On 03/01, Linus Torvalds wrote:
>
> So I'm ok with your alternative
>
> > typedef union {
> > const char __user *const __user *native;
> > compat_uptr_t __user *compat;
> > } conditional_user_ptr_t;
>
> model instead, which moves the pointer into the union.
>
> However, if you do this, then I have one more suggestion: just move
> the "compat" flag in there too!
>
> Every time you pass the union, you're going to pass the compat flag to
> distinguish the cases. So do it like this:
>
> struct conditional_ptr {
> int is_compat;
> union {
> const char __user *const __user *native;
> compat_uptr_t __user *compat;
> };
> };
>
> and it will all look much cleaner, I bet.
Heh. I knew. I swear, I knew you would suggest this ;)
OK, please find v3. I had to deanonymize the union though, otherwise
the initializer in do_execve() becomes nontrivial.
But I don't think this is right. Not only this adds 200 bytes to exec.o.
To me, is_compat is not the private property of argv/envp. Yes, currently
nobody except get_arg_ptr() needs to know the difference. But who knows,
it is possible that we will need more "if (compat)" code in future. IOW,
I think that the explicit argument is a win.
Never mind. I agree with everything as long as we can remove this c-a-p
compat_do_execve().
Oleg.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxxx For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>