Re: [patch v2 4/4] ptrace: Add support for generic PTRACE_GETREGSET/PTRACE_SETREGSET

[Roland McGrath <roland@redhat.com>]

> +static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
> +			struct iovec *uiov)
			       	     ^__user
> +{
> +	const struct user_regset_view *view = task_user_regset_view(task);
> +	const struct user_regset *regset = find_regset(view, type);
> +	struct iovec kiov;
> +
> +	if (!regset)
> +		return -EIO;
> +
> +	if (copy_from_user(&kiov, uiov, sizeof kiov))
> +		return -EFAULT;

Since it's just two words, most places handling struct iovec seem to just
use two get_user() calls, which is probably more efficient.

Also, here is where this function would need to be split in half for
compat_ptrace_request() calls where it has to use struct compat_iovec.

> +	// I am not sure. Afaics it is OK to pass the
> +	// size which is less than n * size. If iov_len
> +	// is bigger, we can silently truncate it, or
> +	// even write the correct value back.

Modifying iov_len to report how much we accessed seems good to me.  If we
do that, we should certainly allow a larger size for get, so userland can
just use a generic large buffer before even knowing the real size.  I'm not
sure whether we should allow a smaller size, especially for set.  I could
go either way.


Thanks,
Roland