Re: [PATCH v2 2/3] kasan: integrate the common part of two KASAN tag-based modes

[Kuan-Ying Lee <kylee0686026@gmail.com>]

On Mon, Jun 14, 2021 at 10:48:27AM +0200, Marco Elver wrote:
> On Sat, 12 Jun 2021 at 17:51, Kuan-Ying Lee <kylee0686026@gmail.com> wrote:
> [...]
> > > > diff --git a/mm/kasan/report_tags.h b/mm/kasan/report_tags.h
> > > > new file mode 100644
> > > > index 000000000000..4f740d4d99ee
> > > > --- /dev/null
> > > > +++ b/mm/kasan/report_tags.h
> > > > @@ -0,0 +1,56 @@
> > > > +/* SPDX-License-Identifier: GPL-2.0 */
> > > > +#ifndef __MM_KASAN_REPORT_TAGS_H
> > > > +#define __MM_KASAN_REPORT_TAGS_H
> > > > +
> > > > +#include "kasan.h"
> > > > +#include "../slab.h"
> > > > +
> > > > +#ifdef CONFIG_KASAN_TAGS_IDENTIFY
> > > > +const char *kasan_get_bug_type(struct kasan_access_info *info)
> > > > +{
> > > [...]
> > > > +       /*
> > > > +        * If access_size is a negative number, then it has reason to be
> > > > +        * defined as out-of-bounds bug type.
> > > > +        *
> > > > +        * Casting negative numbers to size_t would indeed turn up as
> > > > +        * a large size_t and its value will be larger than ULONG_MAX/2,
> > > > +        * so that this can qualify as out-of-bounds.
> > > > +        */
> > > > +       if (info->access_addr + info->access_size < info->access_addr)
> > > > +               return "out-of-bounds";
> > >
> > > This seems to change behaviour for SW_TAGS because it was there even
> > > if !CONFIG_KASAN_TAGS_IDENTIFY. Does it still work as before?
> > >
> >
> > You are right. It will change the behavior.
> > However, I think that if !CONFIG_KASAN_TAG_IDENTIFY, it should be reported
> > "invalid-access".
> 
> There's no reason that if !CONFIG_KASAN_TAG_IDENTIFY it should be
> reported as "invalid-acces" if we can do better without the additional
> state that the config option introduces.
> 
> It's trivial to give a slightly better report without additional
> state, see the comment explaining why it's reasonable to infer
> out-of-bounds here.
> 
> > Or is it better to keep it in both conditions?
> 
> We want to make this patch a non-functional change.
>

Got it.

> [...]
> > > > diff --git a/mm/kasan/tags.c b/mm/kasan/tags.c
> > > > new file mode 100644
> > > > index 000000000000..9c33c0ebe1d1
> > > > --- /dev/null
> > > > +++ b/mm/kasan/tags.c
> > > > @@ -0,0 +1,58 @@
> > > > +// SPDX-License-Identifier: GPL-2.0
> > > > +/*
> > > > + * This file contains common tag-based KASAN code.
> > > > + *
> > > > + * Author: Kuan-Ying Lee <kylee0686026@gmail.com>
> > >
> > > We appreciate your work on this, but this is misleading. Because you
> > > merely copied/moved the code, have a look what sw_tags.c says -- that
> > > should either be preserved, or we add nothing here.
> > >
> > > I prefer to add nothing or the bare minimum (e.g. if the company
> > > requires a Copyright line) for non-substantial additions because this
> > > stuff becomes out-of-date fast and just isn't useful at all. 'git log'
> > > is the source of truth.
> >
> > This was my first time to upload a new file.
> > Thanks for the suggestions. :)
> > I will remove this author tag and wait for Greg's process advice.
> >
> > >
> > > Cc'ing Greg for process advice. For moved code, does it have to
> > > preserve the original Copyright line if there was one?
> 
> Greg responded, see his emails. Please preserve the original header
> from the file the code was moved from (hw_tags.c/sw_tags.c).

Ok. I will do it in v3.
Thanks.

> 
> Thanks,
> -- Marco