Re: [PATCH 1/2] xfs: correct the sb_rgcount when the disk not support rt volume

[Long Li <leo.lilong@huawei.com>]

On Tue, Jan 07, 2025 at 04:32:06PM -0800, Darrick J. Wong wrote:
> On Tue, Jan 07, 2025 at 09:11:23PM +0800, Long Li wrote:
> > On Mon, Jan 06, 2025 at 11:52:20AM -0800, Darrick J. Wong wrote:
> > > On Tue, Dec 31, 2024 at 10:34:22AM +0800, Long Li wrote:
> > > > When mounting an xfs disk that incompat with metadir and has no realtime
> > > > subvolume, if CONFIG_XFS_RT is not enabled in the kernel, the mount will
> > > > fail. During superblock log recovery, since mp->m_sb.sb_rgcount is greater
> > > > than 0, updating the last rtag in-core is required, however, without
> > > > CONFIG_XFS_RT enabled, xfs_update_last_rtgroup_size() always returns
> > > > -EOPNOTSUPP, leading to mount failure.
> > > 
> > > Didn't we fix the xfs_update_last_rtgroup_size stub to return 0?
> > > 
> > > --D
> > 
> > Indeed, when CONFIG_XFS_RT is not enabled, xfs_update_last_rtgroup_size() should
> > return 0, as returning an error is meaningless.
> > 
> > 1) For kernels without CONFIG_XFS_RT, mounting an image with realtime subvolume will
> > fail at xfs_rtmount_init().
> > 
> > 2) For kernels without CONFIG_XFS_RT, mounting an image without realtime subvolume
> > should succeed.
> > 
> > However, in the current scenario, should sb_rgcount be initialized to 0 ? it will 
> > consistent with metadir feature is enabled. The xfs-documentation [1] describes 
> > sb_rgcount as follows:
> > 
> > "Count of realtime groups in the filesystem, if the XFS_SB_FEAT_RO_INCOMPAT_METADIR
> > feature is enabled. If no realtime subvolume exists, this value will be zero."
> > 
> > [1] https://git.kernel.org/pub/scm/fs/xfs/xfs-documentation.git/tree/design/XFS_Filesystem_Structure/superblock.asciidoc
> 
> Ah, I see your point finally -- if there's no realtime section, then
> there's no need to allocate any incore rtgroups, nor is there any point
> to set sb_rgcount==1.
> 
> That said, I think the correct tags here are:
> Cc: <stable@vger.kernel.org> # v6.13-rc1
> Fixes: 96768e91511bfc ("xfs: define the format of rt groups")
> 
> because 96768e91511bfc is the commit that actually added "to->sb_rgcount
> = 1;".
> 

Ok, thanks for point out that, In fact, xfs_grow_last_rtg() needs to be modified together,
otherwise the growfs may be problematic. Therefore, I will release a new version.

Long Li