[Cluster-devel] [GFS2 PATCH] GFS2: flush log and pages for jdata in gfs2_write_inode

[Cluster-devel] [GFS2 PATCH] GFS2: flush log and pages for jdata in gfs2_write_inode

[Bob Peterson]

Hi,

In function gfs2_write_inode, starting with patch a9185b41a4f84, we
only flush the log and call filemap_fdatawait if we're passed in a
wbc sync_mode of WB_SYNC_ALL. We also need to do these things if
we're evicting a jdata inode, because we might have jdata pages
still attached to bufdata descriptors that need to be revoked, but
by the time it gets to evict() it's too late to start a new
transaction. This patch changes it to treat jdata inodes as if
WB_SYNC_ALL had been specified.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
---
diff --git a/fs/gfs2/super.c b/fs/gfs2/super.c
index 769841185ce5..ac78e72de790 100644
--- a/fs/gfs2/super.c
+++ b/fs/gfs2/super.c
@@ -754,14 +754,15 @@ static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 	struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
 	struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
 	int ret = 0;
+	bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
 
-	if (wbc->sync_mode == WB_SYNC_ALL)
+	if (flush_all)
 		gfs2_log_flush(GFS2_SB(inode), ip->i_gl, NORMAL_FLUSH);
 	if (bdi->wb.dirty_exceeded)
 		gfs2_ail1_flush(sdp, wbc);
 	else
 		filemap_fdatawrite(metamapping);
-	if (wbc->sync_mode == WB_SYNC_ALL)
+	if (flush_all)
 		ret = filemap_fdatawait(metamapping);
 	if (ret)
 		mark_inode_dirty_sync(inode);

[Cluster-devel] [GFS2 PATCH] GFS2: flush log and pages for jdata in gfs2_write_inode

[Abhijith Das]

Hi,

Looks good to me. ACK.

Cheers!
--Abhi

----- Original Message -----
> From: "Bob Peterson" <rpeterso@redhat.com>
> To: "cluster-devel" <cluster-devel@redhat.com>
> Sent: Friday, October 6, 2017 8:38:21 AM
> Subject: [Cluster-devel] [GFS2 PATCH] GFS2: flush log and pages for jdata in gfs2_write_inode
> 
> Hi,
> 
> In function gfs2_write_inode, starting with patch a9185b41a4f84, we
> only flush the log and call filemap_fdatawait if we're passed in a
> wbc sync_mode of WB_SYNC_ALL. We also need to do these things if
> we're evicting a jdata inode, because we might have jdata pages
> still attached to bufdata descriptors that need to be revoked, but
> by the time it gets to evict() it's too late to start a new
> transaction. This patch changes it to treat jdata inodes as if
> WB_SYNC_ALL had been specified.
> 
> Signed-off-by: Bob Peterson <rpeterso@redhat.com>
> ---
> diff --git a/fs/gfs2/super.c b/fs/gfs2/super.c
> index 769841185ce5..ac78e72de790 100644
> --- a/fs/gfs2/super.c
> +++ b/fs/gfs2/super.c
> @@ -754,14 +754,15 @@ static int gfs2_write_inode(struct inode *inode, struct
> writeback_control *wbc)
>  	struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
>  	struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
>  	int ret = 0;
> +	bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
>  
> -	if (wbc->sync_mode == WB_SYNC_ALL)
> +	if (flush_all)
>  		gfs2_log_flush(GFS2_SB(inode), ip->i_gl, NORMAL_FLUSH);
>  	if (bdi->wb.dirty_exceeded)
>  		gfs2_ail1_flush(sdp, wbc);
>  	else
>  		filemap_fdatawrite(metamapping);
> -	if (wbc->sync_mode == WB_SYNC_ALL)
> +	if (flush_all)
>  		ret = filemap_fdatawait(metamapping);
>  	if (ret)
>  		mark_inode_dirty_sync(inode);
> 
>