is wrong...
>
> 6.1-rc2 without the patch series:
> root@vm1 ~]# cd /scratch
> [root@vm1 scratch]# mkdir -m 600 subdir
> [root@vm1 scratch]# chown 13 subdir
> [root@vm1 scratch]# echo data > subdir/file
> [root@vm1 scratch]# ls -l subdir/file | awk '{ print $1, $3 }'
> -rw-r--r--. root
>
> 6.1-rc2 with the patch series:
> [root@vm1 hubcap]# cd /scratch
> [root@vm1 scratch]# mkdir -m 600 subdir
> [root@vm1 scratch]# chown 13 subdir
> [root@vm1 scratch]# echo data > subdir/file
> [root@vm1 scratch]# ls -l subdir/file | awk '{ print $1, $3 }'
> -rw-rw-rw-. root
>
> The commit message for the orangefs part of the patch
> says "calculate the correct mode directly before
> actually creating the inode."
>
> Anywho... I'll keep looking...
Thanks for the report. I'm mostly afk this weekend but I've installed
the orangefs-server on Fedora and will reproduce and fix this early next
week.
