Re: Exception with AutoSeededRandomPool - Probably BUG in CryptoPP?

[Voronkov Konstantin]

One more short :)   I propose to solve this problem (in case you agree this is problem)  by doing self-test on attachement of the first thread instead of process attachment to CryptoPP library.   Best Regards and thank you for your attention, Konstantin ----- Original Message ----- From: Voronkov Konstantin To: [EMAIL PROTECTED] Sent: Monday, November 29, 2004 7:31 PM Subject: Re: Exception with AutoSeededRandomPool - Probably BUG in CryptoPP? Hello, again :)   Probably I found the reason of the problem:  the reason may be because in DllMain of CryptoPP there is a self-test, but the advapi dll is not loaded. That is why problem happens. To solve the problem I'll comment self-test of CrypoPP in DllMain,  thus no function from advapi will be used, and I'll have no exceptions.   At least this works for me.   Regards, Voronkov Konstantin     ----- Original Message ----- From: Voronkov Konstantin To: [EMAIL PROTECTED] Sent: Monday, November 29, 2004 3:36 PM Subject: Re: Exception with AutoSeededRandomPool - Probably BUG in CryptoPP? Hello, All!   We still experience the problem. Here is bit of additional information:  - the call stack can be found in the first mail (see below)  - the exception we have is: OS_Rgn:CryptGenRandom operation failed with error: OTHER_ERROR  - we are calling  AutoSeededRandomPool rng with no parameters  -  When exception is catched the GetlastError returns error code 87. Error lookup said: The parameter is incorrect.   In MSDN the explanation on CryptGenRandom given next information  ERROR_INVALID_PARAMETER One of the parameters contains an invalid value. This is most often an illegal pointer. The pointer provided to the CryptGenRandom function is  SecByteBlock seed(seedSize);   Additional information on compilation settings on CryptoPP and our application: - we have multithread Dll run-time library - we do not have _WIN64-platform - we use VS 6 compiler with service pack 5 - we use __cdecl calling conventions - we use default new and delete operators from run-time library     The main problem, that we can not reproduce it in every applciation,  only in our complex application which links and uses MFC.   Can anyone hint where we can start searching the problem, please?   Regards, Voronkov Konstantin   ----- Original Message ----- From: Voronkov Konstantin To: [EMAIL PROTECTED] Sent: Monday, November 29, 2004 12:18 PM Subject: Re: Exception with AutoSeededRandomPool - Probably BUG in CryptoPP? Hello, All!   I made analysis of source code of CryptoPP and came to next question. There is a call to CryptReleaseContext() on Windows Platform,  but no call to CryptContextAddRef(). IMHO, according to MSDN the call to CryptContextAddRef() is missing. Is not it?   Best Regards, Voronkov Konstantin     ----- Original Message ----- From: Voronkov Konstantin To: [EMAIL PROTECTED] Sent: Wednesday, November 17, 2004 7:00 PM Subject: Exception with AutoSeededRandomPool Hello, All!   I have exception using AutoSeededRandomPool. OS_Rgn:CryptGenRandom operation failed with error 0x57  OTHER_ERROR   I'm calling AutoSeededRandomPool rng;   Here is the stack: ADVAPI32! 7c2e7324() ADVAPI32! 7c2ea85b() CRYPTOPP! 42966e6a() AfxWinMain(HINSTANCE__ * 0x00400000, HINSTANCE__ * 0x00000000, char * 0x00133c8e, int 1) line 39 + 11 bytes WinMain(HINSTANCE__ * 0x00400000, HINSTANCE__ * 0x00000000, char * 0x00133c8e, int 1) line 30 WinMainCRTStartup() line 330 + 54 bytes KERNEL32! 7c4e87f5() I'm using modified version of CryptoPP 5.21 This happens in MFC application, and does not reproduced  in little sample console application.   Have anyone any idea about the reasons of this problem?   Best Regards, Voronkov Konstantin