Bytes::Random::Secure(3) User Contributed Perl Documentation Bytes::Random::Secure(3)

Bytes::Random::Secure - Perl extension to generate cryptographically-secure random bytes.

use Bytes::Random::Secure qw(
    random_bytes random_bytes_base64 random_bytes_hex
my $bytes = random_bytes(32); # A string of 32 random bytes.
my $bytes = random_string_from( 'abcde', 10 ); # 10 random a,b,c,d, and e's.
my $bytes_as_base64 = random_bytes_base64(57); # Base64 encoded rand bytes.
my $bytes_as_hex = random_bytes_hex(8); # Eight random bytes as hex digits.
my $bytes_as_quoted_printable = random_bytes_qp(100); # QP encoded bytes.
my $random = Bytes::Random::Secure->new(
    Bits        => 64,
    NonBlocking => 1,
); # Seed with 64 bits, and use /dev/urandom (or other non-blocking).
my $bytes = $random->bytes(32); # A string of 32 random bytes.
my $long  = $random->irand;     # 32-bit random integer.

Bytes::Random::Secure provides two interfaces for obtaining crypto-quality random bytes. The simple interface is built around plain functions. For greater control over the Random Number Generator's seeding, there is an Object Oriented interface that provides much more flexibility.

The "functions" interface provides functions that can be used any time you need a string of a specific number of random bytes. The random bytes are available as simple strings, or as hex-digits, Quoted Printable, or MIME Base64. There are equivalent methods available from the OO interface, plus a few others.

This module can be a drop-in replacement for Bytes::Random, with the primary enhancement of using a cryptographic-quality random number generator to create the random data. The "random_bytes" function emulates the user interface of Bytes::Random's function by the same name. But with Bytes::Random::Secure the random number generator comes from Math::Random::ISAAC, and is suitable for cryptographic purposes. The harder problem to solve is how to seed the generator. This module uses Crypt::Random::Seed to generate the initial seeds for Math::Random::ISAAC.

In addition to providing random_bytes(), this module also provides several functions not found in Bytes::Random: "random_string_from", random_bytes_base64(), "random_bytes_hex", and "random_bytes_qp".

And finally, for those who need finer control over how Crypt::Random::Seed generates its seed, there is an object oriented interface with a constructor that facilitates configuring the seeding process, while providing methods that do everything the "functions" interface can do (truth be told, the functions interface is just a thin wrapper around the OO version, with some sane defaults selected). The OO interface also provides an "irand" method, not available through the functions interface.

There are many uses for cryptographic quality randomness. This module aims to provide a generalized tool that can fit into many applications while providing a minimal dependency chain, and a user interface that is simple. You're free to come up with your own use-cases, but there are several obvious ones:

  • Creating temporary passphrases (random_string_from()).
  • Generating per-account random salt to be hashed along with passphrases (and stored alongside them) to prevent rainbow table attacks.
  • Generating a secret that can be hashed along with a cookie's session content to prevent cookie forgeries.
  • Building raw cryptographic-quality pseudo-random data sets for testing or sampling.
  • Feeding secure key-gen utilities.

Why use this module? This module employs several well-designed CPAN tools to first generate a strong random seed, and then to instantiate a high quality random number generator based on the seed. The code in this module really just glues together the building blocks. However, it has taken a good deal of research to come up with what I feel is a strong tool-chain that isn't going to fall back to a weak state on some systems. The interface is designed with simplicity in mind, to minimize the potential for misconfiguration.

By default "random_bytes" is the only function exported. Optionally "random_string_from", "random_bytes_base64", "random_bytes_hex", and "random_bytes_qp" may be exported.

The functions interface seeds the ISAAC generator on first use with a 256 bit seed that uses Crypt::Random::Seed's default configuration as a strong random seed source.

my $random_bytes = random_bytes( 512 );

Returns a string containing as many random bytes as requested. Obviously the string isn't useful for display, as it can contain any byte value from 0 through 255.

