Re: [RFC PATCH v4 5/6] uretprobes: invoke return probe handlers

[Oleg Nesterov <oleg@redhat.com>]

On 03/04, Anton Arapov wrote:
>
> +static void handle_uretprobe(struct xol_area *area, struct pt_regs *regs)
> +{
> +	struct hlist_head *head;
> +	struct hlist_node *tmp;
> +	struct return_uprobe_i *ri;
> +	struct uprobe_task *utask;
> +	unsigned long orig_ret_vaddr;
> +
> +	/* TODO: uretprobe bypass logic */
> +
> +	utask = get_utask();
> +	if (!utask) {
> +		/* TODO:RFC task is not probed, do we want printk here? */
> +		return;
> +	}
> +	head = &utask->return_uprobes;
> +	hlist_for_each_entry_safe(ri, tmp, head, hlist) {
> +		if (ri->uprobe->consumers) {
> +			instruction_pointer_set(regs, ri->orig_ret_vaddr);

This doesn't look right if ri->orig_ret_vaddr == area->vaddr. We should
splice the list and find orig_ret_vaddr in advance.

> @@ -1589,8 +1639,11 @@ static void handle_swbp(struct pt_regs *regs)
>
>  	if (!uprobe) {
>  		if (is_swbp > 0) {
> -			/* No matching uprobe; signal SIGTRAP. */
> -			send_sig(SIGTRAP, current, 0);
> +			area = get_xol_area();
> +			if (area && bp_vaddr == area->vaddr)
> +				handle_uretprobe(area, regs);
> +			else
> +				send_sig(SIGTRAP, current, 0);

Why? We can check bp_vaddr at the start, before find_active_uprobe().

And I'd suggest to not use area->vaddr directly, imho a trivial helper
makes sense.

Oleg.