ntdll: a simple implementation of cross-process NtAllocateVirtualMemory

[Dan Kegel]

On 6/1/06, tkho at ucla.edu <tkho at ucla.edu> wrote:
> One thought that has recently crossed my mind: Building off his framework and
> putting much of the code in the server may not have been a great idea, since
> the only real need for the server is to get the unix pid of the thread.

Nah, I think it was the right thing to do.  You definitely want the
server to do the dirty work here.


> +    /* return base address of allocated memory in eax */
> +    asm("movl %0, %%eax"
> +        :
> +        :"r"(mem)
> +        :"%eax");
> +    asm("int3"); /* execution doesn't continue past here */

Better surround that with #if defined(LINUX) && defined(__i386__), or whatever
the right symbol is, and explain better why 'int3' is how you return to
the caller of ptrace.

> +            printf("using allocator in kernel32 at 0x%08x\n",
> +                    (unsigned) allocator);

You might want to strip out the debugging prints...

> +enum remote_op_code
> +{
> +    REMOTE_OP_NEW_THREAD,
> +    REMOTE_OP_VM_ALLOC,
> +    REMOTE_OP_VM_FREE,
> +    REMOTE_OP_VM_PROTECT,
> +    REMOTE_OP_VM_QUERY,
> +    REMOTE_OP_VM_MAP,
> +    REMOTE_OP_VM_UNMAP,
> +    REMOTE_OP_VM_FLUSH
> +};

I would have ripped out any part of his patch you're not implementing,
just to keep the size down...


> +        case REMOTE_OP_VM_ALLOC:
> +        {
> +#if defined(linux) && defined(__i386__)
> +            int pid;
> +            struct user_regs_struct oldregs, regs;
> +            struct remote_op_params_vm_alloc *params;
> +            struct remote_op_result_vm_alloc result;
> + ...

You probably want to hide all the ptrace stuff inside a separate
function called, maybe, posix_remote_mmap(...).

Also, which bugzilla bugs / conformance tests will this fix?
If the answer is 'none', maybe you need to write them.
- Dan