The parameter is a byte-count, and must be an integer greater or equal to zero.

my $random_bytes = random_string_from( $bag, $length );
my $random_bytes = random_string_from( 'abc', 50 );

$bag is a string of characters from which "random_string_from" may choose in building a random string. We call it a 'bag', because it's permissible to have repeated chars in the bag (if not, we could call it a set). Repeated digits get more weight. For example, "random_string_from( 'aab', 1 )" would have a 66.67% chance of returning an 'a', and a 33.33% chance of returning a 'b'. For unweighted distribution, ensure there are no duplicates in $bag.

This isn't a "draw and discard", or a permutation algorithm; each character selected is independent of previous or subsequent selections; duplicate selections are possible by design.

Return value is a string of size $length, of characters chosen at random from the 'bag' string.

It is perfectly legal to pass a Unicode string as the "bag", and in that case, the yield will include Unicode characters selected from those passed in via the bag string.

This function is useful for random string generation such as temporary random passwords.

my $random_bytes_b64           = random_bytes_base64( $num_bytes );
my $random_bytes_b64_formatted = random_bytes_base64( $num_bytes, $eol );

Returns a MIME Base64 encoding of a string of $number_of_bytes random bytes. Note, it should be obvious, but is worth mentioning that a base64 encoding of base256 data requires more digits to represent the bytes requested. The actual number of digits required, including padding is 4(n/3). Furthermore, the Base64 standard is to add padding to the end of any string for which "length % 57" is a non-zero value.

If an $eol is specified, the character(s) specified will be used as line delimiters after every 76th character. The default is "qq{\n}". If you wish to eliminate line-break insertions, specify an empty string: "q{}".

my $random_bytes_as_hex = random_bytes_hex( $num_bytes );

Returns a string of hex digits representing the string of $number_of_bytes random bytes.

It's worth mentioning that a hex (base16) representation of base256 data requires two digits for every byte requested. So "length( random_bytes_hex( 16 ) )" will return 32, as it takes 32 hex digits to represent 16 bytes. Simple stuff, but better to mention it now than forget and set a database field that's too narrow.

my $random_bytes_qp           = random_bytes_qp( $num_bytes );
my $random_bytes_qp_formatted = random_bytes_qp( $num_bytes, $eol );

Produces a string of $num_bytes random bytes, using MIME Quoted Printable encoding (as produced by MIME::QuotedPrint's "encode_qp" function. The default configuration uses "\n" as a line break after every 76 characters, and the "binmode" setting is used to guarantee a lossless round trip. If no line break is wanted, pass an empty string as $eol.

The Object Oriented interface provides methods that mirror the "functions" interface. However, the OO interface offers the advantage that the user can control how many bits of entropy are used in seeding, and even how Crypt::Random::Seed is configured.

my $random = Bytes::Random::Secure->new( Bits => 512 );
my $bytes  = $random->bytes( 32 );

The constructor is used to specify how the ISAAC generator is seeded. Future versions may also allow for alternate CSPRNGs to be selected. If no parameters are passed the default configuration specifies 256 bits for the seed. The rest of the default configuration accepts the Crypt::Random::Seed defaults, which favor the strongest operating system provided entropy source, which in many cases may be "blocking".



my $random = Bytes::Random::Secure->new( Bits => 128 );

The "Bits" parameter specifies how many bits (rounded up to nearest multiple of 32) will be used in seeding the ISAAC random number generator. The default is 256 bits of entropy. But in some cases it may not be necessary, or even wise to pull so many bits of entropy out of "/dev/random" (a blocking source).

Any value between 64 and 8192 will be accepted. If an out-of-range value is specified, or a value that is not a multiple of 32, a warning will be generated and the parameter will be rounded up to the nearest multiple of 32 within the range of 64 through 8192 bits. So if 16384 is specified, you will get 8192. If 33 is specified, you will get 64.

