Future allocator enhancements

[brendonhumphrey]

Thoughts on some work to be done to the allocator as time arises:

• Rename in such a way as to fit seamlessly into the SPL
• Remove osif, code directly for in kernel use. The time for testing in user space will soon have passed
• Replace the linked lists with use of the spl generic list if possible
• Provide a specialised slice type with low space overhead for small allocations - this will lift the space efficiency overall.
• Remove suspiciously C++ style type names.

[lundman]

To use built in lists, you do code like

/* Create list */
        list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
            offsetof(znode_t, z_link_node));
/* The list name is then z_all_znodes, which is a list of type znode_t, and inside znode_t
 * the "next,prev" private element is called "z_link_node". The offsetof does the math to work out
 * the offset to the private data inside the struct znode_t
 */

struct z_node_t {
/* your fields */
    list_node_t     z_link_node;  /* list's private node */
};

/* Allocating a new znode_t sutrct ,call */
        list_link_init(&zp->z_link_node);

/* Freeing a node, nothing is required, but you can do sanity */
        ASSERT(!list_link_active(&zp->z_link_node));

/* Freeing the list */
        list_destroy(&zfsvfs->z_all_znodes);

/* list API, should be protected by mutex or similar */
        list_insert_tail(&zfsvfs->z_all_znodes, zp);

        list_remove(&zfsvfs->z_all_znodes, zp);

        zp = list_head(&zfsvfs->z_reclaim_znodes);