On Mon, Jun 10, 2019 at 9:31 PM Vivek Goyal <vgoyal@xxxxxxxxxx> wrote: > > On Sat, Jun 08, 2019 at 12:04:54PM +0300, Amir Goldstein wrote: > > Hi Matt, > > > > Thank you for trying to address this, but I see problems both in Why and > > How you did it. > > > > On Fri, Jun 7, 2019 at 11:51 PM Matt Coffin <mcoffin13@xxxxxxxxx> wrote: > > > > > > [Why] > > > Currently, if the redirect_dir option is set as a kernel or module > > > parameter, then even if metacopy is only enabled config, then both > > > metacopy and redirect_dir will be enabled when one creates a mount > > > point. This is not desirable because /sys/module/overlay/parameters will > > > still report that redirect_dir is not enabled > > > > /sys/module/overlay/parameters reports that redirect_dir is not enabled > > *by default* not per mount. > > > > > and there will be no redirect_dir=on line in the mount options in /proc/mounts. > > > > That is a bug. This code: > > /* Automatically enable redirect otherwise. */ > > config->redirect_follow = config->redirect_dir = true; > > > > Needs to update of config->redirect_mode. > > Hi Amir, > > It took me a while to understand what's the problem. So IIUC, issue is > that kernel has enabled redirect_dir by default. But it was disabled > using module parameter. But it was enabled again as a side affect because > metacopy=on was passed as mount option. And /proc/self/mountinfo does > not show redirect_dir=on and hence the confusion. > > IIRC, once you mentioned that we only show those options which needs > to be specified if same mount has to be reproduced on different machine > with same kernel/module options. If yes, then setting > config->redirect_dir=on is not needed because passing metacopy=on will > ensure that. I agree it is not a must fix bug, more of a nice to have. In all other similar cases (actual differs from module param value) we would have presented the option in /proc/mounts. It is only because redirect_dir is not consistent with redirect_mode that we do not. This seems inconsistent and hence I called it a bug. Thanks, Amir.
