Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <cover.1605305978.git.andreyknvl@google.com> <52518837b34d607abbf30855b3ac4cb1a9486946.1605305978.git.andreyknvl@google.com>
 <CACT4Y+ZaRgqpgPRe5k5fVrhd_He5_6N55715YzwWcQyvxYUNRQ@mail.gmail.com>
In-Reply-To: <CACT4Y+ZaRgqpgPRe5k5fVrhd_He5_6N55715YzwWcQyvxYUNRQ@mail.gmail.com>
From:   Andrey Konovalov <andreyknvl@google.com>
Date:   Mon, 23 Nov 2020 19:54:47 +0100
Message-ID: <CAAeHK+xv2UQyD1MtAiu8d=cRbJDNXQaaA-Qh+Eut3gRnLbJEMA@mail.gmail.com>
Subject: Re: [PATCH mm v3 17/19] kasan: clean up metadata allocation and usage
To:     Dmitry Vyukov <dvyukov@google.com>
Cc:     Andrew Morton <akpm@linux-foundation.org>,
        Catalin Marinas <catalin.marinas@arm.com>,
        Will Deacon <will.deacon@arm.com>,
        Vincenzo Frascino <vincenzo.frascino@arm.com>,
        Andrey Ryabinin <aryabinin@virtuozzo.com>,
        Alexander Potapenko <glider@google.com>,
        Marco Elver <elver@google.com>,
        Evgenii Stepanov <eugenis@google.com>,
        Branislav Rankov <Branislav.Rankov@arm.com>,
        Kevin Brodsky <kevin.brodsky@arm.com>,
        kasan-dev <kasan-dev@googlegroups.com>,
        Linux ARM <linux-arm-kernel@lists.infradead.org>,
        Linux-MM <linux-mm@kvack.org>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Nov 17, 2020 at 2:12 PM Dmitry Vyukov <dvyukov@google.com> wrote:
>
> >  void __kasan_poison_slab(struct page *page)
> >  {
> > @@ -272,11 +305,9 @@ void * __must_check __kasan_init_slab_obj(struct kmem_cache *cache,
> >         struct kasan_alloc_meta *alloc_meta;
> >
> >         if (kasan_stack_collection_enabled()) {
> > -               if (!(cache->flags & SLAB_KASAN))
> > -                       return (void *)object;
>
> Is it a subtle change in behavior?
> Previously we had an early return and also did not set tag, now we
> only skip memset but set tag... was it a bug before?...

This is a change in behavior, see the patch description. We now always
sanitize an object's contents, but only store/update the metadata when
it fits. I'll update the patch title, as it might sound confusing, as
it kind of implies we're not changing the behavior.

> > @@ -135,7 +135,12 @@ static void qlink_free(struct qlist_node *qlink, struct kmem_cache *cache)
> >         if (IS_ENABLED(CONFIG_SLAB))
> >                 local_irq_save(flags);
> >
> > +       /*
> > +        * As the object now gets freed from the quaratine, assume that its
> > +        * free track is now longer valid.
>
> typo: _no_ longer valid

Will fix!

>
> > +        */
> >         *(u8 *)kasan_mem_to_shadow(object) = KASAN_KMALLOC_FREE;
> > +
> >         ___cache_free(cache, object, _THIS_IP_);
> >
> >         if (IS_ENABLED(CONFIG_SLAB))
> > @@ -168,6 +173,9 @@ void quarantine_put(struct kmem_cache *cache, void *object)
> >         struct qlist_head temp = QLIST_INIT;
> >         struct kasan_free_meta *meta = kasan_get_free_meta(cache, object);
> >
> > +       if (!meta)
> > +               return;
>
> Humm... is this possible? If yes, we would be leaking the object here...
> Perhaps BUG_ON with a comment instead.

No, this isn't possible. Will turn this into a warning and add a comment.

Thanks!