Unionfs: flush and release updates
authorErez Zadok <ezk@cs.sunysb.edu>
Sun, 11 Nov 2007 03:01:16 +0000 (22:01 -0500)
committerErez Zadok <ezk@cs.sunysb.edu>
Fri, 6 Aug 2010 03:01:07 +0000 (23:01 -0400)
Remove the totalopens counter which was intended to reduce unnecessary
processing of d_deleted dentries.  Move that processing from file_release to
flush.

Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
fs/unionfs/commonfops.c
fs/unionfs/union.h

index 50e57756e9cbee3b36e55cfdeb6e9fa05f6cbd78..ba845298aaa2b6c305dfac7e885d8bff7dc0ed65 100644 (file)
@@ -551,9 +551,6 @@ int unionfs_open(struct inode *inode, struct file *file)
        bstart = fbstart(file) = dbstart(dentry);
        bend = fbend(file) = dbend(dentry);
 
-       /* increment, so that we can flush appropriately */
-       atomic_inc(&UNIONFS_I(dentry->d_inode)->totalopens);
-
        /*
         * open all directories and make the unionfs file struct point to
         * these lower file structs
@@ -565,7 +562,6 @@ int unionfs_open(struct inode *inode, struct file *file)
 
        /* freeing the allocated resources, and fput the opened files */
        if (err) {
-               atomic_dec(&UNIONFS_I(dentry->d_inode)->totalopens);
                for (bindex = bstart; bindex <= bend; bindex++) {
                        lower_file = unionfs_lower_file_idx(file, bindex);
                        if (!lower_file)
@@ -606,6 +602,7 @@ int unionfs_file_release(struct inode *inode, struct file *file)
        struct unionfs_file_info *fileinfo;
        struct unionfs_inode_info *inodeinfo;
        struct super_block *sb = inode->i_sb;
+       struct dentry *dentry = file->f_path.dentry;
        int bindex, bstart, bend;
        int fgen, err = 0;
 
@@ -628,6 +625,7 @@ int unionfs_file_release(struct inode *inode, struct file *file)
        bstart = fbstart(file);
        bend = fbend(file);
 
+       unionfs_lock_dentry(dentry);
        for (bindex = bstart; bindex <= bend; bindex++) {
                lower_file = unionfs_lower_file_idx(file, bindex);
 
@@ -635,7 +633,15 @@ int unionfs_file_release(struct inode *inode, struct file *file)
                        fput(lower_file);
                        branchput(sb, bindex);
                }
+
+               /* if there are no more refs to the dentry, dput it */
+               if (d_deleted(dentry)) {
+                       dput(unionfs_lower_dentry_idx(dentry, bindex));
+                       unionfs_set_lower_dentry_idx(dentry, bindex, NULL);
+               }
        }
+       unionfs_unlock_dentry(dentry);
+
        kfree(fileinfo->lower_files);
        kfree(fileinfo->saved_branch_ids);
 
@@ -799,11 +805,6 @@ int unionfs_flush(struct file *file, fl_owner_t id)
                goto out;
        unionfs_check_file(file);
 
-       if (!atomic_dec_and_test(&UNIONFS_I(dentry->d_inode)->totalopens))
-               goto out;
-
-       unionfs_lock_dentry(dentry);
-
        bstart = fbstart(file);
        bend = fbend(file);
        for (bindex = bstart; bindex <= bend; bindex++) {
@@ -813,14 +814,7 @@ int unionfs_flush(struct file *file, fl_owner_t id)
                    lower_file->f_op->flush) {
                        err = lower_file->f_op->flush(lower_file, id);
                        if (err)
-                               goto out_lock;
-
-                       /* if there are no more refs to the dentry, dput it */
-                       if (d_deleted(dentry)) {
-                               dput(unionfs_lower_dentry_idx(dentry, bindex));
-                               unionfs_set_lower_dentry_idx(dentry, bindex,
-                                                            NULL);
-                       }
+                               goto out;
                }
 
        }
@@ -830,8 +824,6 @@ int unionfs_flush(struct file *file, fl_owner_t id)
        /* parent time could have changed too (async) */
        unionfs_copy_attr_times(dentry->d_parent->d_inode);
 
-out_lock:
-       unionfs_unlock_dentry(dentry);
 out:
        unionfs_read_unlock(dentry->d_sb);
        unionfs_check_file(file);
index 9b530ec69ab8c3df64ef1daf57396bd2916fb554..8d19b2eee3fe1b6adccda47cff1d6d8f1e5b07b3 100644 (file)
@@ -102,8 +102,6 @@ struct unionfs_inode_info {
 
        /* The lower inodes */
        struct inode **lower_inodes;
-       /* to keep track of reads/writes for unlinks before closes */
-       atomic_t totalopens;
 
        struct inode vfs_inode;
 };