Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 24 Aug 2022 23:02:36 +0100
From:   Catalin Marinas <catalin.marinas@arm.com>
To:     Linus Torvalds <torvalds@linux-foundation.org>
Cc:     Steve French <smfrench@gmail.com>,
        CIFS <linux-cifs@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>,
        linux-arch <linux-arch@vger.kernel.org>
Subject: Re: strlcpy() notes (was Re: [GIT PULL] smb3 client fixes)
Message-ID: <Ywaf/CWYzmPx1lxa@arm.com>
References: <CAH2r5mtDFpaAWeGCtrfm_WPM6j-Gkt_O80=nKfp6y39aXaBr6w@mail.gmail.com>
 <CAHk-=wi+xbVq++uqW9YgWpHjyBHNB8a-xad+Xp23-B+eodLCEA@mail.gmail.com>
 <YwSWLH4Wp6yDMeKf@arm.com>
 <CAHk-=whU7QiwWeO81Rf+KGh0rGS9CEfKUXc5eik+Z0GaVJgu4A@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CAHk-=whU7QiwWeO81Rf+KGh0rGS9CEfKUXc5eik+Z0GaVJgu4A@mail.gmail.com>
Precedence: bulk

On Tue, Aug 23, 2022 at 10:37:38AM -0700, Linus Torvalds wrote:
> On Tue, Aug 23, 2022 at 1:56 AM Catalin Marinas <catalin.marinas@arm.com> wrote:
> > With load_unaligned_zeropad(), the arm64 implementation disables tag
> > checking temporarily. We could do the same with read_word_at_a_time()
> > (there is a kasan_check_read() in this function but it wrongly uses a
> > size of 1).
> 
> The "size of 1" is not wrong, it's intentional, exactly because people
> do things like
> 
>     strscpy(dst, "string", sizeof(dst));
> 
> which is a bit unfortunate, but very understandable and intended to
> work. So that thing may over-read the string by up to a word. And
> KASAN ends up being unhappy.

Good point. We could attempt a single-byte checked read on arm64 as well
and then disable tag checking (the arm64 load_unaligned_zeropad()
doesn't bother with this).

For KASAN, if we want to be more precise, we could move the
kasan_check_read() (or add a new one) in the strscpy() implementation
that actually takes into account how much was copied (non-zero bytes).
Not sure it's worth it though. The check would be post-read though.

-- 
Catalin