Re: [patch 3/6] sys_indirect RFC - sys_indirect core

["Ulrich Drepper" <drepper@gmail.com>]

On 6/29/07, Davide Libenzi <davidel@xmailserver.org> wrote:
> +int indirect_set_context(struct fsa_context *ator,
> +                        const struct indirect_ctx __user * __user *ctxs,
> +                        unsigned int nctxs, struct indirect_op **first)
> +{
> +       unsigned int i;
> +       int error;
> +       u32 ctx;
> +       const struct indirect_ctx __user *pctx;
> +       struct indirect_op *new;
> +
> +       *first = NULL;
> +       for (i = 0; i < nctxs; i++) {
> +               if (get_user(pctx, &ctxs[i]) || get_user(ctx, &pctx->ctx))
> +                       return -EFAULT;
> +               if (unlikely(ctx >= ARRAY_SIZE(inprocs) || !inprocs[ctx].set))
> +                       return -EINVAL;
> +               error = (*inprocs[ctx].set)(ator, pctx, &new);
> +               if (unlikely(error))
> +                       return error;
> +               new->next = *first;
> +               *first = new;
> +       }

If you use one single struct as explained in my last mail all this
shouldn't be necessary.  The sys_indirect syscall would simply points
current->ind_ctx to a kernel-copy of the struct.  Then call the
syscall and on return clear current->ind_ctx.

In the affected syscalls we can then test whether current->ind_ctx is
NULL and if not, enable the extra functionality.

These callbacks etc seem to be far too expensive and complicated.