Re: [PATCH 09/10] uprobes: Change prepare_uretprobe() to (try to) flush the dead frames

[Oleg Nesterov <oleg@redhat.com>]

On 05/04, Oleg Nesterov wrote:
>
> --- a/kernel/events/uprobes.c
> +++ b/kernel/events/uprobes.c
> @@ -1522,6 +1522,16 @@ static unsigned long get_trampoline_vaddr(void)
>  	return trampoline_vaddr;
>  }
>  
> +static void cleanup_return_instances(struct uprobe_task *utask, struct pt_regs *regs)
> +{
> +	struct return_instance *ri = utask->return_instances;
> +	while (ri && !arch_uretprobe_is_alive(&ri->auret, regs)) {
> +		ri = free_ret_instance(ri);
> +		utask->depth--;
> +	}
> +	utask->return_instances = ri;
> +}
> +
>  static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
>  {
>  	struct return_instance *ri;
> @@ -1576,6 +1586,9 @@ static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
>  	ri->orig_ret_vaddr = orig_ret_vaddr;
>  	ri->chained = chained;
>  
> +	if (utask->depth) /* drop the entries invalidated by longjmp() */
> +		cleanup_return_instances(utask, regs);
> +

Self nack ;)

Note that that prepare_uretprobe() does

	if (orig_ret_vaddr == trampoline_vaddr) {
		if (!utask->return_instances) {
			/*
			 * This situation is not possible. Likely we have an
			 * attack from user-space.
			 */
			pr_warn("uprobe: unable to set uretprobe pid/tgid=%d/%d\n",
						current->pid, current->tgid);
			goto fail;

_before_ we do cleanup_return_instances().

This is actually fine in a sense that ->return_instances == NULL after
cleanup_return_instances() is not possible if chained, there should be
another _alive() frame. But malicious user can obviously fool the kernel.

Easy to fix. But after discussion with Srikar and Pratyush (thanks!) I
decided to update 6-10.

Oleg.