Note: In the Perlish spirit of "no arbitrary limits", the maximum number of bits this module accepts is 8192, which is the maximum number that ISAAC can utilize. But just because you can specify a seed of 8192 bits doesn't mean you ought to, much less need to. And if you do, you probably want to use the "NonBlocking" option, discussed below. 8192 bits is a lot to ask from a blocking source such as "/dev/random", and really anything beyond 512 bits in the seed is probably wasteful.


Reserved for future use. Eventually the user will be able to select other RNGs aside from Math::Random::ISAAC.


Reserved for future use.

Other Crypt::Random::Seed Configuration Parameters

For additional seeding control, refer to the POD for Crypt::Random::Seed. By supplying a Crypt::Random::Seed parameter to Bytes::Random::Secure's constructor, it will be passed through to Crypt::Random::Seed. For example:

my $random = Bytes::Random::Secure->new( NonBlocking => 1, Bits => 64 );

In this example, "Bits" is used internally, while "NonBlocking" is passed through to Crypt::Random::Seed.

my $random_bytes = $random->bytes(1024);

This works just like the "random_bytes" function.

my $random_string = $random->string_from( 'abcdefg', 10 );

Just like "random_string_from": Returns a string of random octets selected from the "Bag" string (in this case ten octets from 'abcdefg').

my $random_hex = $random->bytes_hex(12);

Identical in function to "random_bytes_hex".

my $random_base64 = $random->bytes_base64( 32, EOL => "\n" );

Identical in function to "random_bytes_base64".

my $random_qp = $random->bytes_qp( 80 );

You guessed it: Identical in function to "random_bytes_qp".

my $unsigned_long = $random->irand;

Returns a random 32-bit unsigned integer. The value will satisfy "0 <= x <= 2**32-1". This functionality is only available through the OO interface.

my $aref_shuffled = $random->shuffle($aref);

Shuffles the contents of a reference to an array in sitiu, and returns the same reference.

List::Util, which ships with Perl, includes "shuffle" function. But that function is flawed in two ways. First, from a cryptographic standpoint, it uses Perl's "rand", which is not a CSPRNG, and therefore is inadequate.

Second, because Perl's rand has an internal state of just 32 bits, it cannot possibly generate all permutations of arrays containing 13 or more elements.

This module's "shuffle" uses a CSPRNG, and also benefits from large seeds and a huge internal state. ISAAC can be seeded with up to 8192 bits, yielding 2^8192 possible initial states, and 2^8288 possible internal states. A seed of 8192 bits will assure that for arrays of up to 966 elements every permutation is accessible.

Bytes::Random::Secure's interface tries to keep it simple. There is generally nothing to configure. This design, eliminates much of the potential for diminishing the quality of the random byte stream through misconfiguration. The ISAAC algorithm is used as our factory, seeded with a strong source.

There may be times when the default seed characteristics carry too heavy a burden on system resources. The default seed for the functions interface is 256 bits of entropy taken from /dev/random (a blocking source on many systems), or via API calls on Windows. The default seed size for the OO interface is also 256 bits. If /dev/random should become depleted at the time that this module attempts to seed the ISAAC generator, there could be delay while additional system entropy is generated. If this is a problem, it is possible to override the default seeding characteristics using the OO interface instead of the functions interface. However, under most circumstances, this capability may be safely ignored.

Beginning with Bytes::Random::Secure version 0.20, Crypt::Random::Seed provides our strong seed (previously it was Crypt::Random::Source). This module gives us excellent "strong source" failsafe behavior, while keeping the non-core dependencies to a bare minimum. Best of all, it performs well across a wide variety of platforms, and is compatible with Perl versions back through 5.6.0.

And as mentioned earlier in this document, there may be circumstances where the performance of the operating system's strong random source is prohibitive from using the module's default seeding configuration. Use the OO interface instead, and read the documentation for Crypt::Random::Seed to learn what options are available.

