Memory-mapped file - The most effective mapping size when using mmap()

Question

I'm working on 'database' program where I store objects in one file.
I want to read objects using mmap() function.

The storage is document-oriented database, where objects are formed into a tree. The object struct (it has a size of 64 bytes) looks like this:

struct Entity {
    section_off key_ptr;
    size_t key_size;
    section_off val_ptr;
    size_t val_size;
    enum value_type val_type;
    file_off ancestor_ptr;
    file_off child_ptr;
    file_off sibling_ptr; // ptr to next child of the same ancestor
}; 

I store these objects in the sections into which the file is divided. Each section has a size of 8 Kbytes and begins with section_header struct:

struct section_header {
    size_t free_space;
    section_off first_free_cell;
    section_off last_free_cell; 
    file_off section_addr;
};

So the question is:
What size part of the file do I need to map in order to go through all objects in my file? Should it be the size of a single section, multiple sections or the entire file? The file size can be unlimited.

I need to go through not only all the objects in the file, but also get some exact objects by it's address. So I should already have a function, that would map a certain object from a file or the entire section with this object. Therefore, I don't know if I need to implement a function, that would map multiple sections to go through the entire file faster.