]> err.no Git - linux-2.6/commitdiff
ocfs2: don't unconditionally pass LVB flags
authorMark Fasheh <mark.fasheh@oracle.com>
Wed, 13 Sep 2006 05:08:14 +0000 (22:08 -0700)
committerMark Fasheh <mark.fasheh@oracle.com>
Sun, 24 Sep 2006 20:50:47 +0000 (13:50 -0700)
Allow a lock type to specifiy whether it makes use of the LVB. The only type
which does this right now is the meta data lock. This should save us some
space on network messages since they won't have to needlessly transmit value
blocks.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
fs/ocfs2/dlmglue.c

index 3d5c6a0eb9ec5362322115db3e16eb757fb34b46..40562185088a13768764e98d95f914852b4d5c63 100644 (file)
@@ -135,6 +135,11 @@ struct ocfs2_lock_res_ops {
  */
 #define LOCK_TYPE_REQUIRES_REFRESH 0x1
 
+/*
+ * Indicate that a lock type makes use of the lock value block.
+ */
+#define LOCK_TYPE_USES_LVB             0x2
+
 typedef int (ocfs2_convert_worker_t)(struct ocfs2_lock_res *, int);
 static int ocfs2_generic_unblock_lock(struct ocfs2_super *osb,
                                      struct ocfs2_lock_res *lockres,
@@ -150,7 +155,7 @@ static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
 static struct ocfs2_lock_res_ops ocfs2_inode_meta_lops = {
        .get_osb        = ocfs2_get_inode_osb,
        .unblock        = ocfs2_unblock_meta,
-       .flags          = LOCK_TYPE_REQUIRES_REFRESH,
+       .flags          = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
 };
 
 static struct ocfs2_lock_res_ops ocfs2_inode_data_lops = {
@@ -870,6 +875,9 @@ static int ocfs2_cluster_lock(struct ocfs2_super *osb,
 
        ocfs2_init_mask_waiter(&mw);
 
+       if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
+               lkm_flags |= LKM_VALBLK;
+
 again:
        wait = 0;
 
@@ -937,7 +945,7 @@ again:
                status = dlmlock(osb->dlm,
                                 level,
                                 &lockres->l_lksb,
-                                lkm_flags|LKM_CONVERT|LKM_VALBLK,
+                                lkm_flags|LKM_CONVERT,
                                 lockres->l_name,
                                 OCFS2_LOCK_ID_MAX_LEN - 1,
                                 ocfs2_locking_ast,
@@ -2212,11 +2220,15 @@ static int ocfs2_drop_lock(struct ocfs2_super *osb,
 {
        enum dlm_status status;
        unsigned long flags;
+       int lkm_flags = 0;
 
        /* We didn't get anywhere near actually using this lockres. */
        if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
                goto out;
 
+       if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
+               lkm_flags |= LKM_VALBLK;
+
        spin_lock_irqsave(&lockres->l_lock, flags);
 
        mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
@@ -2266,7 +2278,7 @@ static int ocfs2_drop_lock(struct ocfs2_super *osb,
 
        mlog(0, "lock %s\n", lockres->l_name);
 
-       status = dlmunlock(osb->dlm, &lockres->l_lksb, LKM_VALBLK,
+       status = dlmunlock(osb->dlm, &lockres->l_lksb, lkm_flags,
                           ocfs2_unlock_ast, lockres);
        if (status != DLM_NORMAL) {
                ocfs2_log_dlm_error("dlmunlock", status, lockres);