haochenuw/seal-expand-example

Environments

Ubuntu 16.04, g++-7.5.0, cmake-3.13.0, g++-7.5.0

How to run

• Go to either 3.2.0 or 3.5.6 directories, this refers to the version number of SEAL
• Run cmake .
• make
• ./expand

What this example is about

• I tried to use the expand_query function in sealPIR, in the main program, I generated a plaintext of "Ax^1", then encrypt it and ran expand_query on this ciphertext.

• For simplicity, here I only expanded this into 4 ciphertexts instead of the number of poly modulus. Then I decrypted the ciphertexts to see the results. Correct output should be 0, 40, 0, 0.

• I repeated the procedures above for twice in the main function as a for loop.

Issues encountered

• In SEAL 3.2.0, every time I will get the correct output, however, when the program exited, and I believe that is when the memorypool started cleaning up everyting, it showed a segmentation fault bug.

• In SEAL 3.5.6, for the first run, the expansion will get the correct output, but for the second run, results will become incorrect. I decrypted some ciphertexts during the execution of expand_query to see where the ciphertext goes wrong. And it turns out that at line 131, we executed multiply_power_of_X(temp[a], tempctxt_shifted, index_raw, params);, but newtemp[a] changes after that execution, although I believe this function has no impact on newtemp[a].

  You can check that this by seeing the output when running expand.

  You should see something similar to this during the second run:

  0 (c0 + sub) to string: 0

  0 after mul new temp: 3093AEE8x^8191 + 3D45D00Bx^8190 + 2A634A1x^8189 + 3CE3777Fx^8188 + 2463A898x^8187 + 1C662308x^8186...

  These two are the output of line 127 and line 137 where we output the plaintext after decrypting newtemp[a].

Some guess about this

• This might be related to the memorypool part, since it seems to have some memory corruptions?