Question on __inode_passivate() code

[mykaul]

The following code, specifically within the loop, is a bit strange to me:

list_for_each_entry_safe(dentry, t, &inode->dentry_list, inode_list)
{
    if (!__is_dentry_hashed(dentry))
        dentry_destroy(__dentry_unset(dentry));
}

If I understand it, if the dentry is NOT hashed, then let's unset and destroy it. But if it's not hashed, then what does __dentry_unset() do at all? Unhashed means that dentry->hash is empty, therefore __dentry_unset() can't list_del_init() it!

And indeed, we can see in the code coverage, that this path is not really called at all (https://build.gluster.org/job/line-coverage/1550/Line_20Coverage_20Report/libglusterfs/src/inode.c.gcov.html ):

426 10190680 : list_for_each_entry_safe(dentry, t, &inode->dentry_list, inode_list)
427 : {
428 5109382 : if (!__is_dentry_hashed(dentry))
429 0 : dentry_destroy(__dentry_unset(dentry));

[pranithk]

Went through the code,
check __inode_link() code, where it checks if the dentry table has a cycle, in that case I think __dentry_unset()'s code will be useful. I have only ever saw the cycle happen once in the last 12 years because of a bug.

That said, I don't see in what cases the data structure could end up in this state for __inode_passivate(). Since it is called only when nlookup count is non zero. Looks like dead code to me too.
I am ccing @amarts and @xhernandez as well, to check this. Maybe we should drop the loop.