"Victor Bazarov" <v.********@comAcast.netwrote in message
news:e9**********@news.datemas.de...
Jim Langston wrote:
>[..]
The question has already been answered, but I had to try out of
curiosity. This program gives a memory violation trying to write to
memory on the 2nd call to FuncP. I was sure it wouldn't work, but
had to see what would happen anyway.
#include <string>
#include <iostream>
int TestFunction()
{
return 2;
}
typedef int (*Func)();
int main()
{
Func FuncP;
FuncP = TestFunction;
int x = FuncP();
std::cout << x << std::endl;
FuncP = (Func)malloc( 1000 );
So, here 'FuncP' is a pointer to *data*. Even though you cast it to
a pointer to function, it doesn't really point to any function, does it?
Definately. And I knew this when I did the malloc.
> memcpy( FuncP, TestFunction, 1000 );
Now, since 'TestFunction' is not a pointer to an object, the behaviour
of that code is undefined. But even if we assume that you're allowed
to read bytes from the memory location behind 'TestFunction', you're
storing those bytes into data memory.
I actually expected to get an error here about an illegal memory read.
Although I didn't. I guess my OS allows reading of data in the program
segment.
> x = 4;
x = FuncP(); // Crashes here
And here you're asking to treat the data as if it were *code*.
In modern OSes, you cannot execute data unless you have special
permissions or changed permissions (or properties) of the memory
where you want to create code. Of course it doesn't work.
Right, and I didn't think it would. I didn't expect to get a write error,
however, but had expected to get a read or execute error. I really can't
think of what was trying to be written during the execution. Perhaps it
actually was an execution error but they didn't catch that specific error as
such and just explained it as a write error.
> free(FuncP);
std::cout << x << std::endl;
std::string wait;
std::cin >wait;
This actually made me start to think a little bit about what would be
required to create such a function. I think I would have to get down into
the OS level. I think there would be many, may problems with this however,
such as the jmp statements which would get their jump points from a data
program segment which would have to be changed.
I think it would not be trivial and would wind up being more like creating
an OS itself then a simple program.
