rpcrt4: Fix regression in NdrConformantStringUnmarshall by doing the memcpy check in safe_copy_from_buffer

[Robert Shearman]

Maarten Lankhorst wrote:
> @@ -665,7 +665,8 @@ static inline void safe_copy_from_buffer(MIDL_STUB_MESSAGE *pStubMsg, void *p, U
>      if ((pStubMsg->Buffer + size < pStubMsg->Buffer) || /* integer overflow of pStubMsg->Buffer */
>          (pStubMsg->Buffer + size > pStubMsg->BufferEnd))
>          RpcRaiseException(RPC_X_BAD_STUB_DATA);
> -    memcpy(p, pStubMsg->Buffer, size);
> +    if (p != pStubMsg->Buffer)
> +        memcpy(p, pStubMsg->Buffer, size);
>      pStubMsg->Buffer += size;
>  }
>  
> @@ -890,8 +891,7 @@ unsigned char *WINAPI NdrConformantStringUnmarshall( PMIDL_STUB_MESSAGE pStubMsg
>        *ppMemory = NdrAllocate(pStubMsg, memsize);
>    }
>  
> -  if (*ppMemory != pStubMsg->Buffer)
> -    safe_copy_from_buffer(pStubMsg, *ppMemory, bufsize);
> +  safe_copy_from_buffer(pStubMsg, *ppMemory, bufsize);
>  
>    if (*pFormat == RPC_FC_C_CSTRING) {
>      TRACE("string=%s\n", debugstr_a((char*)*ppMemory));
>   

Good work in spotting and fixing the mistake I made, but I think I'd 
prefer to fix it by making the caller of safe_copy_from_buffer do the 
incrementing of the buffer. This is to avoid confusion with the name of 
the function and to avoid the possibility that the buffer is incremented 
twice.

-- 
Rob Shearman