Prior to version 0.20, a heavy dependency chain was required for reliably and securely seeding the ISAAC generator. Earlier versions required Crypt::Random::Source, which in turn required Any::Moose. Thanks to Dana Jacobsen's new Crypt::Random::Seed module, this situation has been resolved. So if you're looking for a secure random bytes solution that "just works" portably, and on Perl versions as far back as 5.6.0, you've come to the right place. Users of older versions of this module are encouraged to update to version 0.20 or higher to benefit from the improved user interface and lighter dependency chain.

If performance is a consideration, you may also install Math::Random::ISAAC::XS. Bytes::Random::Secure's random number generator uses Math::Random::ISAAC. That module implements the ISAAC algorithm in pure Perl. However, if you install Math::Random::ISAAC::XS, you get the same algorithm implemented in C/XS, which will provide better performance. If you need to produce your random bytes more quickly, simply installing Math::Random::ISAAC::XS will result in it automatically being used, and a pretty good performance improvement will coincide.

When programming for parallel computation, avoid the "functions" interface do use the Object Oriented interface, and create a unique "Bytes::Random::Secure" object within each process or thread. Bytes::Random::Secure uses a CSPRNG, and sharing the same RNG between threads or processes will share the same seed and the same starting point. This is probably not what one would want to do. By instantiating the B::R::S object after forking or creating threads, a unique randomness stream will be created per thread or process.

It's easy to generate weak pseudo-random bytes. It's also easy to think you're generating strong pseudo-random bytes when really you're not. And it's hard to test for pseudo-random cryptographic acceptable quality. There are many high quality random number generators that are suitable for statistical purposes, but not necessarily up to the rigors of cryptographic use.

Assuring strong (ie, secure) random bytes in a way that works across a wide variety of platforms is also challenging. A primary goal for this module is to provide cryptographically secure pseudo-random bytes. A secondary goal is to provide a simple user experience (thus reducing the propensity for getting it wrong). A tertiary goal is to minimize the dependencies required to achieve the primary and secondary goals, to the extent that is practical.

The ISAAC algorithm is considered to be a cryptographically strong pseudo-random number generator. There are 1.0e2466 initial states. The best known attack for discovering initial state would theoretically take a complexity of approximately 4.67e1240, which has no practical impact on ISAAC's security. Cycles are guaranteed to have a minimum length of 2**40, with an average cycle of 2**8295. Because there is no practical attack capable of discovering initial state, and because the average cycle is so long, it's generally unnecessary to re-seed a running application. The results are uniformly distributed, unbiased, and unpredictable unless the seed is known.

To confirm the quality of the CSPRNG, this module's test suite implements the FIPS-140-1 tests for strong random number generators. See the comments in "t/27-fips140-1.t" for details.

To keep the dependencies as light as possible this module uses some ideas from Math::Random::Secure. That module is an excellent resource, but implements a broader range of functionality than is needed here. So we just borrowed from it.

The primary source of random data in this module comes from the excellent Math::Random::ISAAC. To be useful and secure, even Math::Random::ISAAC needs a cryptographically sound seed, which we derive from Crypt::Random::Seed. There are no known weaknesses in the ISAAC algorithm. And Crypt::Random::Seed does a very good job of preventing fall-back to weak seed sources.

This module requires Perl 5.6 or newer. The module also uses a number of core modules, some of which require newer versions than those contemporary with 5.6. Unicode support in "random_string_from" is best with Perl 5.8.9 or newer. See the INSTALLATION section in this document for details.

If Test::Warn is installed, test coverage is 100%. For those who don't want to bother installing Test::Warn, you can just take our word for it. It's an optional installation dependency.

It is possible (and has been seen in testing) that the system's random entropy source might not have enough entropy in reserve to generate the seed requested by this module without blocking. If you suspect that you're a victim of blocking from reads on "/dev/random", one option is to manipulate the random seed configuration by using the object oriented interface.

This module seeds as lazily as possible so that using the module, and even instantiating a Bytes::Random::Secure object will not trigger reads from "/dev/random". Only the first time the object is used to deliver random bytes will the RNG be seeded. Long-running scripts may prefer to force early seeding as close to start-up time as possible, rather than allowing it to happen later in a program's run-time. This can be achieved simply by invoking any of the functions or methods that return a random byte. As soon as a random byte is requested for the first time, the CSPRNG will be seeded.

