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;
MIME-Version: 1.0
References: <20220225221625.3531852-1-keescook@chromium.org> <CANpmjNOup5JCjRpRkhsF3Z+dPX6_MQE5u6WhnMit84c1TyRK+A@mail.gmail.com>
In-Reply-To: <CANpmjNOup5JCjRpRkhsF3Z+dPX6_MQE5u6WhnMit84c1TyRK+A@mail.gmail.com>
From:   Daniel Micay <danielmicay@gmail.com>
Date:   Mon, 28 Feb 2022 09:48:41 -0500
Message-ID: <CA+DvKQ+PYtDyejaUoBj51D_Y7muYaR1_FhQGWGWvgQawCbp31Q@mail.gmail.com>
Subject: Re: [PATCH] mm: Handle ksize() vs __alloc_size by forgetting size
To:     Marco Elver <elver@google.com>
Cc:     Kees Cook <keescook@chromium.org>, llvm@lists.linux.dev,
        Pekka Enberg <penberg@kernel.org>,
        David Rientjes <rientjes@google.com>,
        Joonsoo Kim <iamjoonsoo.kim@lge.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Vlastimil Babka <vbabka@suse.cz>, linux-mm@kvack.org,
        stable@vger.kernel.org,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        "Rafael J. Wysocki" <rafael@kernel.org>,
        Christoph Lameter <cl@linux.com>,
        Nathan Chancellor <nathan@kernel.org>,
        Nick Desaulniers <ndesaulniers@google.com>,
        linux-kernel@vger.kernel.org, linux-hardening@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

There aren't many calls to ksize in the entire Linux kernel source
tree. Most use cases are using the memory as some kind of (chunked)
dynamic array, where they either need to realloc or kmalloc a new
chunk when it runs out of space. Changing the approach to this API
would be both more efficient and fully compatible with alloc_size
since it would simply not be added to these functions:

    kmalloc_size(size, &result_size)
    krealloc_size(p, new_size, &result_size)

Nearly every use of ksize could be easily phased out this way. There
are probably a few which are harder to remove. It can be gradually
phased out by keeping around ksize temporarily but documenting that
it's only correct to use it on memory allocated with
kmalloc_size/krealloc_size. I think it can be phased out quite quickly
though. Look at how many calls there are to it. It's really not a lot,
especially if you filter out the uses of the identifier 'ksize' for
variables rather than calls to that function.

I brought this up when I originally submitted CONFIG_FORTIFY_SOURCE.
It's the main reason that I didn't bother trying to submit the
alloc_size attributes myself. The most important ones are for kmalloc
and it isn't technically correct to use it.