For what do I need to use VirtualAlloc/VirtualAllocEx?
An example, one case that I found - if I allocated 4 GB of virtual memory, then if I do not use all of them, then I do not spend physical memory, and if I resize my array, I do not need to do new allocating and copying old data to new array.
struct T_custom_allocator; // which using VirtualAllocEx()
std::vector<int, T_custom_allocator> vec;
vec.reserve(4*1024*1024*1024); // allocated virtual memory (physical memory is not used)
vec.resize(16384); // allocated 16KB of physical memory
// ...
vec.resize(32768); // allocated 32KB of physical memory
// (no need to copy of first 16 KB of data)
And if I used standard allocator, I need to copy of data when I do resize:
std::vector<int> vec;
vec.resize(16384); // allocated 16KB of physical memory
// ...
vec.resize(32768); // allocated 32KB of physical memory
// and need to copy of first 16 KB of data
Or with standatd allocator, I must spend 4GB of physical memory:
std::vector<int> vec;
vec.reserve(4*1024*1024*1024); // allocated 4GB of physical memory
vec.resize(16384); // no need to do, except changing a local variable of size
// ...
vec.resize(32768); // no need to do, except changing a local variable of size
But, why this is better than realloc()?
http://www.cplusplus.com/reference/cstdlib/realloc/
And are there any else cases to use VirtualAlloc[Ex] with benefits?
Another use for VirtualAllocEx
which hasn't been mentioned yet, is to allocate memory in another process' address space. Note that the first parameter is the handle to a process - the function allocates the memory within the virtual address space of that process.
I've used this before when injecting code into another process, by forcing a LoadLibrary
call in the target process. The basic steps are as follows:
- Get the process id of the target process (e.g. with something like
GetWindowThreadProcessId
).
- Get a handle to the process with the appropriate permissions using
OpenProcess
.
- Allocate some memory in that process with
VirtualAllocEx
.
- Copy the name of your DLL into that memory with
WriteProcessMemory
.
- Get the address of the
LoadLibrary
function using GetProcAddress
.
- Call
CreateRemoteThread
to start the LoadLibrary
call in the target process, with the thread parameter being the memory you've allocated with VirtualAllocEx
(containing the name of the DLL).
Not that you needed to know all of that, but I though it was an interesting use case.
VirtualAlloc
and VirtualAllocEx
in very simplistic terms allocate raw pages, all other memory functions from malloc
to GlobalAlloc
all use VirtualAllocEx
underneath. The problem with VirtualAlloc
is that it is basically raw memory, there is no reallocation or relocation available. As such if your address space becomes fragmented you have no recourse but to release and rebuild.
The primary use case for VirtualAlloc
is when you need to write your own memory manager, for say a SQL implementation where it can make a huge difference. Or if you were implementing a Just In Time Compiler (JIT), as you would need to be able to change the protection flags on the page you compile into from read/write to read/execute as to not trigger Data Execution Prevention.