> On Jul 15, 2024, at 5:35 AM, Christian Brauner <brauner@xxxxxxxxxx> wrote: > >> The issue can be easily reproduced. >> #mount -t tmpfs "" /tmp/tdir >> #grep "/tmp/tdir" /proc/$$/mountinfo > > The kernel has accepted "" before the new mount api was introduced. So > the regression was showing "none" when userspace requested "" which got > fixed. The patch proposed right here would reintroduce the regression: The commit messages do not indicate that ’none’ caused a regression and was changed back to being empty string again, unless I missed it. I was under the impression ’none’ was added to address issues with parsing mount entries in /proc/<pid>mountinfo. Below was a patch proposal describing the parsing issues with libmount and other software because the mount source is empty. Don’t think this patch was merged. https://patchwork.kernel.org/project/linux-fsdevel/patch/1524109641-45617-1-git-send-email-cofyc.jackson@xxxxxxxxx/ Due to changes in v5.4 timeframe the mount entry started showing ’none’ for source name. Our Database application ran into an issue with libmount when moving from running on v5.4* kernel to v5.15* kernel version. > > (1) 4.15 > root@b1:~# cat /proc/self/mountinfo | grep mnt > 386 28 0:52 / /mnt rw,relatime shared:223 - tmpfs rw > > (2) 5.4 > root@f1:~# cat /proc/self/mountinfo | grep mnt > 584 31 0:55 / /mnt rw,relatime shared:336 - tmpfs none rw > > (3) 6.10-rc6 > root@localhost:~# cat /proc/self/mountinfo | grep mnt > 62 130 0:60 / /mnt rw,relatime shared:135 - tmpfs rw,inode64
