2021-05-24 20:44:55 +00:00
|
|
|
# INFORMATIONS
|
|
|
|
Organization: Technical University of Kosice (TUKE),
|
|
|
|
|
|
|
|
Department: Department of Electronics and Multimedia Telecommunications (DEMT/KEMT),
|
|
|
|
|
|
|
|
Faculties: Faculty of Electrical Engineering and Informatics (FEI),
|
|
|
|
|
|
|
|
Feld of study: Informatics,
|
|
|
|
|
|
|
|
Study program: Computer Networks,
|
|
|
|
|
|
|
|
School year: 3., Bachelor study, 2020/2021,
|
|
|
|
|
|
|
|
Author: Marek Rohac -- MR,
|
|
|
|
|
|
|
|
Compiler: Winlibs GCC -- MinGW-W64 x86_64-posix-seh, built by Brecht Sanders, v. 10.2.0,
|
|
|
|
-- works also with v.11.1.0
|
|
|
|
|
|
|
|
Version: 1.3, 20.05.2021.
|
|
|
|
# COMPILING
|
|
|
|
## VIA MAKEFILE
|
|
|
|
|
|
|
|
For compiling all source codes i prepare makefile, so just use command:
|
|
|
|
|
|
|
|
```bash
|
|
|
|
make all
|
|
|
|
```
|
|
|
|
That will create .exe files. For more details see makefile.
|
|
|
|
|
|
|
|
## SELECTIVE COMPILING
|
|
|
|
|
|
|
|
winAPIprng:
|
|
|
|
```bash
|
|
|
|
gcc winAPIprng.c -o winAPIprng -Wall -Wextra -Werror -g -std=c11 -lbcrypt
|
|
|
|
```
|
|
|
|
amdSPRNG:
|
|
|
|
```bash
|
|
|
|
gcc ./amdINC/secrng.c amdSPRNG.c -Wall -Wextra -Werror -g -I./amdINC -mrdrnd -mrdseed -o amdSPRNG -std=c11
|
|
|
|
```
|
|
|
|
opessl_rng:
|
|
|
|
```bash
|
|
|
|
gcc -o openssl_rng openssl_rng.c -Wall -Wextra -g -I./openssl/include -L./openssl/lib -llibcrypto
|
|
|
|
```
|
|
|
|
|
|
|
|
## REMOVAL
|
|
|
|
|
|
|
|
For removing all executables and files which was generated by executions these programs use command:
|
|
|
|
|
|
|
|
```bash
|
|
|
|
make clean
|
|
|
|
```
|
|
|
|
That remove all file with .exe , .bin and .txt suffixes
|
|
|
|
|
|
|
|
# USAGE
|
|
|
|
NO INPUT PARAMATERS IN EACH CASE
|
|
|
|
|
|
|
|
## WINDOWS API PRNG - winAPIprng.c
|
|
|
|
```c
|
|
|
|
39 #define RNG_PER_CYKLE_BITS 268435456 //per API call output size
|
|
|
|
41 #define RNG_CYCLES 1024 //total count of calls
|
|
|
|
```
|
|
|
|
|
|
|
|
In macro RNG_PER_CYKLE_BITS define how much output in bits you want from one call of API and in RNG_CYCLES number of api calls.
|
|
|
|
|
|
|
|
Program generate output according APIs and save values to .bin files with name by current API.
|
|
|
|
|
|
|
|
If you want to generate with specific API, just comment all other functions im main().
|
|
|
|
|
|
|
|
### MEASURMENT
|
|
|
|
For measurment of API i use function cpucyclesS() and cpucyclesE(). They returns numbers of cycles of current API.
|
|
|
|
Speed or time of execution we then can calculate via formula:
|
|
|
|
|
|
|
|
Total_cpycycles_of_API * (1/Processor_ghz_during_execution_API) * 10^-9 = Time_of_executions s
|
|
|
|
|
|
|
|
On modern processors its hard to determine how much GHz was used during executions(power plans,turbo modes,...).
|
|
|
|
Therefore i also implemented timer of execution, but keep in mind, it counts also with storing of random data.
|
|
|
|
Results are store in Results.txt file.
|
|
|
|
|
|
|
|
### NOTE FOR MEASURMENT
|
|
|
|
If you want better precision of Total execution time, don't save random data (comment function store_Data()).
|
|
|
|
Also for absolutely same conditions for every api, you should wait certain time before calling next api.
|
|
|
|
Reason is to give system time to retrieve respectively gain more data
|
|
|
|
into entropy pool from system entropy sources. (1 hours to full regen).
|
|
|
|
Also random data have much better quality after giving a time to regen entropy.
|
|
|
|
|
|
|
|
### TIME OF EXECUTIONS
|
|
|
|
Time of executions of program strongly depend on your procesor and output size.
|
|
|
|
If you don't want spend to much time lover RNG_OUTPUT_BITS (do it correctly).
|
|
|
|
|
|
|
|
WIN API for measurment is used via macros.
