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: <YMe8ktUsdtwFKHuF@nuc10> <CACT4Y+ZjSbioNS8oPwUcyOrLhB6-Sf-WZmadAoAm0H-JYRLo1g@mail.gmail.com>
 <YMpCEu9yM5Ppj5jj@nuc10> <YNTCsqQUaYnlXGbO@nuc10> <20210625150132.GF20835@arm.com>
 <YNX1/6FS4z0eVaLF@nuc10>
In-Reply-To: <YNX1/6FS4z0eVaLF@nuc10>
From:   Dmitry Vyukov <dvyukov@google.com>
Date:   Fri, 25 Jun 2021 17:36:44 +0200
Message-ID: <CACT4Y+bWQ69-QhQswvzuhm=CvjYovhdGpDQXAkMSj819fc+MTA@mail.gmail.com>
Subject: Re: kmemleak memory scanning
To:     Rustam Kovhaev <rkovhaev@gmail.com>
Cc:     Catalin Marinas <catalin.marinas@arm.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Linux-MM <linux-mm@kvack.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Fri, Jun 25, 2021 at 5:28 PM Rustam Kovhaev <rkovhaev@gmail.com> wrote:
>
> Hi Catalin,
>
> On Fri, Jun 25, 2021 at 04:01:33PM +0100, Catalin Marinas wrote:
> > On Thu, Jun 24, 2021 at 10:36:50AM -0700, Rustam Kovhaev wrote:
> > > On Wed, Jun 16, 2021 at 11:25:22AM -0700, Rustam Kovhaev wrote:
> > > > On Tue, Jun 15, 2021 at 07:15:24AM +0200, Dmitry Vyukov wrote:
> > > > > On Mon, Jun 14, 2021 at 10:31 PM Rustam Kovhaev <rkovhaev@gmail.com> wrote:
> > > > > >
> > > > > > hello Catalin, Andrew!
> > > > > >
> > > > > > while troubleshooting a false positive syzbot kmemleak report i have
> > > > > > noticed an interesting behavior in kmemleak and i wonder whether it is
> > > > > > behavior by design and should be documented, or maybe something to
> > > > > > improve.
> > > > > > apologies if some of the questions do not make sense, i am still going
> > > > > > through kmemleak code..
> > > > > >
> > > > > > a) kmemleak scans struct page (kmemleak.c:1462), but it does not scan
> > > > > > the actual contents (page_address(page)) of the page.
> > > > > > if we allocate an object with kmalloc(), then allocate page with
> > > > > > alloc_page(), and if we put kmalloc pointer somewhere inside that page,
> > > > > > kmemleak will report kmalloc pointer as a false positive.
> > > > > > should we improve kmemleak and make it scan page contents?
> > > > > > or will this bring too many false negatives?
> > > > >
> > > > > Hi Rustam,
> > > > >
> > > > > Nice debugging!
> > > > > I assume lots of pages are allocated for slab and we don't want to
> > > > > scan the whole page if only a few slab objects are alive on the page.
> > > > > However alloc_pages() can be called by end kernel code as well.
> > > > > I grepped for any kmemleak annotations around existing calls to
> > > > > alloc_pages, but did not find any...
> > > > > Does it require an explicit kmemleak_alloc() after allocating the page
> > > > > and kmemleak_free () before freeing the page?
> > > >
> > > > hi Dmitry, thank you!
> > > > yes, as Catalin has pointed out, there are a few places where we call
> > > > kmemleak_alloc()/kmemleak_free() explicitly in order for the pages to be
> > > > scanned, like in blk_mq_alloc_rqs()
> > > >
> > > > > If there are more than one use case for this, I guess we could add
> > > > > some GFP flag for this maybe.
> > > >
> > > > and this way kernel users won't have to use kmemleak fuctions mentioned
> > > > above including some or most kmemleak_not_leak() calls and basically
> > > > kmemleak will be kind of "transparent" to them? and they will only need
> > > > to use the GFP flag to instruct kmemleak to scan the page contents?
> > > > it sounds like a good idea to me..
> > > >
> > >
> > > i've been thinking about this and it seems like in the scenario where we
> > > want kmemleak to scan only some part of the page, we will have to either
> > > do separate alloc_page() calls with different flags or use
> > > kmemleak_scan_area() to limit the memory scan area. maybe this approach
> > > won't simplify things and will produce more code instead of reducing it
> >
> > Since page allocation is not tracked by kmemleak, you can always do an
> > explicit kmemleak_alloc() call with a smaller size than a full page.
> >
> right, but if i understood Dmitry's idea correctly, he was thinking
> about using a new GFP flag, like GFP_KMEMLEAK, and burying
> kmemleak_alloc() in page allocator

I don't have a strong opinion either way. Especially since we have
only 1 use case so far.