[Bug 20984] New: CRYPT_GetDefaultProvider-related leak in crypt32/tests/message.c

http://bugs.winehq.org/show_bug.cgi?id=20984

Summary: CRYPT_GetDefaultProvider-related leak in
                    crypt32/tests/message.c
           Product: Wine
           Version: 1.1.34
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Keywords: download, source, testcase
          Severity: normal
          Priority: P2
         Component: crypt32
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]

Juan writes:

-- snip --
The key leaked in
http://kegel.com/wine/valgrind/logs/2009-12-08-09.24/vg-crypt32_message.txt

972 bytes in 1 blocks are definitely lost 
    at  notify_alloc (heap.c:247)
    by  RtlAllocateHeap (heap.c:1697)
    by  new_object (handle.c:359)
    by  new_key (rsaenh.c:834)
    by  import_public_key (rsaenh.c:2777)
    by  import_key (rsaenh.c:2973)
    by  RSAENH_CPImportKey (rsaenh.c:3020)
    by  CryptImportKey (crypt.c:1827)
    by  CRYPT_ImportRsaPublicKeyInfoEx (encode.c:4786)
    by  CryptImportPublicKeyInfoEx (encode.c:4819)
    by  CryptImportPublicKeyInfo (encode.c:4757)
    by  CDecodeSignedMsg_VerifySignatureWithKey (msg.c:2554)
    by  CDecodeSignedMsg_VerifySignature (msg.c:2613)
    by  CDecodeMsg_Control (msg.c:2668)
    by  CryptMsgControl (msg.c:2805)
    by  CryptVerifyDetachedMessageSignature (message.c:164)
    by  test_verify_detached_message_signature (message.c:403)

should be suppressed.  Feel free to open a bug if you like so you can
reference it.  The existing bug about a leak in rsaenh, 20684, doesn't
seem appropriate, as it may not cover this case.

Here's the scoop:  the leak comes from
test_verify_detached_message_signature.  This calls
CryptVerifyDetachedMessageSignature, with one of its arguments being a
pointer to a CRYPT_VERIFY_MESSAGE_PARA structure.  This calls
CryptMsgOpenToDecode, passing as the crypto provider the provider set
in the CRYPT_VERIFY_MESSAGE_PARA structure... except that it's 0.  If
it's 0, it calls CRYPT_GetDefaultProvider(), and sets an internal flag
indicating the the crypto provider shouldn't be unloaded, since it's
process-wide.  There's that famous CRYPT_GetDefaultProvider again:  it
can't be unloaded by crypt32, so it leaks whatever it's got, including
the last set of keys that get set into it.
-- snip --

Filing bug for reference, I'll add a suppression.