The "random_string_from" function, and "string_from" method permit the user to pass a "bag" (or source) string containing Unicode characters. For any modern Perl version, this will work just as you would hope. But some versions of Perl older than 5.8.9 exhibited varying degrees of bugginess in their handling of Unicode. If you're depending on the Unicode features of this module while using Perl versions older than 5.8.9 be sure to test thoroughly, and don't be surprised when the outcome isn't as expected. ...this is to be expected. Upgrade.

No other functions or methods in this module get anywhere near Perl's Unicode features. So as long as you're not passing Unicode source strings to "random_string_from", you have nothing to worry about, even if you're using Perl 5.6.0.

Care is taken so that there is no modulo bias in the randomness returned either by "random_bytes" or its siblings, nor by "random_string_from". As a matter if fact, this is exactly why the "random_string_from" function is useful. However, the algorithm to eliminate modulo bias can impact the performance of the "random_string_from" function. Any time the length of the bag string is significantly less than the nearest greater or equal factor of 2**32, performance will degrade. Unfortunately there is no known algorithm that improves upon this situation. Fortunately, for sanely sized strings, it's a minor issue. To put it in perspective, even in the case of passing a "bag" string of length 2**31 (which is huge), the expected time to return random bytes will only double. Given that the entire Unicode range is just over a million possible code-points, it seems unlikely that the normal use case would ever have to be concerned with the performance of the "random_string_from" function.

This module should install without any fuss on modern versions of Perl. For older Perl versions (particularly 5.6 and early 5.8.x's), it may be necessary to update your CPAN installer to a more modern version before installing this this module.

Another alternative for those with old Perl versions who don't want to update their CPAN installer (You must know you're crazy, right?): Review "Makefile.PL" and assure that you've got the dependencies listed under "PREREQ_PM" and "BUILD_REQUIRES", in at least the minimum versions specified. Then proceed as usual.

This module only has two non-Core dependencies. But it does expect that some of the Core dependencies are newer than those supplied with 5.6 or early 5.8's. If you keep your CPAN installer up-to-date, you shouldn't have to think about this, as it will usually just "do the right thing", pulling in newer dependency versions as directed by the module's META files.

Test coverage for Bytes::Random::Secure is 100% (per Devel::Cover) on any system that has Test::Warn installed. But to keep the module light-weight, Test::Warn is not dragged in by default at installation time.

Math::Random::Secure and Crypt::Random provide strong CSPRINGs and even more configuration options, but come with hefty toolchains.

Bytes::Random::Secure::Tiny is a stand-alone adaptation of Bytes::Random::Secure with no dependencies. It will, however, detect if Math::Random::ISAAC, Math::Random::ISAAC::XS, and Crypt::Random::Seed are installed on the target system, and if they are, it quietly upgrades to using them.

David Oswald "<davido [at] cpan (dot) org>"

Please report any bugs or feature requests to "bug-bytes-random-secure at", or through the web interface at I will be notified, and then you'll automatically be notified of progress on your bug as I make changes.

You can find documentation for this module with the perldoc command.

perldoc Bytes::Random::Secure

You can also look for information at:

Dana Jacobsen ( <> ) for his work that led to Crypt::Random::Seed, thereby significantly reducing the dependencies while improving the portability and backward compatibility of this module. Also for providing a patch to this module that greatly improved the performance of "random_bytes".

Dana Jacosen also provided extensive input, code reviews, and testing that helped to guide the direction this module has taken. The code for the FIPS-140-1 tests was taken directly from Crypt::Random::TESHA2. Thanks!

Bytes::Random for implementing a nice, simple interface that this module patterns itself after.

Copyright 2012 David Oswald.

This program is free software; you can redistribute it and/or modify it under the terms of either: the GNU General Public License as published by the Free Software Foundation; or the Artistic License.

See for more information.

2023-07-25 perl v5.38.0