[ BUG during rename ]

Well, sometimes even the simple things don't work right.  I was testing
this patch by forcing an -EEXIST return value from reiserfs_rename,
which should have stopped the entire mv operation, but somehow the
rename was proceeding.  After leaving it alone for a bit, I did an
strace on mv, stupid thing unlinks the target if rename returns -EEXIST,
and then does a copy by hand.

Anyway, I think the bug you are hitting is just a symptom of a problem zam
found, where you've got a hidden entry lying around from some aborted 
operations.  debugreiserfs has a nifty feature that allows us to 
corrupt blocks at will, so I added a flag that turns a visible name into
a hidden one and did some extra tests.

The short version is that instead of an oops, you should now see clm-3000 
and zam-7002 messages in your log files.  Please let me know if/when they are
triggered.

-chris

diff -Nru a/fs/reiserfs/namei.c b/fs/reiserfs/namei.c
--- a/fs/reiserfs/namei.c	Tue May  8 20:29:33 2001
+++ b/fs/reiserfs/namei.c	Tue May  8 20:29:33 2001
@@ -1106,16 +1106,12 @@
     retval = reiserfs_add_entry (&th, new_dir, new_dentry->d_name.name, new_dentry->d_name.len, 
 				 old_inode, 0);
     if (retval == -EEXIST) {
-	// FIXME: is it possible, that new_inode == 0 here? If yes, it
-	// is not clear how does ext2 handle that
 	if (!new_inode) {
-	    printk ("reiserfs_rename: new entry is found, new inode == 0\n");
-	    BUG ();
+	    printk("clm-3000: name already found in dir %lu\n", new_dir->i_ino);
+	    goto error_out ;
 	}
     } else if (retval) {
-	pop_journal_writer(windex) ;
-	journal_end(&th, old_dir->i_sb, jbegin_count) ;
-	return retval;
+	goto error_out ;
     }
 
     reiserfs_update_inode_transaction(old_dir) ;
@@ -1250,8 +1246,11 @@
     if (new_inode)
 	reiserfs_update_sd (&th, new_inode);
 
+    retval = 0 ;
+
+error_out:
     pop_journal_writer(windex) ;
     journal_end(&th, old_dir->i_sb, jbegin_count) ;
-    return 0;
+    return retval ;
 }