Jelly, 11 bytes

WẋLŒDUṙLFsL

Try it online!

Explanation

               Input: array z
WẋL            length(z) copies of z
   ŒD          Diagonals (starting with main diagonal)
     U         Reverse each
      ṙL       Rotate left length(z) places
               (now the top-left diagonal is in front)
        F      Flatten
         sL    Split into chunks of size length(z)

Python 2, 105 96 bytes

-1 and -4 and -4 bytes thanks to Flp.Tkc

a=input()
n=len(a)
L,M=[],[]
for i in range(n):L+=a[i::-1];M+=a[:i:-1]
print zip(*[iter(L+M)]*n)

The for loop adds the items like in the description, the real magic happens in the zip which is from here

JavaScript (ES6) 101 105

a=>(u=>{for(r=[],i=l=a.length;i+l;i--)for(j=l;j--;v&&((u%=l)||r.push(s=[]),s[u++]=v))v=a[j-i]})(0)||r

Less golfed

a => {
  u = 0
  for(r=[], i=l=a.length; i+l>0; i--)
    for(j=l; j--; )
    {
      v = a[j-i]
      if (v) 
      {
        u %= l
        if (u==0) r.push(s=[])
        s[u++] = v
      }
    }
  return r
}

Test

F=
a=>(u=>{for(r=[],i=l=a.length;i+l;i--)for(j=l;j--;v&&((u%=l)||r.push(s=[]),s[u++]=v))v=a[j-i]})(0)||r

function update() {
  var a=I.value.match(/\d+/g)
  if (a) {
    var r=F(a)
    O.textContent = r.join`\n`
  }
}

update()

<input id=I value='1 2 3 4 5' oninput='update()'>
<pre id=O></pre>

MATL, 17 bytes

!Gg*tRwZRhPXzGne!

Try it online!

How it works

The following explanation uses input [1 2 3 4 5] as an example. To visualize the intermediate results, insert % (comment symbol) after any statement in the code.

Note that ; is the row separator for matrices. So [1 2] is a row vector, [1; 2] is a column vector, and [1 0; 0 1] is the 2×2 identity matrix.

!     % Implicitly input a row vector. Transpose. Gives a column vector
      % STACK: [1; 2; 3; 4; 5]
Gg    % Push input with all (nonzero) values replaced by ones
      % STACK: [1; 2; 3; 4; 5], [1 1 1 1 1]
*     % Multiply, with broadcast. Gives a square matrix
      % STACK: [1 1 1 1 1;
                2 2 2 2 2;
                3 3 3 3 3;
                4 4 4 4 4;
                5 5 5 5 5]
tR    % Duplicate. Upper triangular part
      % STACK: [1 1 1 1 1;
                2 2 2 2 2;
                3 3 3 3 3;
                4 4 4 4 4;
                5 5 5 5 5],
               [1 1 1 1 1
                0 2 2 2 2;
                0 0 3 3 3;
                0 0 0 4 4;
                0 0 0 0 5]
wZR   % Swap. Lower triangular part, below main diagonal 
      % STACK: [1 1 1 1 1;
                0 2 2 2 2;
                0 0 3 3 3;
                0 0 0 4 4;
                0 0 0 0 5],
               [0 0 0 0 0;
                2 0 0 0 0;
                3 3 0 0 0;
                4 4 4 0 0;
                5 5 5 5 0]
h     % Concatenate horizontally
      % STACK: [1 1 1 1 1 0 0 0 0 0;
                0 2 2 2 2 2 0 0 0 0;
                0 0 3 3 3 3 3 0 0 0;
                0 0 0 4 4 4 4 4 0 0;
                0 0 0 0 5 5 5 5 5 0]
P     % Flip vertically
      % STACK: [0 0 0 0 5 5 5 5 5 0;
                0 0 0 4 4 4 4 4 0 0;
                0 0 3 3 3 3 3 0 0 0;
                0 2 2 2 2 2 0 0 0 0;
                1 1 1 1 1 0 0 0 0 0]
Xz    % Column vector of nonzeros, taken in column-major order
      % STACK: [1;2;1;3;2;1;4;3;2;1;5;4;3;2;1;5;4;3;2;5;4;3;5;4;5]
Gne   % Reshape into a matrix with as many rows as input size
      % STACK: [1 1 5 5 4;
                2 4 4 4 3;
                1 3 3 3 5;
                3 2 2 2 4;
                2 1 1 5 5]
 !    % Transpose. Implicitly display
      % STACK: [1 2 1 3 2;
                1 4 3 2 1;
                5 4 3 2 1;
                5 4 3 2 5;
                4 3 5 4 5]

JavaScript (ES6), 116 bytes

a=>a.map(_=>b.splice(0,a.length),b=[].concat(...a.map((_,i)=>a.slice(~i)),...a.map((_,i)=>a.slice(0,~i))).reverse())

Well, it's a start...

Mathematica 93 Bytes

Partition[Flatten[Table[If[i>n,#[[n;;(i-n+1);;-1]],#[[i;;1;;-1]]],{i,1,2(n=Length@#)-1}]],n]&

Here's how I'd ordinarily write this code (109 Bytes):

Partition[Reverse@Flatten[Table[Reverse@Diagonal[ConstantArray[Reverse@#,n],k],{k,-(n=Length@#)+1,n-1}]],n]&

This matrix plot gives a good idea from the structure due to a sequentially increasing input vector.

Here's the matrix plot with a random input vector. Obviously some structure still exists.

Mathematica, 92 bytes

n=NestList[#2,(r=Reverse)@#,(l=Length@#)-1]&;{Most@r[#~n~Rest],#~n~Most}~ArrayReshape~{l,l}&

Unnamed function taking a list as its argument. There might be other structures to such a function, but hopefully I golfed this structure pretty good....

The first part n=NestList[#2,(r=Reverse)@#,(l=Length@#)-1]& defines a function n of two arguments: the first is a list of length l, and the second is a function to apply to lists. n applies that function l-1 times to the reversed argument list, saving all the results in its output list. (Defining r and l along the way is just golfing.)

n is called twice on the original list, once with the function being Rest (drop the first element of the list) and once with the function being Most (drop the last element). This produces all the desired sublists, but the whole list is there twice (hence the extra Most) and the first half is there in backwards order (hence the r[...]). Finally, ~ArrayReshape~{l,l} forgets the current list structure and forces it to be an lxl array.

Mathematica, 85 bytes

Literally performing the steps suggested:

(l=Length@#;Partition[Flatten@Table[Reverse@Diagonal[Table[#,l],i],{i,-l+1,l-1}],l])&

My gut says that there should be a clever way to use Part to do this shorter, but every attempt I've made has been longer than 85 bytes.