[Cluster-devel] [PATCH 0/2 v2] new vfs freeze hooks and gfs2 freeze rewrite

Steven Whitehouse swhiteho at redhat.com
Mon Nov 17 12:04:16 UTC 2014


Hi,

Both patches are now in the -nmw tree. Thanks,

Steve.

On 14/11/14 02:42, Benjamin Marzinski wrote:
> The existing gfs2 freeze code can't properly make use the kernel vfs
> freezing code on all the nodes of a cluster. This is because the only hook
> into the gfs2 freeze code comes inside of freeze_super, which holds the
> sb->s_umount lock. In order to notify the other nodes in the cluster to
> freeze the filesystem, gfs2 must grab a glock. Thus, if two nodes tried to
> freeze the filesystem at the same time and all cluster nodes used
> freeze_super to enforce the freeze, both nodes would first grab the
> sb->s_umount lock in freeze_super, and then they would race to grab the
> gfs2 freeze glock. The loser would be stuck in freeze_super, waiting on the
> freeze glock. This means that gfs2 can't rely on freezing all the nodes in
> the cluster by calling freeze_super, because one node would already be
> stuck in that function with sb->s_umount held. So, currently gfs2 only
> calls freeze_super on the node that initiates the freeze and deals with
> blocking the writes entirely within the gfs2 code. This involves some
> pretty hacky code.
>
> To fix this, the first patch of this patchset adds two new vfs hooks,
> freeze_super and thaw_super.  If a filesystem implements these hooks, they
> will be called instead of the vfs freeze_super and thaw_super functions.
> The second patch of this set makes use of these hooks to simplify the gfs2
> freezing code. This allows gfs2 to grab the freeze glock first and call the
> actual vfs freeze functions in response to node grabbing the freeze glock.
> Thus, all nodes can use the vfs freeze_super and thaw_super code to freeze
> the filesystem. This vesion also fixes some issues with a race between
> reserving log space and making the filesystem read-only.
>
> The current hook names were suggested by Jan Kara, over my original
> prepare_freeeze and prepare_thaw names, which were more confusing.
>
> Benjamin Marzinski (2):
>    fs: add freeze_super/thaw_super fs hooks
>    gfs2: update freeze code to use freeze/thaw_super on all nodes
>
>   fs/block_dev.c       |  10 ++++-
>   fs/gfs2/glops.c      |  26 ++++++------
>   fs/gfs2/glops.h      |   2 +
>   fs/gfs2/incore.h     |  18 ++++++---
>   fs/gfs2/inode.c      |  40 +++++-------------
>   fs/gfs2/log.c        |  42 +++++++++----------
>   fs/gfs2/main.c       |  11 ++++-
>   fs/gfs2/ops_fstype.c |  18 +++------
>   fs/gfs2/super.c      | 112 ++++++++++++++++++++++++++++++++++-----------------
>   fs/gfs2/super.h      |   1 +
>   fs/gfs2/trans.c      |  17 ++++++--
>   fs/ioctl.c           |   6 ++-
>   include/linux/fs.h   |   2 +
>   13 files changed, 176 insertions(+), 129 deletions(-)
>




More information about the Cluster-devel mailing list