On Wednesday, December 10, 2014 6:25:21 AM UTC-5, Eric Cano wrote:
>
> Dear Crypo++ mainteners,
>
> We experienced several cases of crashes in the constructor of the Base64
> class. The problem arises when constructing CryptoPP::Base64Decoder in
> parallel in 2 threads for the first time in a program. We identified the
> race condition:
>
> The constructor of Base64Decoder calls GetDecodingLookupArray:
>
> class Base64Decoder : public BaseN_Decoder
> {
> public:
> Base64Decoder(BufferedTransformation *attachment = NULL)
> : BaseN_Decoder(GetDecodingLookupArray(), 6, attachment)
> {}
> void IsolatedInitialize(const NameValuePairs ¶meters) {}
> private:
> static const int * CRYPTOPP_API GetDecodingLookupArray();
> };
>
> GetDecodingLookupArray initializes a static structure, and the
> initialization is protected by a racy mechanism:
>
> const int *Base64Decoder::GetDecodingLookupArray()
> {
> static volatile bool s_initialized = false;
> static int s_array[256];
> if (!s_initialized)
> {
> InitializeDecodingLookupArray(s_array, s_vec, 64, false);
> s_initialized = true;
> }
> return s_array;
> }
>
>
> When thread 1 passes the if (!s_initialized) test, thread 2 has ample
> time to pass it as well and to concurrently initialize the array, before
> thread 1 reaches s_initialized = true;.
> A solution would be to use a mutex, which is a multiplatform issue unless
> CryptoPP gets compiled in C++11 (the language now has built-in mutexes).
>
> Another solution would be to unconditionally initialize all those
> structures at library load time by using a singleton's constructor.
> Drawback is longer load time and memory use.
>
> Yet another way would be to just make those structures compile-time static
> arrays, pre-computed in the source. Drawback is bigger binary, and usage of
> external script to pre-cook the sources.
>
The easiest solution may be to create a Base64{Encoder|Decoder} before the
program switches to multiple threads. That way, there is no race during the
lazy initialization.
I don't think its the best solution. Only the easiest.
Jeff
--
--
You received this message because you are subscribed to the "Crypto++ Users"
Google Group.
To unsubscribe, send an email to [email protected].
More information about Crypto++ and this group is available at
http://www.cryptopp.com.
---
You received this message because you are subscribed to the Google Groups
"Crypto++ Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
