Re: [PATCH 1/3] uprobes: install_breakpoint() should fail if is_swbp_insn() == T

[Peter Zijlstra <peterz@infradead.org>]

On Wed, 2012-05-30 at 23:07 +0530, Srikar Dronamraju wrote:
> > > 
> > > Signed-off-by: Oleg Nesterov <oleg@redhat.com>
> > > ---
> > >  kernel/events/uprobes.c |    2 +-
> > >  1 files changed, 1 insertions(+), 1 deletions(-)
> > > 
> > > diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
> > > index 8c5e043..1593b43 100644
> > > --- a/kernel/events/uprobes.c
> > > +++ b/kernel/events/uprobes.c
> > > @@ -704,7 +704,7 @@ install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
> > >  			return ret;
> > >  
> > >  		if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
> > > -			return -EEXIST;
> > > +			return -ENOTSUPP;
> > >  
> > >  		ret = arch_uprobe_analyze_insn(&uprobe->arch, mm);
> > >  		if (ret)
> > 
> > IIRC this -EEXIST existed because the vma iteration it does is racy and
> > one can encounter the same vma twice or so. See the special -EEXIST
> > handling in register_for_each_vma().
> > 
> > Changing it like this would break stuff. 
> > 
> 
> Peter, 
> 
> is_swbp_insn() is looking at the copy of the instruction thats read from
> the file. This path is only taken even before any mm's are inserted with
> the breakpoint instruction.
> 
> We still check and return -EEXIST if the memory while inserting the breakpoint
> instruction  already has a breakpoint.
> 
> Hence this change is correct. 

Oh, indeed. I overlooked copy_insn() is taking the page from the
page-cache instead of the page-tables and will thus get the original.

OK, no worries then.