|
|
|
|
Experimental time results on my computer (see SPEC),if you run all api
|
|
|
|
in program with storing random data:
|
|
|
|
|
|
|
|
1,875GB output size --> CryptGenRandom --> EXECUTED in 1.752000 s
|
|
|
|
BCryptGenRandom --> EXECUTED in 21.232000 s
|
|
|
|
RtlGenRandom --> EXECUTED in 23.052000 s
|
2021-05-27 22:53:37 +00:00
|
|
|
rand_s --> EXECUTED in 45.193655 s
|
|
|
|
|
2021-05-24 20:44:55 +00:00
|
|
|
|
|
|
|
SPEED OF EXECUTION IS ALSO INFLUENCED BY WRITING SPEED OF YOUR STORAGE DEVICE
|
|
|
|
(In case store_Data() is enable)
|
|
|
|
### NOTE FOR OUTPUT SIZE
|
|
|
|
To count output size in GB use formula:
|
|
|
|
|
|
|
|
--> RNG_OUTPUT_BITS/RNG_PER_CYKLE_BITS/8/1000/1000 = output_size_in_GB
|
|
|
|
|
|
|
|
For faster execution downsize macros
|
|
|
|
|
|
|
|
## AMD SECURE PRNG - amdSPRNG.c
|
|
|
|
Program use 4 AMD APIs (defined in secrng.h, realized in secrng.c)
|
|
|
|
to generate random bytes:
|
|
|
|
|
|
|
|
```c
|
|
|
|
is_RDRAND_supported();
|
|
|
|
is_RDSEED_supported();
|
|
|
|
get_rdseed_bytes_arr(unsigned char *rng_arr, unsigned int N, unsigned int retry_count)
|
|
|
|
get_rdrand_bytes_arr(unsigned char *rng_arr, unsigned int N, unsigned int retry_count)
|
|
|
|
```
|
|
|
|
|
|
|
|
Version of AMD libraries: 3.0 -- March, 2021
|
|
|
|
|
|
|
|
### DESCRIPTION
|
|
|
|
In source code, amdSPRNG.c, macros
|
|
|
|
|
|
|
|
```c
|
|
|
|
29 #define CYCLES 1000 // calls count
|
|
|
|
31 #define N 625000 //per API call output size
|
|
|
|
36 #define NT 10 //retry times
|
|
|
|
```
|
|
|
|
|
|
|
|
N define how much random bytes you want per one generating of certain API.
|
|
|
|
CYCLES defined how much times you want to repeat generating process of that API.
|
|
|
|
|
|
|
|
Program then generate output according APIs and save values
|
|
|
|
to .bin files with name by current RD name.
|
|
|
|
|
|
|
|
IN CASE OF PRINTING FAILURE TO STDOUT, RAISE NT macro.
|
|
|
|
Maximum of retry attempts is set in secrng.c,
|
|
|
|
under MAX_RETRY_COUNT on 51 line. (equal 20, default 10).
|
|
|
|
|
|
|
|
For informations about file output size check NOTE FOR OUTPUT SIZE
|
|
|
|
|
|
|
|
### MEASURMENT
|
|
|
|
Look on MEASURMENT in WINDOWS API PRNG.
|
|
|
|
Results are store in RDResults.txt file.
|
|
|
|
For better precision comment checking confition and storing data.
|
|
|
|
NOTICE: NO REGEN TIME NEEDED FOR REGAINING ENTROPY CAUSE OF USING TRNG.
|
|
|
|
|
|
|
|
### TIME OF EXECUTIONS
|
|
|
|
Look on TIME OF EXECUTION in WINDOWS API PRNG. Same principle.
|
|
|
|
Note that these will take longer time to execute program.
|
|
|
|
It depend on NT times, your processor and writing speed of storage device.
|
|
|
|
|
|
|
|
Experimental time results (same hardware & software like above):
|
|
|
|
|
|
|
|
0,625GB output size --> RDSEED --> EXECUTED in 319.976000 s
|
|
|
|
RDRAND --> EXECUTED in 71.733000 s
|
|
|
|
|
|
|
|
### NOTE FOR OUTPUT SIZE
|
|
|
|
To count output size in GB use formula:
|
|
|
|
|
|
|
|
--> N*CYCLES/1024/1024/1024 = output_size_in_GB
|
|
|
|
|
|
|
|
## OPENSSL RNG APIs 64-bit libraries used
|
|
|
|
We implement 2 API --> RAND_bytes and RAND_priv_bytes. To right function don't remove libcrypto dll.
|
|
|
|
|
|
|
|
Metods implemented in winAPIprng are use there also.
|
|
|
|
|
|
|
|
Time of executions is litle bit longer than win api.
|
|
|
|
|
|
|
|
Results are available in Ossl_rng.txt
|
|
|
|
|
|
|
|
## !!ALWAYS CHECK OUTPUT IN TERMINAL!!
|
|
|
|
## REFERENCIES
|
|
|
|
AMD RNG API Libraries + examples
|
|
|
|
|
|
|
|
--> https://developer.amd.com/amd-aocl/rng-library/
|
|
|
|
|
|
|
|
WIN API Documentations:
|
|
|
|
|
|
|
|
Deprecated
|
|
|
|
|
|
|
|
CryptGenRandom() --> https://docs.microsoft.com/en-us/windows/win32/api/wincrypt/nf-wincrypt-cryptgenrandom
|
|
|
|
RtlGenRandom() --> https://docs.microsoft.com/en-us/windows/win32/api/ntsecapi/nf-ntsecapi-rtlgenrandom
|
|
|
|
|
|
|
|
Currently supported:
|
|
|
|
|
|
|
|
BCryptGenRandom() --> https://docs.microsoft.com/en-us/windows/win32/api/bcrypt/nf-bcrypt-bcryptgenrandom
|
|
|
|
rand_s() --> https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/rand-s?view=msvc-160
|
|
|
|
|
|
|
|
MEASURMENT FUNCTIONS:
|
|
|
|
|
|
|
|
--> https://github.com/newhopecrypto/newhope/tree/master/ref
|
|
|
|
|
|
|
|
WINLIBS GCC
|
|
|
|
|
|
|
|
--> http://winlibs.com/
|
|
|
|
|
|
|
|
OPENSSL LIBRARIES
|
|
|
|
--> https://curl.se/windows/
|
|
|
|
|
|
|
|
(See specification--> OpenSSL )
|