On Fri, Jun 16, 2023 at 1:28 PM Christian Brauner <brauner@xxxxxxxxxx> wrote:
>
> Cleanup sb->s_fs_info when ovl_fill_super() fails before setting
> sb->s_root. The logic is a bit convoluted but tl;dr: If sget_fc() has
> succeeded fc->s_fs_info will have been transferred to sb->s_fs_info. So
> by the time ->fill_super()/ovl_fill_super() is called fc->s_fs_info is
> NULL consequently fs_context->free() won't call ovl_free_fs(). If we
> fail before sb->s_root() is set then ->put_super() won't be called which
> would call ovl_free_fs(). IOW, if we fail in ->fill_super() before
> sb->s_root we have to clean it up.
>
> Amir reported an issue when running xfstests overlay/037 which made me
> investigate and detect this.
>
> Fixes: fc0dc3a9b73b ("ovl: port to new mount api")
> Signed-off-by: Christian Brauner <brauner@xxxxxxxxxx>
> ---
>
> ---
> I would just fold this into ("ovl: port to new mount api"). The patch
> here is mostly so we have a track record of the issue.
Folder (with commit message) and applied to
https://github.com/amir73il/linux/commits/fs-overlayfs-mount_api
I've split out two prep patches, so "port to new mount api" patch
is now as clean as it can be.
>
> Passes both
>
> sudo ./check -overlay overlay/*
> sudo ./check -overlay -g overlay/union
>
> I somehow must've missed the earlier failure of overlay/037.
No failures on my tests so far.
Thanks!
Amir.
