Binary Bomb - Phase 4

I am having a very difficult time tracing the assembly code for the following binary bomb (An assignment from school where a bomb has to be defused, this bomb contains 6 phases which all have 1 correct input to proceed to the next phase). I am currently on phase_4 and it has a recursive function called func4. I have identified that the input is "%d %d" which is two integers. However, I cannot quite figure out what func4 is doing, even after getting the info on all registers throughout every step.

Phase_4:

    (gdb) disas
Dump of assembler code for function phase_4:
=> 0x08048e24 <+0>: sub    $0x2c,%esp
   0x08048e27 <+3>: lea    0x1c(%esp),%eax
   0x08048e2b <+7>: mov    %eax,0xc(%esp)
   0x08048e2f <+11>:    lea    0x18(%esp),%eax
   0x08048e33 <+15>:    mov    %eax,0x8(%esp)
   0x08048e37 <+19>:    movl   $0x804a7f1,0x4(%esp)
   0x08048e3f <+27>:    mov    0x30(%esp),%eax
   0x08048e43 <+31>:    mov    %eax,(%esp)
   0x08048e46 <+34>:    call   0x80488d0 <__isoc99_sscanf@plt>
   0x08048e4b <+39>:    cmp    $0x2,%eax
   0x08048e4e <+42>:    jne    0x8048e5d <phase_4+57>
   0x08048e50 <+44>:    mov    0x18(%esp),%eax
   0x08048e54 <+48>:    test   %eax,%eax
   0x08048e56 <+50>:    js     0x8048e5d <phase_4+57>
   0x08048e58 <+52>:    cmp    $0xe,%eax
   0x08048e5b <+55>:    jle    0x8048e62 <phase_4+62>
   0x08048e5d <+57>:    call   0x8049470 <explode_bomb>
   0x08048e62 <+62>:    movl   $0xe,0x8(%esp)
   0x08048e6a <+70>:    movl   $0x0,0x4(%esp)
   0x08048e72 <+78>:    mov    0x18(%esp),%eax
   0x08048e76 <+82>:    mov    %eax,(%esp)
   0x08048e79 <+85>:    call   0x8048dbb <func4>
   0x08048e7e <+90>:    cmp    $0x25,%eax
   0x08048e81 <+93>:    jne    0x8048e8a <phase_4+102>
   0x08048e83 <+95>:    cmpl   $0x25,0x1c(%esp)
   0x08048e88 <+100>:   je     0x8048e8f <phase_4+107>
   0x08048e8a <+102>:   call   0x8049470 <explode_bomb>
   0x08048e8f <+107>:   add    $0x2c,%esp
   0x08048e92 <+110>:   ret    
    End of assembler dump.

func4:

Breakpoint 2, 0x08048dbb in func4 ()
(gdb) disas
Dump of assembler code for function func4:
=> 0x08048dbb <+0>: sub    $0x1c,%esp
   0x08048dbe <+3>: mov    %ebx,0x14(%esp)
   0x08048dc2 <+7>: mov    %esi,0x18(%esp)
   0x08048dc6 <+11>:    mov    0x20(%esp),%eax
   0x08048dca <+15>:    mov    0x24(%esp),%edx
   0x08048dce <+19>:    mov    0x28(%esp),%esi
   0x08048dd2 <+23>:    mov    %esi,%ecx
   0x08048dd4 <+25>:    sub    %edx,%ecx
   0x08048dd6 <+27>:    mov    %ecx,%ebx
   0x08048dd8 <+29>:    shr    $0x1f,%ebx
   0x08048ddb <+32>:    add    %ebx,%ecx
   0x08048ddd <+34>:    sar    %ecx
   0x08048ddf <+36>:    lea    (%ecx,%edx,1),%ebx
   0x08048de2 <+39>:    cmp    %eax,%ebx
   0x08048de4 <+41>:    jle    0x8048dfd <func4+66>
   0x08048de6 <+43>:    lea    -0x1(%ebx),%ecx
   0x08048de9 <+46>:    mov    %ecx,0x8(%esp)
   0x08048ded <+50>:    mov    %edx,0x4(%esp)
   0x08048df1 <+54>:    mov    %eax,(%esp)
   0x08048df4 <+57>:    call   0x8048dbb <func4>
   0x08048df9 <+62>:    add    %eax,%ebx
   0x08048dfb <+64>:    jmp    0x8048e16 <func4+91>
   0x08048dfd <+66>:    cmp    %eax,%ebx
   0x08048dff <+68>:    jge    0x8048e16 <func4+91>
   0x08048e01 <+70>:    mov    %esi,0x8(%esp)
   0x08048e05 <+74>:    lea    0x1(%ebx),%edx
   0x08048e08 <+77>:    mov    %edx,0x4(%esp)
   0x08048e0c <+81>:    mov    %eax,(%esp)
   0x08048e0f <+84>:    call   0x8048dbb <func4>
   0x08048e14 <+89>:    add    %eax,%ebx
   0x08048e16 <+91>:    mov    %ebx,%eax
   0x08048e18 <+93>:    mov    0x14(%esp),%ebx
   0x08048e1c <+97>:    mov    0x18(%esp),%esi
   0x08048e20 <+101>:   add    $0x1c,%esp
   0x08048e23 <+104>:   ret    
End of assembler dump.

Solution 1:

I hope it's obvious that phase4 is checking that the first number is in the range 0..14 inclusive (see lines +44..+57) Then it invokes func4 with three arguments: the first number entered, 0 and 14 (lines +62..+85). Next it checks that the return value is 0x25 (37 decimal) on line +90 and that the second number entered is also 37 (line +95)

Let's move on to func4. I'll call the three arguments x, low and high. Initially you don't know what they are of course. Lines +23..+34 calculate (high - low) / 2. The ugly mess is because the compiler generates code to handle negative numbers with truncation to zero. We won't see any negative numbers though. Line +36 is just a fancy addition, so in ebx we now have low + (high - low) / 2 which is also known as the average of the two numbers. The code then compares this average to the number x that has been provided as first argument. Lines +43..+62 get executed if x < average and they invoke func4(x, low, average - 1) and add the returned value to the average. Similarly, lines +70..+89 get executed if x > average and calculate average + func4(x, average + 1, high). If x == average then just the average itself is returned.

It's basically doing a binary search and summing up the guesses as it goes. Given that the interval has 15 elements, it will need at most 4 guesses. The first guess is going to be 7, so to get the required result of 37 we need 30 more. We have at most 3 more tries and all the guesses will be either less than 7 or more than 7. Since 7 * 3 = 21 and that can't give us 30 it means the number has to be greater than 7. Second guess is thus going to be (8 + 14) / 2 = 11, making our sum 18 with 19 more to go. If the number was above 11 that would mean we overshoot the target, so the number must be more than 7 and less than 11. Third guess is thus (8 + 10) / 2 = 9 which brings the sum to 27 with 10 more to go and just a single guess, so that means the number is 10.

TL;DR: the correct input should be 10 and 37