ocfs2: always unmap in ocfs2_data_convert_worker()

Mmap-heavy clustered workloads were sometimes finding stale data on mmap
reads. The solution is to call unmap_mapping_range() on any down convert of
a data lock.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

1 files changed, 9 insertions, 1 deletions

diff --git a/fs/ocfs2/dlmglue.c b/fs/ocfs2/dlmglue.c
index e6220137bf69..e335541727f9 100644
--- a/fs/ocfs2/dlmglue.c
+++ b/fs/ocfs2/dlmglue.c
@@ -2718,6 +2718,15 @@ static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
         inode = ocfs2_lock_res_inode(lockres);
         mapping = inode->i_mapping;
 
+        /*
+         * We need this before the filemap_fdatawrite() so that it can
+         * transfer the dirty bit from the PTE to the
+         * page. Unfortunately this means that even for EX->PR
+         * downconverts, we'll lose our mappings and have to build
+         * them up again.
+         */
+        unmap_mapping_range(mapping, 0, 0, 0);
+
         if (filemap_fdatawrite(mapping)) {
                 mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
@@ -2725,7 +2734,6 @@ static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
         sync_mapping_buffers(mapping);
         if (blocking == LKM_EXMODE) {
                 truncate_inode_pages(mapping, 0);
-                unmap_mapping_range(mapping, 0, 0, 0);
         } else {
                 /* We only need to wait on the I/O if we're not also
                  * truncating pages because truncate_inode_pages waits