Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [RFC PATCH v2 3/3] kasan: add interceptors for all string
 functions
To:     Andrey Ryabinin <aryabinin@virtuozzo.com>,
        Benjamin Herrenschmidt <benh@kernel.crashing.org>,
        Paul Mackerras <paulus@samba.org>,
        Michael Ellerman <mpe@ellerman.id.au>,
        Nicholas Piggin <npiggin@gmail.com>,
        "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>,
        Alexander Potapenko <glider@google.com>,
        Dmitry Vyukov <dvyukov@google.com>,
        Daniel Axtens <dja@axtens.net>
Cc:     linux-mm@kvack.org, linuxppc-dev@lists.ozlabs.org,
        linux-kernel@vger.kernel.org, kasan-dev@googlegroups.com
References: <f13944c4e99ec2cef6d93d762e6b526e0335877f.1553785019.git.christophe.leroy@c-s.fr>
 <51a6d9d7185de310f37ccbd7e4ebfdd6c7e9791f.1553785020.git.christophe.leroy@c-s.fr>
 <3211b0f8-7b52-01b7-8208-65d746969248@c-s.fr>
 <ae9262e5-0917-b7c9-52c7-fe21db2ecacb@virtuozzo.com>
From:   Christophe Leroy <christophe.leroy@c-s.fr>
Message-ID: <d5d2fa6f-c2ba-8234-e412-9c4ccd6cc4c6@c-s.fr>
Date:   Tue, 2 Apr 2019 22:36:43 +0200
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:60.0) Gecko/20100101
 Thunderbird/60.6.1
MIME-Version: 1.0
In-Reply-To: <ae9262e5-0917-b7c9-52c7-fe21db2ecacb@virtuozzo.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: fr
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


Le 02/04/2019 à 18:14, Andrey Ryabinin a écrit :
> 
> 
> On 4/2/19 12:43 PM, Christophe Leroy wrote:
>> Hi Dmitry, Andrey and others,
>>
>> Do you have any comments to this series ?
>>
> 
> I don't see justification for adding all these non-instrumented functions. We need only some subset of these functions
> and only on powerpc so far. Arches that don't use str*() that early simply doesn't need not-instrumented __str*() variant.
> 
> Also I don't think that auto-replace str* to __str* for all not instrumented files is a good idea, as this will reduce KASAN coverage.
> E.g. we don't instrument slub.c but there is no reason to use non-instrumented __str*() functions there.

Ok, I didn't see it that way.

In fact I was seeing the opposite and was considering it as an 
opportunity to increase KASAN coverage. E.g.: at the time being things 
like the above (from arch/xtensa/include/asm/string.h) are not covered 
at all I believe:

#define __HAVE_ARCH_STRCPY
static inline char *strcpy(char *__dest, const char *__src)
{
	register char *__xdest = __dest;
	unsigned long __dummy;

	__asm__ __volatile__("1:\n\t"
		"l8ui	%2, %1, 0\n\t"
		"s8i	%2, %0, 0\n\t"
		"addi	%1, %1, 1\n\t"
		"addi	%0, %0, 1\n\t"
		"bnez	%2, 1b\n\t"
		: "=r" (__dest), "=r" (__src), "=&r" (__dummy)
		: "0" (__dest), "1" (__src)
		: "memory");

	return __xdest;
}

In my series, I have deactivated optimised string functions when KASAN 
is selected like arm64 do. See https://patchwork.ozlabs.org/patch/1055780/
But not every arch does that, meaning that some string functions remains 
not instrumented at all.

Also, I was seeing it as a way to reduce impact on performance with 
KASAN. Because instrumenting each byte access of the non-optimised 
string functions is a performance genocide.

> 
> And finally, this series make bug reporting slightly worse. E.g. let's look at strcpy():
> 
> +char *strcpy(char *dest, const char *src)
> +{
> +	size_t len = __strlen(src) + 1;
> +
> +	check_memory_region((unsigned long)src, len, false, _RET_IP_);
> +	check_memory_region((unsigned long)dest, len, true, _RET_IP_);
> +
> +	return __strcpy(dest, src);
> +}
> 
> If src is not-null terminated string we might not see proper out-of-bounds report from KASAN only a crash in __strlen().
> Which might make harder to identify where 'src' comes from, where it was allocated and what's the size of allocated area.
> 
> 
>> I'd like to know if this approach is ok or if it is better to keep doing as in https://patchwork.ozlabs.org/patch/1055788/
>>
> I think the patch from link is a better solution to the problem.
> 

Ok, I'll stick with it then. Thanks for your feedback

Christophe