Re: [PATCH 11/15] uprobes: move BUG_ON(UPROBE_SWBP_INSN_SIZE) from write_opcode() to install_breakpoint()

[Oleg Nesterov <oleg@redhat.com>]

On 06/15, Srikar Dronamraju wrote:
>
> > @@ -699,6 +694,10 @@ install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
> >  		if (ret)
> >  			return ret;
> >
> > +		/* write_opcode() assumes we don't cross page boundary */
> > +		BUG_ON((uprobe->offset & ~PAGE_MASK) +
> > +				UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
> > +
> >  		uprobe->flags |= UPROBE_COPY_INSN;
> >  	}
>
> I am now thinking if we really need a BUG_ON?

I was thinking about this too.

> I am now thinking I should
> have had a check at the start in uprobe_register() and failed the request.
>
> Something like
> 	if ((offset & ~PAGE_MASK) + UPROBE_SWBP_INSN_SIZE > PAGE_SIZE)
> 		return -EINVAL;

Perhaps. Or we can simply remove it. arch_uprobe_analyze_insn()
should be careful anyway, and all this validation should be moved
into uprobe_register/alloc_uprobe.

I do not really mind, I only wanted to simplify write_opcode() which
does a lot of unnecessary things (say, lock_page, I am going to kill
it).

So. Do you want me to redo this patch? Or do you think we can keep
this "must not happen after arch_uprobe_analyze_insn" check?

Oleg.