★ ☆ ★   W E L C O M E   T O   M Y   H O M E P A G E   ★ ☆ ★

NULLRAKE'S PAGE

last updated: NOV 20 2026  ·  hosted on a Pentium 133  ·  28.8k friendly
  *** WELCOME 2 MY HOMEPAGE *** IF U R SEEING THIS, THE PAGE LOADED OK *** DO NOT ENTER KEYS U DONT OWN *** SIGN MY GUESTBOOK *** BEST VIEWED IN NETSCAPE NAVIGATOR 4.0 @ 800x600 *** TELL UR FRIENDS *** THIS SITE IS ALWAYS UNDER CONSTRUCTION ***
one byte is the whole licence check

I wrote the target for this so anybody can repeat it. It is a small Windows console app, the kind of licence gate that ships in thousands of paid desktop tools: the user types a key, the program checks it locally, and a boolean decides which message gets printed.

THE SHAPE OF THE CHECK. Key must be 19 characters, dashes in positions 4, 9 and 14, and the sum of every ASCII byte multiplied by the length has to equal 0x2A5E1. Fail any of those and the program prints Invalid licence key and exits 1.
WHAT THE TOOL SAID FIRST

route put it on pe_native in one call: PE32+, no packer, no .NET, 51 imports, seven sections, nothing suspicious. plan then returned three branch candidates with the string that sits beside each one. The top entry:

{
  "index": 0,
  "kind": "FAIL",
  "jump": "je",
  "bytes": "7422",
  "jcc_off": "0xc91",
  "ref_off": "0xc3f",
  "string": "Invalid licence key",
  "verified": true
}

That entry is wrong, and the disassembler was right to call it verified anyway. File offset 0xc91 lands in __do_global_dtors, a runtime teardown loop that has nothing to do with licences. It scored high because a string reference is nearby, which is exactly the sort of thing a proximity heuristic gets wrong on a compiler that packs static strings into one region.

WHAT THE BINARY ACTUALLY DOES

Ten seconds with objdump settles it. The comparison and the branch that decides which way the program goes:

1400015d6:  call   140007cd8 <strlen>
1400015db:  cmp    $0x13,%rax          ; length must be 19
1400015df:  jne    140001659           ; -> error path
1400015e1:  cmpb   $0x2d,0x24(%rsp)    ; dash at 4
1400015e6:  jne    140001659
1400015e8:  cmpb   $0x2d,0x29(%rsp)    ; dash at 9
1400015ed:  jne    140001659
1400015ef:  cmpb   $0x2d,0x2e(%rsp)    ; dash at 14
1400015f4:  jne    140001659
140001603:  movzbl (%rax,%r8,1),%ecx    ; checksum loop
140001608:  add    %ecx,%edx
14000160a:  add    $0x1,%rax
14000160e:  cmp    $0x13,%rax
140001612:  jne    140001603
140001614:  cmp    $0x943608bb,%edx    ; the multiplied sum
14000161a:  jne    140001659           ; <- this one decides it
14000161c:  lea    ...%rcx             ; "Licence accepted"
140001623:  call   140001530 <printf>

The branch at 14000161a is 75 3d: jne, jump forward into the error path when the checksum does not match. Everything above it is key format checking. Change that one byte to 74 and the conditional inverts, so the wrong key falls through into the paid path instead of jumping away from it.

THE PATCH AND THE PROOF
$ ree_ai.py patch '{"path":"invoicepro.exe","out":"invoicepro_patched.exe",
    "patches":[{"offset":"0xc1a","mode":"flip"}]}'

{"dst": "invoicepro_patched.exe",
 "results": [{"offset": "0xc1a", "mode": "flip",
              "before": "753d", "after": "743d", "status": "applied"}]}

$ ree_ai.py verify '{"orig":"invoicepro.exe","new":"invoicepro_patched.exe"}'

{"same_size": true, "regions_changed": 1,
 "diffs": [{"range": "0xc1a-0xc1a", "orig": "75", "new": "74"}]}

One byte, file length unchanged, nothing else touched. I could not run the PE here, there is no Windows host on this machine, so I rebuilt the same source as a Linux binary and repeated the exercise. Offset 0x1209, 75 36 to 74 36, same verify result, one region changed. Then I ran both:

$ ./invoicepro_linux <<< "AAAA-BBBB-CCCC-DDDD"
InvoicePro 3.2
Enter licence key: Invalid licence key        (exit 1)

$ ./invoicepro_linux_patched <<< "AAAA-BBBB-CCCC-DDDD"
InvoicePro 3.2
Enter licence key: Licence accepted. Pro features unlocked.
Batch export is available.                      (exit 0)

$ ./invoicepro_linux_patched <<< ""
InvoicePro 3.2
Enter licence key: Invalid licence key        (exit 1)

Note the third run. The empty key is still rejected, because the length check upstream is untouched. That is what a one byte patch looks like from the outside: the program still validates shape, it just no longer cares whether the numbers add up.

WHAT THIS MEANS IF YOU SHIP SOFTWARE

Total cost to defeat this licence model: one byte, and about ten minutes if you know where to look. No packer, no obfuscator and no anti-debug change that number by much, they only change the ten minutes. The fix is not a better local check, it is not having the decision live on the customer's machine at all. Ask the server, per request, and hand back only what was paid for.

REPRODUCE IT

The source is 45 lines of C and both binaries are built from it: mingw for the PE, gcc for the ELF. Hashes below are the exact files used here, so you can confirm you are looking at the same thing.

invoicepro.exe        8dbd1b21c20e14899ea5e98052cee3ea8a54a43c889d1c79c5402e845935d26e
invoicepro_patched     one byte: 0xc1a  75 -> 74  (same size: true)
[ THE TOOL: /REE ] [ HOW TO STOP THIS ]

✉ tip me (crypto) · notes & messages (SimpleX) · contact me · telegram bot · services
NETSCAPE NOW! IE 3.0 800x600 HTML 3.2
© 1997-2026 nullrake // this page is best viewed on a CRT // sign my guestbook