MSC32-C. Properly seed pseudorandom number generators

A pseudorandom number generator (PRNG) is a deterministic algorithm capable of generating sequences of numbers that approximate the properties of random numbers. Each sequence is completely determined by the initial state of the PRNG and the algorithm for changing the state. Most PRNGs make it possible to set the initial state, also called the seed state. Setting the initial state is called seeding the PRNG.

Calling a PRNG in the same initial state, either without seeding it explicitly or by seeding it with the same value, results in generating the same sequence of random numbers in different runs of the program. If a PRNG function is called 10 times consecutively to produce a sequence of 10 random numbers without being seeded, running the code for the first time produces the sequence S = {r1, r2, r3, r4, r5, r6, r7, r8, r9, r10}. If the PRNG is subsequently seeded with the same initial seed value, then it will generate the same sequentce S.

As a result, after the first run of an improperly seeded PRNG, an attacker can predict the sequence of random numbers that will be generated in the future runs. Improperly seeding or failing to seed the PRNG can lead to vulnerabilities, especially in security protocols.

The solution is to ensure that the PRNG is always properly seeded. A properly seeded PRNG will generate a different sequence of random numbers each time it is run.

Not all random number generators can be seeded. True random number generators that rely on hardware to produce completely unpredictable results do not need to be and cannot be seeded. Some high-quality PRNGs, such as the /dev/random device on some UNIX systems, also cannot be seeded. This rule applies only to algorithmic pseudorandom number generators that can be seeded.

Noncompliant Code Example (POSIX)

This noncompliant code example generates a sequence of 10 pseudorandom numbers using the random() function. When random() is not seeded, it behaves like rand(), producing the same sequence of random numbers each time any program that uses it is run.

#include <stdio.h>
#include <stdlib.h>
 
void func(void) {
  for (unsigned int i = 0; i < 10; ++i) {
    /* Always generates the same sequence */
    printf("%ld, ", random());
  }
}

The output is as follows:

1st run: 1804289383, 846930886, 1681692777, 1714636915, 1957747793, 424238335, 719885386, 1649760492, 596516649,
         1189641421,
2nd run: 1804289383, 846930886, 1681692777, 1714636915, 1957747793, 424238335, 719885386, 1649760492, 596516649,
         1189641421,
...
nth run: 1804289383, 846930886, 1681692777, 1714636915, 1957747793, 424238335, 719885386, 1649760492, 596516649,
         1189641421,

Compliant Solution (POSIX)

Call srandom() before invoking random() to seed the random sequence generated by random(). This compliant solution produces different random number sequences each time the program is run:

#include <stdio.h>
#include <stdlib.h>
#include <time.h>
 
void func(void) {
  struct timespec ts;
  if (timespec_get(&ts, TIME_UTC) == 0) {
    /* Handle error */
  }
  else {
    srandom(ts.tv_nsec ^ ts.tv_sec);
    for (unsigned int i = 0; i < 10; ++i) {
      /* Generates different sequences at different runs */
    printf("%ld, ", random());
  }
}

The output is as follows:

1st run: 198682410, 2076262355, 910374899, 428635843, 2084827500, 1558698420, 4459146, 733695321, 2044378618, 1649046624,
2nd run: 1127071427, 252907983, 1358798372, 2101446505, 1514711759, 229790273, 954268511, 1116446419, 368192457,
         1297948050,
3rd run: 2052868434, 1645663878, 731874735, 1624006793, 938447420, 1046134947, 1901136083, 418123888, 836428296,
         2017467418,

This may not be sufficiently random for concurrent execution, where it may lead to correlated generated series in different threads, or for small embedded systems that have an unsigned int type with a width of 16 bits. Note that the POSIX standard specifies a minimum width of 32 bits for unsigned int. Depending on the application and the desired level of security, a programmer may choose alternative ways to seed PRNGs. In general, hardware is more capable than software of generating real random numbers (for example, by sampling the thermal noise of a diode).

Compliant Solution (Windows)

The CryptGenRandom() function does not run the risk of not being properly seeded because its arguments serve as seeders:

#include <Windows.h>
#include <wincrypt.h>
#include <stdio.h>
 
void func(void) {
  HCRYPTPROV hCryptProv;
  long rand_buf;
  /* Example of instantiating the CSP */
  if (CryptAcquireContext(&hCryptProv, NULL, NULL,
                          PROV_RSA_FULL, 0)) {
    printf("CryptAcquireContext succeeded.\n");
  } else {
    printf("Error during CryptAcquireContext!\n");
  }

  for (unsigned int i = 0; i < 10; ++i) {
    if (!CryptGenRandom(hCryptProv, sizeof(rand_buf),
                        (BYTE *)&rand_buf)) {
      printf("Error\n");
    } else {
      printf("%ld, ", rand_buf);
    }
  }
}

The output is as follows:

1st run: -1597837311, 906130682, -1308031886, 1048837407, -931041900, -658114613, -1709220953, -1019697289, 1802206541,
         406505841,
2nd run: 885904119, -687379556, -1782296854, 1443701916, -624291047, 2049692692, -990451563, -142307804, 1257079211,
         897185104,
3rd run: 190598304, -1537409464, 1594174739, -424401916, -1975153474, 826912927, 1705549595, -1515331215, 474951399,
         1982500583,

Risk Assessment

Related Vulnerabilities

Search for vulnerabilities resulting from the violation of this rule on the CERT website.

Related Guidelines

Bibliography