Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.lang.c++ > #70409

Re: One agnostic random function for big project (4 GB of code)

Show all headers | View raw

On 12.12.2019 13:53, Frederick Gotham wrote:
>
> I'm working on a large embedded Linux project with about 4 gigs of source code.
>
> I'm trying to create one header file that can be included in any sub-project, which will have a function for generating random bytes of data. The only two requirement of using this random number generator will be:
> (1) You must include "agnostic_random.hpp" in the source file(s) where it's used
> (2) When linking, you must link with "libdl.so"
>
> In this header file, I'm taking advantage of the "inline" keyword because of its unique linking characteristics, specifically that I can have a static object inside an inline function and it will only exist once (and it will only be initialised once for the current process).
>
> This random number generator must be fully thread-safe.
>
> At the moment I've written it to use "RAND_bytes" provided by OpenSSL, but this might change later (e.g. I might use tpm2-tss directly).
>
> Please give me a critique of this header file and let me know if I've overlooked anything.

TBH, at the moment it looks like you are trying to sneak in "accidental" 
back-doors into this large Linux project. More below.

>
> #ifndef HPP_INCLUSION_GUARD_AGNOSTIC_RANDOM
> #define HPP_INCLUSION_GUARD_AGNOSTIC_RANDOM
>
> #ifndef __cplusplus
> #	error "This is a C++ header file but you've included it in a C source file (or you've used a C compiler on a C++ source file). Don't do that."
> #endif
>
> #include <cstdint>  /* uintmax_t */
>
> #include <atomic>   /* atomic_flag */
> #include <limits>   /* numeric_limits<int> */
>
> // The next two lines are declarations of functions found in
> // "libdl.so". (Unfortunately the compiler won't let me put
> // these declarations inside the body of a function).
> extern "C" void *dlopen(char const *, int);
> extern "C" void *dlsym(void *,char const *);
>

Namespace is missing. All C++ code should be in a namespace, especially 
if used so widely.

> inline void Agnostic_Random(void *pvoid, std::uintmax_t len_bytes)  // len_bytes can be 0

Buffer length should be in size_t, not some random type.

> {
> 	static std::atomic_flag is_not_first_time = ATOMIC_FLAG_INIT;
>
> 	typedef int (*RAND_bytes_t)(char unsigned *, int);
> 	static std::atomic<RAND_bytes_t> ptr_to_RAND_bytes(nullptr);
> 	
> 	if ( !is_not_first_time.test_and_set() )
> 	{
> 		void *const handle = dlopen("libcrypto.so", 1);  /* 1 == RTLD_LAZY */

The handle is leaked. There should be a matching call to dlclose(). This 
could be achieved pretty easily by having a static C++ class object with 
a destructor.

> 		
> 		if ( nullptr != handle )
> 			ptr_to_RAND_bytes = reinterpret_cast<RAND_bytes_t>(  dlsym(handle, "RAND_bytes")  );
> 	}

A race condition. There is a time slice when the flag is set, but the 
function pointer is not yet assigned. If another thread calls this 
function during this time, it will see a wrong (NULL) function pointer.

Instead suggesting a simple:

static RAND_bytes_t ptr_to_RAND_bytes = SomeInitFunction();

This is guaranteed to do the correct thing since C++11.

> 	if ( nullptr == ptr_to_RAND_bytes )
> 		return;

No error return or exception throw. Extremely bad style, especially if 
this has anything to do with crypto.

> 		
> 	while ( 0 != len_bytes )
> 	{
> 		int bytes_this_time;  // RAND_bytes unfortunately takes an 'int' for length
> 		
> 		if ( len_bytes > std::numeric_limits<int>::max() )

Signed-unsigned comparison, compilers will produce warnings.

> 			bytes_this_time = std::numeric_limits<int>::max();
> 		else
> 			bytes_this_time = len_bytes;
>
> 		ptr_to_RAND_bytes( static_cast<char unsigned *>(pvoid), bytes_this_time );

No error check! Extremely bad style, especially in anything related to 
crypto!

> 		
> 		pvoid = static_cast<char*>(pvoid) + bytes_this_time;  // Cast to char pointer for arithmetic
> 		len_bytes -= bytes_this_time;
> 	}
> }
>
> #endif  //  ifndef HPP_INCLUSION_GUARD_JCI_INSIGHT_RANDOM

Comment is wrong.

Back to comp.lang.c++ | Previous | Next — Previous in thread | Next in thread | Find similar | Unroll thread

Thread

One agnostic random function for big project (4 GB of code) Frederick Gotham <cauldwell.thomas@gmail.com> - 2019-12-12 03:53 -0800
  Re: One agnostic random function for big project (4 GB of code) Bart <bc@freeuk.com> - 2019-12-12 12:01 +0000
    Re: One agnostic random function for big project (4 GB of code) Frederick Gotham <cauldwell.thomas@gmail.com> - 2019-12-12 04:06 -0800
      Re: One agnostic random function for big project (4 GB of code) boltar@nowhere.co.uk - 2019-12-12 12:22 +0000
      Iceream is your friend (was Re: One agnostic random function for big project (4 GB of code)) Vir Campestris <vir.campestris@invalid.invalid> - 2019-12-13 21:51 +0000
  Re: One agnostic random function for big project (4 GB of code) Paavo Helde <myfirstname@osa.pri.ee> - 2019-12-12 14:31 +0200
    Re: One agnostic random function for big project (4 GB of code) Frederick Gotham <cauldwell.thomas@gmail.com> - 2019-12-12 04:47 -0800
      Re: One agnostic random function for big project (4 GB of code) Paavo Helde <myfirstname@osa.pri.ee> - 2019-12-12 15:15 +0200
        Re: One agnostic random function for big project (4 GB of code) Frederick Gotham <cauldwell.thomas@gmail.com> - 2019-12-12 05:27 -0800
          Re: One agnostic random function for big project (4 GB of code) Paavo Helde <myfirstname@osa.pri.ee> - 2019-12-12 16:25 +0200
            Re: One agnostic random function for big project (4 GB of code) Frederick Gotham <cauldwell.thomas@gmail.com> - 2019-12-12 06:33 -0800
            Re: One agnostic random function for big project (4 GB of code) Paavo Helde <myfirstname@osa.pri.ee> - 2019-12-12 16:40 +0200
              Re: One agnostic random function for big project (4 GB of code) Frederick Gotham <cauldwell.thomas@gmail.com> - 2019-12-12 07:26 -0800
                Re: One agnostic random function for big project (4 GB of code) Paavo Helde <myfirstname@osa.pri.ee> - 2019-12-12 17:37 +0200
                Re: One agnostic random function for big project (4 GB of code) Frederick Gotham <cauldwell.thomas@gmail.com> - 2019-12-12 07:48 -0800
                Re: One agnostic random function for big project (4 GB of code) Frederick Gotham <cauldwell.thomas@gmail.com> - 2019-12-12 08:27 -0800
    Re: One agnostic random function for big project (4 GB of code) Manfred <noname@add.invalid> - 2019-12-12 17:41 +0100

csiph-web