smb3: fix perf regression with cached writes with netfs conversion

Write through mode is for cache=none, not for default (when
caching is allowed if we have a lease). Some tests were running
much, much more slowly as a result of disabling caching of
writes by default.

Reviewed-by: David Howells <dhowells@redhat.com>
Reviewed-by: Enzo Matsumiya <ematsumiya@suse.de>
Fixes: 3ee1a1fc3981 ("cifs: Cut over to using netfslib")
Signed-off-by: Steve French <stfrench@microsoft.com>

diff --git a/fs/smb/client/inode.c b/fs/smb/client/inode.c
index e8bfeea23660e47a6cf2e59544bafdfa0f913b16..5d4b0fd3a59e45de4abac84cad9e51d88211fe36 100644 (file)
--- a/fs/smb/client/inode.c
+++ b/fs/smb/client/inode.c
@@ -34,11 +34,8 @@
 static void cifs_set_netfs_context(struct inode *inode)
 {
        struct cifsInodeInfo *cifs_i = CIFS_I(inode);
-       struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
 
        netfs_inode_init(&cifs_i->netfs, &cifs_req_ops, true);
-       if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
-               __set_bit(NETFS_ICTX_WRITETHROUGH, &cifs_i->netfs.flags);
 }
 
 static void cifs_set_ops(struct inode *inode)